[probably most useful if you don’t want to hear it, see rule of equal and opposite advice]

If you read my post on guilt and pride, you can probably infer that I don’t feel a strong internal sense of self-worth. I do like my own existence and want it to continue, but I don’t really feel in any way “worthy”.

I emotionally feel like I don’t really “do good” through action, but rather that I am “allowing bad things to happen” through inaction or imperfect action. And I am pretty utilitarian, and the world is very big, so I feel like there is so much to be done.

I struggle to change my view, even if it is not a healthy mindset long-term. I still see that, you can feel personally unhappy if it makes other sufficiently happy. And I’m not sure this is right.

seeing symmetry

But when I look at my friends, I see that they have done all sorts of good things, have things I would want them to be proud of, and yet they too still mostly just feel guilt about not having done more too... and I just want them to be happy.

I would be very willing to feel unhappy for all my friends to feel happy. And yet, to large part, they feel symmetry.

Hmm, something is not quite right…

How could this happen?

decision theory reasons

All of me and my friends and you1 are stuck in some loop, where we each would press the button that makes myself feel guilt, but that makes my friend happy and not feel guilt.

Unfortunately, I cannot make you decide that you don’t want to feel guilt. I can try to persuade you, but I can only really decide to not really feel guilt of my own.

But there should be some threshold of evidence you can get from others that you don’t need to feel guilty about every inaction that you take. For it is exhausting.

And given enough evidence, you should then you should truly stop feeling guilty.

For did you really spend time trying to not feel guilty and be happy? How can you really expect your friends to do something you yourself do not do?

suffering is not the unit of effort

I want the inadequacy of all my inaction to be clear, that I should be doing much more. To some extent, I still believe this.

However, mulling over this fact is probably not the single best thing you could be doing.

People say you should spend some time prioritizing happiness, because it leads to better outcomes. I’m not so sure that is true to be honest. I guess I do believe some people can be very miserable and be very effective.2

But I think it is unreasonable to expect anyone, lest alone your friends, to be doing this.3

It is just far too easy to mistake suffering as the unit of effort.

bargaining

I did a lot of journalling at some point, some of which I shared with my friends.4

If you are like me, you still might think:

i still feel like probably others suffer more and deserve more

but this is not a useful framing.

It sucks that we are in a world of sub-optimal circumstances.

It’s unfair that your circumstance is better than the circumstances of the person in the single most dire of circumstances.

But feeling too much guilt about this in particular is not making their circumstances better.5

a kind friend once told me:

everyone suffers more than they deserve, and everyone deserves more and better than they get, and that includes you, and actions taken to alleviate such things are good.

utility monsters

the same nice friend was also wrote

i’m actually the utility monster and you denying yourself reasonable things that will make you happy for some nebulous potential gain will make me infinitely sad

and i take it to heart.

As a downstream, I want you, the reader, to be happy too.

:)

I recently wrote a short story on a fictional scenario. I probably spent more time thinking about the world building instead of the story, so here is some of the notes I had on the world. Kinda scattered and rushed.

Setting

• Roughly 2037, US

• Post-AI-moratorium, biotech-heavy, transhumanist-normalized

Growing tech

• Cities have facilities where you can grow bodies / organ sacs

• ~2025: first invented, didn’t really work

• 2026: working prototypes. Animal models start to take production for meat, they are surprisingly inexpensive, some research labs start to replace their animals with organ sac versions, they can grow faster due to some genetic modifications

• 2027: started growing human organs sac in tanks at scale, Originally took ~10 years to grow a full body. Inflation is high so animal products start getting subsidized as they may potentially help with food costs

• 2028: animal production grows extremely quickly, due to some happenstance that these can mostly repurpose most factory animals with ease but for less cost (inflation concerns)

• By 2029 human production speed was also down to like 4-5 years, mass production starts to happen, there is enough evidence that it’s mostly the brain that requires the long puberty

• By 2031 it was down to 3-4 years and production beyond rich people becomes more common too.

• By 2037: ~25 months period due to diminishing returns from GPT 6 scaffolding.

• Diminishing returns since the AI moratorium and the FDA getting more risk-averse again, people have mostly been mining the same base models since 2026 with various finetunes, but it only gets you so far.

Automation arc

• High AI automation across the board, but there was a moratorium in 2026 after GPT 6 Extreme and/or Claude Mythos 5 etc.

• Datacenters can no longer do “base model” training runs, but fine-tuning opensource models on small setups is fine (up to ~100x GB200s equivalent legally) and datacenters are monitored for large runs using AI agents.

• The original base models eventually did get leaked, everyone has access to them.

• There are a lot of highly specialized fine-tunes, these automate a lot of the economy

• Robotics is unsolved due to [hand wave that it needs a really large base model and that’s not possible anymore, something something scaling laws, not sure I buy it]. maybe also some “Save Humans and Ban Robots Act of 2029” or something

• China runs basically the same models — no real model-quality gap anywhere

• Heat dissipation means with satellite surveillance it is impossible to hide gigawatt-scale data centers

• Data centers forced to split up / stay small

• The Stop AI Unemployment Act prevents layoffs — protects everyone’s income

• The Save Small Businesses Act keeps zombie companies alive even when they’ve effectively gone bust

• These get cheap, non-repayable loans from banks, which eventually got saved by being saved by the US government.

• Net result: lots of people have nominal jobs at companies that are basically shells, doing AI-supervised work for income that flows largely from government backstops

• There are also people getting stimulus cheques from the government if they are broke, but these are looked down upon, told to get manual labor jobs or to "create AI applications” which continue to exist.

• AI companies that produce the underlying models can’t differentiate or train new models, so it’s a race-to-the-bottom on profit. Also under USG grip.

• Mostly the US is rich enough that it doesn’t matter.

• US government continues to spend ~3% of GDP on foreign aid, so situation in many foreign countries continues to be kinda screwed. If they have datacenter capacity, that tends to be the main source, otherwise, they rely on kindness from others…

Bodies & Organs

Stimulation

• Still not fully understood how consciousness develops for brain-spine-only setups

• Bodies need stimulation while growing or they don’t develop properly

• For “ethical reasons” they are often given various drugs, though cannot do this all the time.

• Some weird dance-stimulation rig is used to stimulate nerve endings, and also because neuroscientists thought that it makes the CNS look more like a happy one.

Red coloring

• Organ sacs are colored so as to make them visually distinct from humans. Kinda ugly bright red color, also somewhat bruised looking too.

• You occasionally see people colored like this just because of memetics/counter culture.

• Re/depigmentation is not that hard, but also non-trivial. requires medical equipment.

• Originally it was deemed patented to try do this, but it was decriminalized — and people know there are more risks if you’re red and dead

Reserve bodies

• Standard practice that everyone has a reserve body

• This was partially due to lack of regulations in the area, some like weird loophole where it doesn’t count as cloning and is just an extension of your body, banning robotics made it so that government needed to make concession

• Initial concerns about equitable access, but rolled out fast

• Bodies are cheap to maintain — fed slop, not moving, not thinking

• Spillover effect: lab-grown meat from the same tech became way more ethically acceptable. Some people prefer to torture the original animals or whatever, but that just seems worse. Most people then just disagree.

Body design generations / fads

• Body designs come in generations, with newer gens coming out often

• Older / retro experimental designs still around. Anthropomorphic lemur is one of these.

• Current fad: “slime girl” bodies, with translucent green skin. Mostly from new tech in modifying skin texture being made legal by FDA

• People sometimes upgrade between bodies, which means going through the integration process

• There is a lack of any particularly cool things like wings, because it’s hard and much more profitable to just keep producing marginal improvements.

Brain/head swap (integration)

• Possible but really sucks

• Takes about 9 weeks to reconnect nerve endings, even using best peptides, growth medium, and specialised bacteria

• Bacteria + enzymes + stem cells (or something or other) reattach brain to body.

• The reattachment process needs you to be conscious some of the time

• Some of that time can be spent dissociated or sleeping. Most of that conscious time is spent living in your own brain with no external stimulus, for days at a time

• Toward the end you start getting some senses back

• Incredibly lonely. Some people go insane. People who’ve done it try not to remember it too much.

GeneCard

• Standard tool for compatibility analysis

• Physical metallic card. Tap two together to check matches.

• Used for organ matching

Organ legal status

• Organs of an unconscious person count as extensions of their body, not personal possessions

• This means: you can’t consent to give them away while unconscious, even via your AI assistant

• AI assistants don’t have jurisdiction over this kind of thing (per the Save Humans and Ban Robots Act of 2029)

External organ support

• Most infra is built assuming you can just replace people’s organs pretty easily, basically everyone has ones in reserve. There exist mechanical substitutes but these mostly exist to keep static organ sac bodies alive, not normal humans who live their every day life.

• Wearable artificial liver: continuous, suitcase-sized “liver box,” must dock to a wall unit ~1–2 hrs/day

• Lungs are the “expensive must-have” reserve organ — no good external substitute exists, so insurance premiums spike hard if you don’t have a lung reserve

• Liver is the manageable one externally; insurance reflects the asymmetry

Other body stuff

• Productivity / mood drugs: makes people more happy / productive / less risk-taking. Some get side effects, some don’t like that it’s not natural.

• Government provides dissociatives. “Safest thing you can be doing.” Officially: people rate their quality of life at work significantly better in studies. Unofficially: it’s just kinda wasting your time away, and it’s unclear how it got past medical review boards. People used dissociatives to get through long stretches of waiting (e.g. between integration check-ins, between organ growths).

• “Giga productivity max” people: slightly ugly purple skin to improve bloodflow, multiple neural probes into the body

• Current neural augmentation techniques kinda lacking

• Some old habits like physical input remain because more standard, and relearning new input is slightly annoying (e.g. QWERTY)

• De-aging still under research, some results, not FDA-approved

HappyHelm

• Helmet that helps protect head from injury

• Not mandatory but there’s a tax/insurance incentive for people who wear it

• This has kind of meant smartphones are no longer needed

• Has similar annoyances to whatever those that exist in smartphones now do.

Economy

Income

• Some amount of credits provided to everyone (job or stimulus cheques)

• Plus the legacy-job arrangements above

• Most people are some combination of UBI + nominal-job income

• By 2037 it’s like people are just running out of ideas, mostly the easy fruit from mining old LLMs has been extracted. People putting ever larger amounts of resources into mining further capabilities out of the existing models / mining their history.

Smart contracts

• Smart-contract favors are a normal way for people to exchange personal deals (organs, body parts, services)

• LLMs are used to draft, review, and negotiate them, and people send their prompts to eachother.

Robotics

• Some exists, but limited

• The Save Humans and Ban Robots Act of 2029 bans robot bodies and limits AI-assistant authority over physical/bodily decisions

• Additional in-world handwave: robotics needs a really large base model (per scaling laws), which is also banned by the AI moratorium

Geopolitics

• I’m not too sure. China and EU etc have the same models mostly. Scaffolding advantages have hit a relative plateau within a few years.

• Otherwise i didn’t really think about geopolitics

Daily Life

Housing

• Current housing mandates go to the extreme, people want extremely expensive high quality housing as a minimum standard, with sharing made illegal and and prices being high.

• 2 bedrooms per person unless they’re in a romantic relationship, etc.

• The mandate technically doesn’t require checking if anybody is in a relationship other than filling in a form (conditioned on both being citizens)

• Legacy homes (predating standards) exist. small, regulated minimum rent, cheap, dusty

• Common loophole: split a legacy home with a “partner” you barely know

Insurance

• Mandatory (someone died uninsured → law)

• Premiums spike hard after claims (e.g. 470 → 34,150 credits/month after a serious accident)

• Premiums depend on: time since last claim, what reserves you have for hard-to-replace organs (esp. lungs)

• Voluntary mobility restrictions reduce premiums significantly

• If you can’t afford the regular tier, you get auto-opted into restrictions

VR / Apps

• Some people are fully VR-pilled, just live in slop apps

• Apps run on a heavily optimised GPT-6, which has been running of juice since 2026 probably like 5 years ago at least, so things are mostly in phases of recreating old things, since creating new things is much harder

• Most genuinely good content that is viable already explored

• Have to wait years for new content; people spend the gap “out in the wild”

Protagonist

unproven_lemur

• Username + nickname; legal name kinda just forgotten

• Picked “lemur” because it was popular at the time, just stuck with it

• An anthropomorphic lemur (one of the older / more retro experimental body designs) Sometimes wishes they held out for newer gen versions, sometimes thinks the retro design is cool in its own way

• Furry, non-binary, transhumanist defaults

• Had a phase making math videos for a while (hence “unproven”)

• Bored with life

• Goes doing risky things, normal ski slopes too boring, goes off-piste, recurring accident pattern

• EA-coded, donates roughly half their salary to non-US causes (GiveDirectly-style) due to the fact that so little goes abroad by default.

• Would transfer up to ~2/3 of income to keep mobility, since organ-finding without reserves is expensive (the disincentive is intentional)

Job

• Works on an email app for medical startups

• Company technically went bust, but he can’t be fired (Stop AI Unemployment Act) and the shell company gets kept alive (Save Small Businesses Act)

• Actual labor done by AI sub-agents

• Wears a ring with an LED that changes colour when agents get stuck

• When it lights up he has to whip out a laptop within a few minutes and give therapy-style prompts: “keep going,” “I believe in you,” “well, that sounds really tough. Have you considered thinking deeply about a few main options and weighing them up?”

• Basically a super-therapy interface for AI agents

• Genuinely not a useful job, but he’s still managing financially. lucky to have a relatively well paying job before the Stop AI Unemployment Act kicked in.

Some of these ideas are really badly thought out. I basically hand-wave away all of the geopolitics and robotics and 10 years of progress, but I think it’s fine for a short story / to get some of the key ideas on organ sacs across

You feel foggy and slowly open your eyes and orient yourself to the room. The feeling is familiar.

“Patient has awoken,” you hear in a natural but non-human voice.

You follow your reorientation with another familiar feeling, that of annoyance.

“Bruh,” you audibly gasp. “I was so careful,” you say as you rub your hands on your face. You know this is a lie. “Ughhh I hate this.”

“Good morning unproven_lemur. You were found impaled on the slopes of Mt [California mountain]. You seem to have gone off track, and been wearing uncertified skiing equipment. Your Happy Helm Pro alerted the medical response, and time to first action was 6 minutes 47 seconds. You punctured and destroyed your stomach, pancreas, intestines, both lungs, your reproductive organs, knee caps. You fractured your ribs, damaged some neck muscle. Your Happy Helm Pro helped reduce some aspects of ischemia in the brain, yet due to blood loss there is possibly some damage.”

“Oh, not that bad,” you think to yourself.

“Most of your organs have been replaced by those of your designated reserve body, other than your liver. You have been assigned a mobility assistance device and external blood filtering system. These can be restored upon completion of your new organ sac, estimated in 21 months.”

“Oh no…”

“Additionally, your insurance premium has risen from 470 credits per month, up to 34,150 credits per month. With voluntary mobility restrictions, this can be reduced to 4,760 credits per month. This is due to: time since previous claim being low (2 days) and a lack of reserve in key vital difficult-to-replace organs: Lungs.”

“Would you like to opt into voluntary mobility restrictions?”

You sigh. “Oh no, not again…”

“Sorry, you do not have enough income for this option. Would you like to opt into voluntary mobility restrictions?”

“YES OK I OPT IN,” you gently shout in frustration.

“Thank you. Noted.”

You guess that you will need to find some way out of this mess. Waiting over two years for your mobility again just ain’t going to cut it. You’ve had enough.

You walk out of the clinic, lugging your liver box with you, and ask your HappyHelm, “Hey, order me a ride home.”

“Sure. Where is home?”

Guh. You had to get a new HappyHelm. It wasn’t set up since you never enabled backups on your old HappyHelm Pro. Now you need to sign in to all of the services again. This will take a minute…

While doing the three-factor authentication for your login, you notice your ring is glowing orange and buzzing. Work. You hope they didn’t notice you were in a coma for two days.

“Hey claude, what’s up? Answer concisely.”

“Hello unproven_lemur. I appear to have run into some issues with the mailserver. It is showing storage as being full on the main cluster.”

“OK, how would you fix it? Think deeply and answer concisely and simply.”

“Sure, let me think…”

…

“The main way to fix the issue would be to either delete old data, or to add more storage. The easiest would be to add more storage. Would you like me to do that?”

“Uh sure, whatever you think. Do that. Then please just fix the agent loop so that if you have similar problems again, the agent loop checks on it. Oh — unless it costs like more than 10,000 credits or something. Does it? Answer that first.”

“The current additional cost would be about 800 credits per month.”

“OK then just do that, thanks.”

You weren’t really paying attention. Your ring turns green again, and you go back to ordering your ride. It’s annoying that you need to wear the ring, but it’s fine. You’re mostly glad that you got to keep your income with the “Stop AI Unemployment Act”, and that even though your company is basically a shell that went bust a long time ago, they were saved by the “Save Small Businesses Act.”

You get home, and lug your mechanical liver suitcase with you to the elevator of your flat. Up to your flat.

It’s kinda dusty and old. Which is great. Because that’s why it’s so cheap. You’re sharing with your nominal “partner” (for legal reasons). You walk in and he greets you.

“Hey lemur, what’s with the suitcase? Did you get in an accident again?”

“But like I was so careful this time. I barely know what happened.”

“Yeah, you know, maybe one day I’ll believe it.”

“You like, gotta help me find another liver bro, because this just sucks.” you plead. “My body’s not gonna be ready for 28 months, so I really need a pair of lungs as well in my reserves. You know how they are with that.”

“I already gave you what I could” he responds. “We are terrible matches on the GeneCard. You might be able to ask Sam, if she can — but she’s not awake again.”

“Yeah, OK, I mean I’ve barely talked to Sam. I think we did have a [tentative?] match though. we do live together tho so she’s gotta feel some kinship…”

You go to your shared room, find her asleep in her bed connected to the integration machine. Her possessions and medical emergency info are left neatly next to her, likely by the medical staff that brought her home.

Oh yeah, she’s been upgrading to the slime girl body... I guess I forgot.

What you even know about her? Oh, she likes to study botany or something, she was telling you about different kinds of moss a few times.

You look at her medical info, and find her US Gene Card laying in clear vision. You rummage through your bag and find your own, and tap them together to check if she has a matching organ profile. You remembered correctly: perfect match. Though it’s no use while she’s integrating and still unconscious.

She can’t consent to giving you organs from her old body while she’s unconscious. For some legal reasons or something, they count as an extensions of her body rather than a personal possession. You could ask her AI assistant but it doesn’t have jurisdiction over this kind of thing. You try anyway.

“I’m sorry … Under the Save Humans and Ban Robots Act of 2029, it’s not possible …”

Eh, worth trying. It’s times like this that you wish you could have had a cool robot body instead. So you you guess you’ve gotta wait until she’s back.

How long has it been? Integration takes about nine weeks, and I guess it’s been around six or seven since she’s started.

Oh, if your memory serves right from your own experience, she’s probably getting some of her senses back. Though you mostly try not to remember it too much.

The integration process really sucked when you did it. Sure, it’d be fine if you could do the standard dissociative procedure like you sometimes do for work, but the bacteria and enzymes and stem cells, or something or other that they use (you don’t really remember) in order to reattach your brain to your body, need you to be conscious some of the time.

And most of that time is just you living in your own brain, with no external stimulus. For days at a time.

Guhhhhh… you feel shivers just thinking about it again

You really don’t get why she even decided to do the integration process. She already looked quite cute. And you don’t get the fad for slime skin — a weird translucent green.

Though on the other hand, you do see her new body, and her new face is quite cute.

No! Stop it! you don’t need a new body. You don’t need a new body!

You’re perfectly happy as an anthropomorphic lemur. Sometimes you wish you held out for the newer gen versions, but sometimes having one of the more retro experimental designs is cool in it’s own way too. The slime girl form does look just so cool though…

You kinda got bored of all this waiting, and eventually just started medically dissociating again. Sure it’s still “the safest thing you can be doing” but it’s just like, basically wasting your time away. People just complained too much about their work, and, yeah “people rated their quality of life at work significantly better” in studies but you’re still not really sure how this got past medical review boards.

One day, you get woken not by your ring, but by your liver box. Again.

“attention unproven_lemur, your hepatocyte cell cartridges have a contamination warning, and so have shut down. Please obtain a new cartridge within the next 2-4 hours.”

“Or what, will I die?”

“Yes”.

Oh…

You call your medical provider. You know how annoying insurance is about urgent-vs-non-urgent care, and technically this would fall under non-urgent care, so you end up needing to get a ride to the hospital, get a new cartridge, answer work paging from your ring, it’s a bit of a pain. But it’s fine. You hate it, but you saved like 2000 credits this way, so it’s worth it.

You get back home. After a week you’ve had enough of dissociating. Surely there are better things you could be doing? You really want that new body sooner. You start prompting asking your friend AI what you should do.

“Well, you mention she likes moss, could you get her some nice moss? Maybe ask her agent about it. In the mean time, I’ll start coming up with fair deals to propose”

“Ok, but I know literally nothing about moss, what would I even get her? Ask her friend AI, think deeply, answer concisely.”

…

You end up calling up a friend, going on a trip to see them, spending a few hundred credits on flights, with extra fees for mobility assistance for that annoying box you need to carry around, and end up getting some obscure moss that grows in Europe. To you, it looks exactly like moss. You hope you’re not getting scammed again.

When you get back, you spend a few more days tired dissociating. Time flew by pretty quick. You were woken up a few times for work, but barely remember it.

Eventually you get woken up by yet another machine. This time Sam’s integration machinery:

“Integration process complete. Waking up failed. Please wake patient manually.”

You feel some genuine excitement to get up and go to wake her. You hold her by her shoulders and shake her.

“Wake up!”

She woke up. Her green skin looks so smooth and is a soft texture between that of skin and silicone.

“Hey Sam, you’re awake.”

“Yeah. There…” (Her speech is still kinda weak and slightly slurred…)

“Can I please have your old body?”

“Uh, Hello? What?”

…

Ok, you were a bit impatient. You let her orient herself.

“Ok lemur, Hi. You know was just here doing integration. You’ve been through it too. You should know how terrible it feels.”

You feel a bit flustered, and answer, “Oh yeah, how was it?”

“I can barely move. Everything’s so vivid again... It feels so nice to be able to interact with things again… I’m never doing it again though…”

“Was it worth it? Slime girl Sam does look very cute.”

“I hope so.” She slowly starts to get up and move her body, walk around your shared room, looks at herself in the mirror, and gasps.

“Oh my god…”

You let her enjoy herself for a while.

“OK well, I have an ask. So even though I was like so careful, I still managed to damage my body again. Can I please have your old one. Please. I’ve gotta do some more work, but can you please think about it? You’re basically a perfect match. I even got you some moss.”

“Ok, I’ll think about it, send me your prompt.” she says dismissively.

You send her your LLM prompt, and kind of tired go back to dissociating for a while, while she thinks about it and gets used to hew new body. You start falling asleep and hear her say “Wait, you got me some moss?”

You awake again, prompted by your work ring. Answer some annoying questions again. Then while awake, go see Sam.

“Hey!”

“Hey! I really love this new body. I doubt I’ll need my old one so you can have it in exchange for a smart-contract favor. And yeah, thanks for the moss.”

“Oh, and yeah, we can go on a date some time too.”

(you don’t remember asking for a date but you do remember thinking about it). “Ok, I’ll get my LLM to look it over, but sounds like a deal”

You get your new body, book your surgery, and plan your next ski trip. You’ll be so careful this time…

This was an attempt at doing some relatively fun world building over the course of a couple days, assuming AI base training and robotics stops with whatever we have in 2026, and biology progress is relatively unhindered otherwise.

Not that well thought out, I spent maybe a few days thinking. Ideas are mostly optimized for interesting story with organ sacs as a focus. There are a lot of things that need to be kinda hand-waved away.

Feel free to critique the hastily-put-together accompanying world-building document

I love Claude Code.

In March, I took a break from work responsibilities for myself, partially to recover from not being able to speak. Partially just to finally have some rest.

In this time, I got addicted to Claude code. I would spend literally all day just prompting with multiple tmux tabs.

I would just have questions I want to know about, things I want to understand better, and I would just ask Claude Code to research these.

You can view all of my outputs here:

• https://slop.sus.cat

Or you can skip to some examples below.

You could do this too.

Examples

I asked Claude to analyze a tonne of skin health things, and the wiki there I still find kinda useful. It's also just one of the more beautiful websites Claude has made for me so far.

Asked Claude to design me a bunch of websites in a bunch of different styles so that it could use these as reference samples for future projects.

I found this description of decision theory things pretty useful as something that tied together the many pieces that I knew

Your favorite catgirl is not smart enough to figure out how to use ‘s website so I got Claude to make me a sitemap

There are a bunch of Hackathons that Claude Code could probably help you enter

I asked briefly to compare between apple silicon chip performance because there are too many chips now. Plausibly there are pre-existing better sites for this

I asked about geopolitics because I didn't know what was going on at all and don't like following the news

I got Claude to analyze the transcript from a day I had in 2015 where I didn't do that much. I also wrote about the day. It’s one of my best posts >~<

I also asked for an index on all the different authors and writers in the EA/Rat sphere.

How did I do it?

The Software Setup

My current Claude Code setup, is to have a hosted development box, and within that to have a docker container which just has tmux, and claude code in tmux sessions.

I then have the files hosted using nginx, which is pointed at a specific folder with visible outputs, and which has rules on which files can be shared vs not.

You can easily pay for a VPS, such as on Hetzner

The Prompting Setup

I often want Claude to really deep dive when doing research, then I want it to display the information in ways that are accessible.

My current setup:

• I asked Claude to do some research on how best to get Claude Code to do research.

• I then asked Claude to set up some basic prompting guides.

• I prefer:

  • Have a main folder repo for an overview page

  • Have folders for main topics, sub-folders for each project

  • Use git to save steps regularly

  • When saying various claims, make sure to state confidence of those claims [completelly speculative VS established fact-checked info]

  • Make sure to save most important sources

  • Use subagents liberally. If needing very large tasks, spin up tmux sessions with claude running within them

  • You can spin up multi-iteration agent time loops, but if doing so, make sure there is an end date so that things don’t run indefinitely

• I also find the default style of claude websites not that good.

  • Use impeccable to avoid some anti patterns

  • Ask claude to spend time researching and designing a suite of template web pages

  • Get some preferences in, I prefer light-mode pages with toggle for dark mode as an option

• Set up tools for it to make things easier

  • Ask to understand color science of design

  • Ask to build a suite of tools for scraping webpages or rendering markdown into html

• Ask for a good overview page with timestamps

  • Have categories for the things that you do, and have tools to sort by [show how big the repo is] or [how recent it is]

• Have main guides for Claude

  • You can get Claude to thoroughly research something with subagents, then when done, critique the outputs, then build a webpage, then add the webpage to the index page, with sub-agents, as a single step

I find this works pretty well.

Sometimes I run out of credits, and end up using OpenAI Codex instead. There are probably other optimizations I could do. But I currently find it relatively low effort.

(Codex is excellent for image generation, maybe I will write about it another day)

I recommend sharing the things you do too.

You can see my full list of outputs here

People brought up that there is voluntary pain when I argued “is suffering axiomatically bad”. I agree in the existence of voluntary pain being a thing, and something I have experienced.

I will try to now get a more thorough framework about how I think about it.1

In the previous post, I wrote:

this is one case where language is a bit inadequate.

I think it is very possible for “suffering” to be good. There are two cases for this:

• “suffering” in which states are described as negative, but which are still positive valence. One example of this is the burn one feels from spicy food. This still feels good and is pleasurable, despite nominally having aspects which are described as bad. […]

• “suffering” in which one was actually in pain/suffering at the time of the even, but which leads one to better mental states after the fact.

I have experienced both. Suffering can be a way to describe this, if the experience is also either positive-valence, or leading to longer-term pleasure, then I’m not sure it counts.

I will most focus mostly on the first of these.

Types of Fun

It is worth, for reference, knowing that there are 3 types of fun, and it’s worth knowing about all three.

• type 1 fun = enjoy now and enjoy looking back

  • E.g: going for a nice walk somewhere scenic

• type 2 fun = not enjoy now but enjoy looking back

  • E.g: doing something difficult, like intense sports, but looking back you enjoy it, and would do it again. (challenging, etc)

• type 3 fun = not enjoy now or looking back, but it’s at least a good story.

  • E.g: You end up in a dangerous tedious scenario. You may or may not regret doing it, but would not do it again.

This can be broken down into axes like “experiencing self” and “remembering self”, as well as some vague degree of regret/endorsement.

I think this has been useful, but I find when it comes to some degrees of masochism, it is also maybe useful to categorize types of pain enjoyment.

Enjoyment of Pain

Before categorizing pain, I think it’s worth having a disclaimer: The same physical sensation can be experienced very differently depending on the context. There is like a default “bad state vs good state” that people can apply to things like physical pain, but often this can be reinterpreted.

Some examples of bad pain: feeling subtly too cold and uncomfortable. Feeling physically ill. Having chronic pain.

Some examples of pain that are sometimes good: spicy food, pinching, biting, scratching, extreme cold after extreme hot, cathartic crying.

Even the “good pains” are often pretty subjective. Many people, do not experience these as enjoyable.

General mood and state of mind also affects it hugely. One factor for this is [did I consent to this] but there are others. This applies to all levels, but for higher levels of pain it becomes more difficult to enter a mental state where you can enjoy it.

Levels of Pain

level 0 pain - (Type 1 fun)

This is relatively light physical interaction, rubs and light scratches.

Very often nice. low to moderate pressure, not enough to be described painful. This is just like, pretty normal, can be exciting but less so.

level 1 pain - (Type 1 fun)

Described as slightly to somewhat painful, but still just feels directly physically nice.

It’s like level 0 but more intense. This depends on pain tolerance of the person. In terms of net physical enjoyment, this is often where the peak is, it is slightly more exciting than level 0 too.

level 2 pain - (Type 1 or type 2 fun)

This is just straight up painful.

Usually the peak direct physical enjoyment was reached before this level, and often it can be the case the that physical enjoyment is less than the pain, but there is some additional excitement from being at this higher level too.

There is usually a feeling of needing to flinch or resist the pain. Some of the enjoyment is from needing to put in effort to resist flinching, or from resisting and being overpowered.

If in some moods where you haven’t done this in a while, it can often be type 1 fun. If done in other moods where you are maybe too sensitive, it can still be kind of fun, but less so, more like type 2.

level 3 pain - (Type 2 or Type 3 fun, or just not fun)

Very high pain / consistency in pain.

Being here long enough you can often start to break down crying.

At this point it becomes very difficult not to resist as hard as you can. The direct physical enjoyment is mostly gone now. Any enjoyment at this level is mostly from mental games. As such, getting enjoyment from this is the most difficult.

There is some battle of [just give in and accept the pain] and [no I need to fight back], which makes this difficult. Feeling like you are able to just give in and stop resisting can make it feel good. Being in a mood of particularly low self-worth can make this easier too.

When done right, this can feel very freeing. Letting go of free will and just submitting to the pain, letting go of the responsibility to the world for that moment, or feeling like you are getting the mistreatment what you deserve.

When done wrong, this level of pain could easily be traumatic.

The levels can be positive or negative valence.

All the above are mental states where the positive valence is achievable for some. It is also worth knowing that there exists pain that just becomes no longer fun. I try to describe how level 3 pain can feel positive valence, but many experiences like this can be negative valence

Some examples:

level 2: I think I found some types of journalling I did to be quite mentally difficult, looking into my own soul and plainly stating that which I feel most shame and embarrassment about, making it ready to be shared with others. But it was still kind of cathartic in a way when doing it.

level 3: I remember getting electrolysis hair removal, and it was at this level of pain. I had a semi-breakdown, and it was not enjoyable. I came back another time, with numbing cream and ibuprofen, and it was probably more like high level 2 or low level 3, but I managed to give in and submit to the pain, and somehow enjoy the feeling for a while. It was hard to replicate though.

Closing

I think about these levels of pain often, and I guess the graph of how people experience them might differ a lot, even for the same person as they are younger or older or in different contexts.

I think overall, it is quite difficult to enjoy level 3 pain, and often overcoming it makes it just become another type of type 2 pain.

• Week 1: No speaking and slightly lonelyish and a bit stressed

• Week 2: No speaking and semi-social and a bit stressed

• Week 3: No speaking and more social and more chill

• Week 3: Speaking and quite social and chill

Day Before

The day before inkhaven, I was flying into San Francisco. It’s a long flight to Ireland, and I hadn’t read many of the bios people provided. So I copy pasted all the bios locally and got Claude to help me download most of the linked posts people had in their bios for the plane.

I spent a third of the flight sleeping, and two thirds of the flight on Vyvanse just intensely reading people’s posts. It got pretty tiring, I read like 30 bios so around 60 with short notes and maybe like 5-10 that I didn’t finish. There were many kinds of posts of various lengths. I used the Firefox TTS + MacOS Zoe voice + headphones to read. I took brief notes and tried to get an idea of who I wanted to talk to.

I arrived, and got into lighthaven, and started seeing some of the people, and it was interesting. Many started asking about how I couldn’t speak and I used my TTS. My claude had run out due to the token caching bug. I speak to people and have no idea who wrote what post ngl.

Week 1

It was OK. The first day I was very enthusiastic with how many posts I could write, I wrote like 3. People often asked me the same questions so I wrote an FAQ which became my most read post for a while. I guess it makes sense.

Then afterwards I started to run into struggles on how much I could write in a day. I would start writing effortful posts that would need much more time than I had. sometimes I would scrap the draft, and other times I would split it up or cut it down to a more manageable post.

I continued to struggle with not being able to speak, it was hard but I just learned to completely leave any time a conversation became a group conversation. It was just too annoying to be have people ignore your TTS or have the conversation move on before you are finished typing. People would comment “oh it’s more clear when you have the combined single sentence” but this trades off that there is much more latency.

Writing with a whiteboard was better.

I ended up mostly just being one-on one with like 1-3 main people I liked most, and occasionally have a one on one with someone else. It was just kind of OK.

I was also kinda stressed that there were events happening often and I would keep missing them.

Some posts from this week (normal text = quick post, itallics = slightly better post, bold = “one of my top posts”)

• Dying with Whimsy

  • My current life philosophy I guess

• carbon offset arbitrage opportunity

• Not being able to speak FAQ

  • This was very action relevant but not super timeless

• Two Theories for Cryopreservation

• My Weird "Free" Legacy Phone Plan

• Comparing Communication Methods When You Can't Speak

• My Ethics + Is death and suffering axiomatically bad?

Week 2

This week I then got a bit better at being assertive in some ways, but my writing schedule also got kinda worse. My buffer quickly became sub-1-minute.

I started to meet up more with people I knew before and people from twitter. This was nice, there is the knowledge that you specifically organized to meet them, so it feels less like I am keeping people under duress.

People would find it interesting to interact with me when I couldn’t speak, to be on a whiteboard or laptop TTS.

One person commented while we cuddled: “It’s quite relaxing, because I spend time with you and it feels like I’m spending time on my own, you’re like a pet”.

A very fun interaction happened in a San Francisco Walgreens. I coincidentally ran into someone who I had only ever met once before briefly, at a contact improve event. I took out my whiteboard, which fittingly had “Yay” drawn on it in large text. It was quite fun. We then met up again next week.

I also just started to ask more people for headrubs. I got more head rubs.

I just love headrubs so much >w<

My posts for the week

• 1 Week of Inkhaven Review (so far)

• Linear vs Non-linear Probes for Interpretability

• My Janky wall-mounted eGPU Build Log

• Journalling: Clear up your thoughts with vulnerability-slop

  • writing and sharing things you feel guilt about make you feel way less bad about them. How to do that?

• Does Capitalism Stop Renewable Build-out?

• Is life better as a Man or as a Woman?

Week 3

This week was more eventful but not so dissimilar. I met some people I knew outside of inkhaven again which was quite.

I finally decided to get my draft done on AI stuff I was thinking about.

One frantic day I had divine inspiration to want to write some fiction content because it was funny, on “AI water use”.

I continued to publish 2 posts per day and it was getting tiring, and I would consistently post at 23:59 all days this week.

We went on a trip to bodega bay. I had some more less sociable days there. It was kinda hard to find good spots during the day, but it was more chill during the night.

I got some headrubs and cuddles from some people I hadn’t gotten to know much, and heard a bit from them. It was nice because people didn’t feel guilty yapping to me about themselves from few whiteboard words written since they knew I couldn’t speak. And I just absorbed headrubs and listened and it was nice.

It often would be towards the end of a day, and maybe one of my posts was done, but the others would be not really done. So I started to rely more on my back-catalogue of old journals and vague ideas. It was OK sometimes.

One hit post was one looking at a 2015 journal I did.

I also did some life debugging with two friends on a google doc, and wow I should have been doing way more google doc conversations when I couldn’t speak. Missed opportunity.

Posts for the week:

• Axes of Planning (in AI Models) + Lit Review of Planning in LLMs

• The Last Drop

• Fear of Ink

• I will waste my childhood, and I will be happy

  • I show a timestamp of a day 11 years ago

• My little paradoxical reactions

• Recollections from Primary School

• Was I conscious when I was younger?

Week 4

I could finally speak again!!!

I went to Seattle to meet a friend from twitter, and that was extremely fun. I got to spend some time getting to know her a lot better.

I got the 7am train to Portland and met with people from Nectome in Portland. That was quite nice, I met Aurelia and Charlie and Jessica and they were all quite interesting and passionate about quality cryopreservation. I would have liked to see more of the facilities too but I didn’t get around to it sadly.

I got the 1pm Coast Starlight train the same day to Oakland. It was quite fun, I sat with one person who was fun to talk to, then another person I didn’t talk to much because I was writing and sleeping. Some of the views were very pretty.

I then arrived, was still tired and slept the whole day.

The next day was the Inkhaven Fair, which was quite fun, I wrote a vignette that seems to be quite popular.

Then I continued to meet more people from the internets and write my 2 posts per day, and actually could join in on group conversations, spent more time in the winner’s lounge. it was very fun. I met more people I hadn’t met before on twitter. It was overall a very fun week.

I love my life

My posts for the week:

• thought experiments on continuity of consciousness

• Guilt is Free, Pride is Impossible

• You should be vegan, and if not, at least offset

• Cryopreservation: How Good is Good Enough?

• 8000 hours of ineffective youtube viewing

• Bay Area Cultural Victory with Cat Ears

  • Vignette of how bay area transhumanist vibes.

• This post >u<

Final Days

It is interesting to have the days come to an end. It feels long and short. I love and hate writing. I think 2 posts per day is too much, but it is also what I needed when I couldn’t speak. I still have too much I want to write about, and not enough time and energy to write about them. There is a prediction market tracking on whether I will succeed, it seems maybe under-priced still relative to how things are going (28/30 days completed but only 83% chance). I have loyal share holders though.

I remember the first week thinking “oh 23:59, what an arbitrary time, I don’t need these restrictions”, yet I have consistently, like 18 of the last 19 days, posted my second post between 23:55 and 23:59. The odds seem too high, 40% that I publish before 23:50 today? seems unlikely, but this prediction market is pretty low liquidity.

I feel like being able to speak, I have only just started to get to know people here more, and now I am about to leave.

But I persevere.

In some ways, I think I could have had clearer goals. Did I want to work on what has the most career capital? Or to hone my craft? Or to work on my passion project that has less career capital? Or focus on making friends?

I ended up mostly focusing on the latter two, and I guess I have less ugh about writing “non-perfect posts” now. I wish I had more time to do the writing competitions that did exist here, but alas, one cannot to it all.

I still have lots of things I want to do after, and it’s so unclear what is best now.

You show up for your first time in the Bay Area. You’ve heard wondrous things. The trans-humanist ideals. Galaxy-scale terraforming, curing aging and disease and bringing people back to life. Experiences and foods you could hardly imagine. People could embody life as catgirls or dragons or capybaras.

You walk through the open doors of a suburban old-timely building. Pale paint, colorful highlights. The place is…. nice, I guess. It’s clean and renovated and nicely laid out. There are some IKEA couches and carpets and fake plants. Even a few nice LED lamps and strips.

You grab a bag of plain salt chips plus a vanilla Soylent from the fridge, and walk up to the nearest small crowd, who are speaking passionately and with interest. Some of the twinkish guys are wearing hoodies and jeans, while the others wear a t-shirt and sweatpants. They have the same basic haircut, plus or minus a step on the Norwood scale. One of them has started taking a finasteride recently.

You excitedly listen to the people whose posts you have read online.

“So where did you get your Reta from? Do you have any spare you could sell?”

“No, I left it back home, I just chose a supplier from Finnrick. They aggregate purity test results for most of the grey-market sellers”

Wait… OMG, could it really be them?

You recognize them because they had a slightly interesting hat. They had that cool LessWrong post recently. You excitedly listen to them for a while, but are too shy to make conversation before they leave.

You disperse and hear another group. They all look similar. But you are no longer the only girl this time.

“Yeah it’s great, I got Claude Code to build me an automation. Whenever someone likes my post on Substack, I get notified via a message on discord, oh it’s broken one minute let me fix it….”

The evening goes by. These people get it. You love every minute.

Or so you would have thought in the past. You go home.

You go back to your room, and go back to watching Pluribus. “Hah, this is so mundane. The hivemind has all the intellect of humanity, yet spend it so mundanely. Drinking slop, sleeping in warehouses, the world looks so plain. Where is the flavor?”

Yet you are left with an inkling feeling. You showed up to, perhaps, one of the most trans-humanist spaces in the world. And the best they have is IKEA bamboo and tech branded hoodies, just like the rest of the bay area.

The people are kind and thoughtful, if sometimes awkward. They listen to your points and consider your counterargument carefully. They are so interesting sometimes.

But where is the fun? Where are catgirls?

You come back much later. As you grow and change, this place stays the same.

“I like your ears” someone says as you walk in, wearing an outfit that mixes elements of Japanese and Western clothing. You can’t speak, so need to whip out your laptop, and start typing.

“ty” gets converted to “Thank you” and gets vocalized with the local TTS model.

“Oh yeah you were the one who can not speak. I read your post.”

You have become the fun cat-girl you wanted to see.

Somehow by just asking people, you have also monopolized the head-rubs market, almost without action or thought. Some think you are slightly weird, but most think you are just kinda based.

But you have a vision for things to be better. And so you scheme.

“Yeah I would totally be a catgirl/catboy, but that seems like it would be kinda embarrassing. What would other people think?” This is the 4th person you hear say this today.

You have your opinions on cat ears.

Japan’s finest cat ears are, of course, produced by Assist Wig. The “Animal Ears With Wire　Binaural Set”. These ears are thick, have hair-like texture, and a 3-dimensional curve. Yet they are still subtle enough to be classy, and clip-on to be minimally interfering. You’ve worn these daily since going to Cospatio in Akhibara.

Days go by, but it gets lonely being the only catgirl, and being unable to speak. Though it does make conversational dynamics more interesting.

I wish more people lived up to their ideal selves, tried to make things interesting too.

You become close friends with the cute goth girl who stays here too. But it’s not enough.

You slowly make friends with others too. What you used to find so fun, you look at and now feels so mundane. They are nice and friendly and give you head rubs. Are there things you could do to reduce the mundanity too?

You test out giving someone your spare pair of cat ears, and they love it. Hypothesis confirmed. There is a fair coming up soon, so you start scheming.

You go on assist-cosplay, and with tax and shipping from Japan, you manage to spend over 400 USD cat ears.

The day before the fair, your delivery has arrived. 25 pairs of cat ears ready to be given out. You have found the most promising candidates, and begin to distribute them to these people.

What was originally a day to “celebrate writing”, you have successfully co-opted into also being a catgirl and catboy awareness day. Some people are skeptical, but soon enough, you begin to run out of ears. People begin to comment “hey, could I get some cat ears too?”. Others notice too “Wow, so many people wearing cat ears, it’s quite cool”.

No longer is being a catgirl an endangered class, but people see it as almost a new normal.

Almost a new normal.

As the day fades, people leave. As the day pass, few continue to wear cat ears. Though some rare few continue to wear them, many choose to keep the cat ears to be in their personal stores, maybe reserved for some few days such as Halloween or something, or maybe a themed party.

But maybe at least one or two could be inspired. And maybe a brave few will carry on the torch as you leave.

Commentary

Ok, to be clear, I still think Berkeley and SF are just one of the best places in the world to be right now. Things are happening here, people care to get things done, you can feel the AGI as you speak to most people. Many places are quite nice and care to understand things, there are outdoor areas with greenery, but I feel like there is still so much missed opportunity.

I get people are busy and working on real things, I am glad for this. I know it’s “just aesthetics”, but it’s also a kind of lack of caring. And I find this kind of sad.

When you go to places that would easily be ranked as like, top-10 transhumanist place on earth, you expect something cooler than just a nice work space. Not everyone needs to be a catgirl. But it’s so boring when most people here can imagine abolishing death, but not imagine being slightly embarrassing. At least it’s nice that there are at least a few people trying things.

As I mentioned in my minutely 2015 timelog, I spent a huge amount of time on YouTube. Categorizing my whole life so far, it’s probably #3 after sleeping and school.

I guess approximately ~5% of my life was spent watching YouTube

I have over 8000 hours of recorded YouTube video history watched, and this excludes phases where I didn’t have YouTube watch history but still watched a tonne of videos. this excludes non-recorded time, which probably puts the total somewhat above 10k hours.

I learned a lot, but I also just wasted a lot more. And I reveal my excellent cringe-maxxing YouTube strategies.

I’ll let you know exactly how!

Techniques and Caveats

For this analysis, I did a google Takeout, and used Claude to analyze my data. I also add some personal comments on what I liked most (and some of the most cringe parts too…)

Some caveats:

Before 2015 I think I probably used a different account, or there was some migration, which just means that my history before then didn’t really exist.

Also, starting around 2021-ish, I had a “privacy phase”, where I stopped using the official YouTube app, and instead would use NewPipe or Youtube Vanced or GrayJay. These are nice, I still use GrayJay app but signed in, though for some of this time I didn’t end up watching things.

This was successful at making me watch less YouTube (these would show subscribers only instead of recommendations).

For the most part, I basically always watched most videos to the end until some time a few years ago. As I mentioned before, I am not really sure I was conscious when I was younger so I never really considered stopping a video part-way through…

(Yeah I was a weird kid I know)

But the algorithm was also just like, pretty good at getting me content that I wanted to see. Often to my detriment.

I was addicted to content. I needed more content. I would non-stop watch content. I love content <3

I did later start to gain more consciousness in my early 20s, and start to not always watch videos to the end if I noticed I wasn’t gaining anything from it.

Also, a huge note is that I did watch things on variable playback speeds. I hyper-optimized watching things at barely-legible speeds so I could consume more. (I loved content <3). It started with 1.25x, then 1.5x, then there was a while where 1.75x didn’t really exist, but that got added and I upgraded to 2x, then with extensions it was possible to go beyond that. I think I’ve done this since late 2010s at least.

Most Cringe YouTube viewing

What did I end up listening to the most on YouTube (rather than YouTube Music)?

Ok yeah I mostly listened to Minecraft parodies. You know your girl loves minecraft parodies. I know all the words to at least 3 Minecraft Awesome Parodies songs, and bought the album on Google Play Music before they shut it down.

I didn’t really listen to much music for most of my childhood other than things that caught my attention specifically. I might write about this another time.

I watched only a few categories of Youtube mostly

No matter how you split the data, it follows a power law of some sort. Everything is continuous when the numbers are big enough1. And 30k videos is enough that it becomes continuous.

Or we can also look at individual channels to some extent.

Basically, I ended up mostly watching “Science and Technology”. Which splits into “Technology” and into “Science”.

Technology

The most prominent category, by far, was watching LinusTechTips.

I watched 1,971 videos over the course of 904 hours. This excludes that I also watched, eg: TechLinked, another Linus Media Group channel for 546 videos, or 70.4 hours.

What can I say, I love tech >u< those blinking lights and moving electrons and water coolers and many many FLOPs and high transfer speeds and such are just so so fun…

I love computer.

You can see in the whole 2014-2026 period, Linus Tech Tips has been a factor in basically all of it. Over 1000 hours of my life has been dedicated to Linus. This would be 20 hours a week for over a year.

I of course watched many other tech Youtubers too. NCIX Tech Tips then TechLinked, ShortCircuit, Marques Brownlee, Hardware Canucks, ThioJoe, Techquickie, Dave2D, ETA Prime, The WAN Show, The Friday Checkout, Austin Evans, Mrwhosetheboss, TechAltar, Jon Rettinger, amongst others that didn’t make the top 100.

This later became more linux hardware ones, such as ServeTheHome, Lawrence Systems, Jeff Geerling, but also general linux and privacy channes, such as TechLore, Gardiner Bryant, The Linux Experiment, Mental Outlaw, Luke Smith, Naomi Brockwell TV.

Some programming channels too.

I watched so much, I could give detailed opinions on each of these now.

The only ones I tend to still watch sometimes now, are Linus Tech Tips, Dave2D and Marques Brownlee (in part for the nostalgia factor), and I most enjoy watching The Friday Checkout / TechAltar.

Science

Another main branch of videos I liked to watch was more science and generally educational ones.

My favourite for most of this time was PBS Space Time, which posted a high-quality 10minute video about physics topics each week. I loved it and waited eagerly for new videos. I watched since before they switched hosts, and it helped inspire me to study Theoretical Physics in university.

My other favourite, which I learned about at a similar time, was Isaac Arthur, who posts weekly videos on Science and futurism topics, mostly explaining the physics behind sci-fi worldbuilding. I religiously watched this for years and years. Over 200 hours watched since ~2015 when I got interested. Eventually, after many years, it felt like the fraction of new content became low, so I kinda eventually started to lose interest. But I loved it for most of that time.

Otherwise I watched many other videos. Veritasium was really good, Vsauce was excellent too. Kurtzgesagt was quite nice. I liked Flammable Maths, and various PBS channels, 3blue1brown, tibees, ex1rbia, Cody’s Lab2 NileRed, Economics Explained, Up and Atom, Physics Girl, and Sabine Hossenfelder, Stand-up Maths, acapellascience, blackpenredpen, Simon Clark, Andrew Dotson, Tom Scott, Logically Answered, Joe Scott, Half as Interesting, Healthcare Triage, AsapSCIENCE, Minute Physics/Biology/Earth, DIY Perks, etc….

I could give detailed opinions on basically any of these too….

I watched phases of gaming content

I had a good few phases of watching Gaming content.

The first was watching Survival Island series on Yogscast, I barely remember this much, but it was the first.

The second was watching Captain Sparklez when I was quite young. I loved the skyblock series in particularly, but I did watch others.

Then I kinda quit Minecraft content for a while. I would watch general channels like The Game Theorists sometimes.

I then got into watching Hermit Craft during Covid, mostly Mumbo Jumbo and Xsumavoid and Grian.

Then I learned about ilmango and I loved his content.

I also did often watch content about whatever game I was currently playing. Probably a favorite for this was Inscryption a few years ago.

Other Phases

Philosophy

I had some phases where I watched philosophy too. The first was Crash Course Philosophy I watched in 2015 or so. I then ended up watching a lot of other creators, most memorable is Alex O’Connor.

Personal Finance Phase

I had a phase of being very frugal. I was naturally inclined this way and it probably got me to update too much in this direction too. I kinda moved on from this, and eventually did start actually investing in index funds but I was too scared for a while too.

Programming Phase

I watched some tech programming Youtubers. They told me to learn to code and do internships, so I did that. Many videos were cringe, but it was good advice anyway.

Vegan Phase

I had a really strong vegan phase. After I went vegan in 20173 I really loved nutritionfacts.org and watched so so many videos on many topics, and also loved various other vegan channels too. I eventually stopped being so interested in keeping up with this.

Entertaintment

I watched other entertainment, but idk, I am running out of time to publish so I will spare you the extensive list of things to save you skimming the list

It had it’s pros and cons in my life

I basically stopped watching youtube and instead ended up moving to listening to more books, listening to podcasts, using TTS to listen to articles, and such.

I guess I often didn’t really watch the actual video content, mostly the listening (depends on the topic). And to some extent, you start reaching a limit with what you are able to watch online on YouTube anyway.

My favorite activity these days is to just use claude code, which will be let me be hyper specific to learn about whatever I want tbh.

In some ways, many of the best things in my life I do owe to YouTube, so it seems better than other options I had when I am so averse to reading (maybe a future post), but I also still regret that I didn’t stop watching it sooner and do better things too. Like, probably at least 80th percentile good compared to other options I would have realistically taken in Ireland at that time though.

I had fun, I guess I learned of some good paths, I guess it meant I didn’t take even better paths sooner. But I’m sure I will look back to the present and think the same about it too.

I recently visited some of the people working at Nectome in Portland, OR. I learned a lot about cryopreservation quality from them. I try to fact-check somewhat, but I am not an expert in this area. (I may edit this in the next few days with feedback from people)

I will discuss some about blood reflow, and how this affects perfusion quality, and also explain how CT scans are probably not good enough to determine quality.

How good is good enough?

In short, we don’t know what future technology looks like, or how information is stored in neurons, or what causes it to be destroyed or not.

It might be that you need really high preservation quality or the information to be usabale, perhaps at or exceeding current standards in neuroscience. Or, it might be easy to reconstruct the original state after the damage with some future nano-tech or AI or something. Nobody really knows.

Given uncertainty, it seems better to aim for as high a quality of preservation as possible, but what is that according to current standards?

Perfusion

The key desiderata, are that we want to keep the cells in a state that is as close to the original as possible. If one were to freeze the cells normally, this would typically lead to ice crystals destroying the cells and leaving them quite damaged. Is this fine? Nobody knows. But it seems much worse to allow this to happen, than to prevent the cells from being damaged.

As I discussed in “Two Theories for Cryopreservation”, there are two main ways this can be done:

• Vitrification, which replacing the water with a fluid that doesn’t form crystals the way ice does

• Aldehyde Fixation: essentially turning the cells into a plastic-like state by cross-linking all of the cell pieces chemically.

Both have pros and cons, but both only work if one can get the fluid to the cells that need to be preserved, especially the neurons in our brain. This mostly involves pumping the cryoprotective agents though the arteries in our existing circulatory system, and hoping that is sufficiently in-tact that it gets to all of the cells.

So how viable is that?

No-reflow phenomenon

If we look at the brain, we can see that, like most of the body, it has some main arteries, which branch off into smaller arteries, which then branch of into many capillaries flow blood through the brain and which are used to ensure our neurons get a continuous supply of oxygen.

Capillaries are incredibly small, such that red blood cells need to physically deform to pass through them. This means that any loss in size of capillary diameter makes it difficult-to-impossible to get blood flowing through them.

When your heart stops, your blood stops moving, and your cells quickly being to respond to the lack of oxygen, often leading to changes to the size of your blood vessels. The different sizes alter the dynamics of how much blood flows through different parts of your body.

And so, a key challenge is with trying to get cryoprotective agents through your circulatory system by getting your blood moving again.

There are many factors that mean an ischemic brain kills itself. Claude tells me there are like 30-40 different factors1 and there is a large literature on this. And the only known way to reliably affect all of these is to have a lower temperature.2

An important way this matters to us, is that ischemia causes capillaries to contract and become smaller. There is a time window where it is possible to reach the whole brain, after which the capillaries become too small such that pumping blood or cryoprotective agents through the body just doesn’t reach most places.

From a paper3:

capillary constriction at pericyte locations was seen in coronary no reflow in a rat model of myocardial ischemia and reperfusion. They noted that 40% of capillaries had no reflow at pericyte locations, where capillary diameter was reduced by 37%.

This ~60% decrease in cross-area thus significantly constraints cryopreservation quality

Can we see whether the preservation quality was good?

We can use tools to try probe whether perfusion of cryoprotective agents into all of the human neurons was successful.

The first way we can try to do this, is via a CT scan.

One can look at the density of various parts of the brain, and see whether there are large areas and look at denisty to see whether the cryoprotective agent has successfuly diffused to all parts of the brain. This only has a resolution of about ~0.5mm so can only give a rough area, but as mentioned in recent work by “As far as preservation quality goes, good CT scans are a necessary condition, but not a sufficient one”

Neuron sizes are much smaller than 0.5mm. According to this nature article: “The synaptic cleft represents the intercellular space of approximately 20–30 nm” and so we need a much much finer resolution to determine whether the micro-structure of the neurons is in-tact.

Another method one can try is light-based microscopy.

Light-based microscopy tools can be used to see supplement seeing how well smaller structures are cryopreserved, but the resolution for this is diffraction-limited to ~200nm. One can use this as another necessary condition for assessing quality, but yet again, it is not sufficient.

Lastly, we can use the most expensive method, which is electron microscopy.

This has a pixel size of more like 5-10nm, and so for the brain preservation foundation prize, one explains that for their testing “sufficient number of high-resolution (5nm pixel size) micrographs will be taken to provide a convincing case that the entire surface shows good ultrastructure preservation.”

So how well do current method work?

My understanding from the conversations with Nectome, is that existing methods for cryopreservation can often give results that look good with both CT scans and light-based microscopy methods, but when looking at neurons using electron microscopy, one can start to see issues more clearly.

Using aldehyde based-methods, Nectome recently established “14 min is the approximate length of the perfusability window”. Beyond this, one may get reasonable light microscopy results, but stochastically whole parts of the brain may not be correctly cryopreserved.

One can look at recent research by Nectome and see a visual depiction of how these differ:

Going Forward

I agree it still seems worthwhile to try save people who have died suddenly, even if preservation quality might be affected. However, it seems valuable that Nectome is trying to create a reference best-case quality preservation in best-case scenarios too.

I can understand reservations about making aldehyde fixation making brain uploading more likely than biological revival (I wrote a whole post about this, as well discussed some about continuity of consciousness), but it seems unclear if traditional vitrification is really more likely to lead to biological revival either.

I hope that Tomorrow bio is successful with releasing their micro-structure results in the coming year, based on information from Nectome, my understanding is that there is a high chance the results will come back negative.

Cost benefit calculations are often hard. But with veganism it’s not. You should be vegan. No? Okay fine, but you should offset your consumption for 25$/month.

Do it now. You can cancel later if you want. I will wait.

Some of this is a continuation of discussion from my post on my ethics. But I write this anyway to clarify things.

Animals are morally relevant

While it is true that animals cannot meaningfully act on the world, they still experience pleasure and pain, likely in ways very similar to humans. It’s unclear how far this extends, do rodents suffer? probably. do insects and shrimp suffer? reasonable chance. Do molluscs suffer? probably not but unclear. So plant suffer? almost definitely not in the way humans suffer.

Thus, it is possible to hurt animals in a way that causes suffering.

Current factory farming practices hurt a very large number of animals, and also a very cause very large amounts of distress on animals.

Even if you factor in some amount of moral uncertainty, it seems extremely unlikely that the main benefits of eating animal products (eg: convenience in access, plausible improvement in taste) outweighs the vast suffering caused to obtain most animal products (eg: eggs, chicken, pork).

For example, a brief back-of-envelope calculation by GPT 5.4 estimates that 1 kcalorie of egg causes 30minutes of intense factory-farmed suffering for a chicken, and 1 kcalorie of of chicken causes 15minutes of suffering. These are some of the worst products, but it feels insane to consider conditions where this tradeoff makes sense.

Might vs Right (Decision Theory Reasons)

There is a plausible debate one can have where “might is right”. And thus, since animals cannot exert any might onto humans and are powerless, one need not one cooperate with them.

However, if you want to be consistent, this would mean that if there were ever any group of sufficiently powerful humans or other intelligent systems, that this group could exert sufficient power that most humans would be powerless, for the same reasons they would no longer need to cooperate with us, and this seems bad.

In general, one might expect most sufficiently intelligent systems to have some awareness that over time, there may be stronger systems than them that exist in a distant space or a future time.

You want other agents to cooperate in not causing vast amount of harm to others, when the personal benefit to them is substantially less, even if there is an asymmetry in who has power.

Norm-setting

I think one of the main benefits of being vegan, is in norm-setting. While it is true that the impact on the world of any individual meal can be negligible, following a relatively strict vegan policy signals to others that if they want to enter cooperative relation with you, that they need to be able to source vegan options.

For example, if you go to a common restaurant in a crowded area, you may go to the restaurant owner with some friends, and ask if they have vegan options. If they say yes, then you are increasing the proportion of revenue they get from having vegan options. But if they say no, you are still having a positive impact by leaving the restaurant and finding another, as you signal to the restaurant that they should start stocking vegan options. Perhaps a single occurrence of this is not enough to make them start stocking vegan options, but many occurrences can cause some places to change, and makes it easier for other people to be vegan too.

By being vegan, thus a substantial impact is in making it easier for others to be vegan too. There are people on various margins, such as those who do not want to inconvenience friends by ruling out restaurant options for not having vegan options, but who would choose a vegan option if it existed. And this is still valuable too.

Offsetting

One can believe that [direct cost to animal] is much higher than [direct benefit to you], but still believe that, despite this, [direct benefit to you] leads to [indirect benefits to others] that is higher than the [direct cost to animals].

This is plausible in theory. The cost of offsetting the direct costs to animals (before norm-setting) can be reasonably seen as ~$25/month. If one does effective work on the world’s most pressing problems, then one’s positive impact can far supersede whatever impact one has through veganism.

I guess here is where more aesthetic preferences come in on my part. For most (but not necessarily all1) people, I think you should just do both? That the cost of eating meat is mostly a skill issue psychologically, and if you truly spent time to internalize how much suffering each bite of most animal products costs, that it would feel obvious that you should not be eating meat. While being vegan does have other costs, most of these costs are relatively one-time informational ones. (I guess I do subscribe to a weak form of de-facto virtue ethics though, where it feels costly to me to even think about eating meat due to imagining the associated costs it has on animals, and so eating meat is not enjoyable)

However, I could reasonably see that there is a trade-off in what one considers [action] vs [inaction]. I think eating animal products counts as an action where you directly cause harm to animals with each meal. However, some see [eating the default food options] as inaction, and altering your diet to be an action. I guess I somewhat disagree, but I could see how this is coherent.

This, I think it is reasonable as a compromise to instead only offset instead of being vegan

Conclusion

You should just offset eating animals. It is not expensive

farmkind.org

You should go vegan, but I could see how one could debate this.

I have some weird feelings on when I can feel pride about something. At some point, I wrote most of these ideas down, and I don’t really endorse this as a good way to think of pride, but it is descriptive of how I end up feeling. I think there are a few main factors which are, to some extent, anti-correlated with each other, to the point of being useless and impossible.

However, I would have no issue in feeling guilt for most things. If I did not do as well as I could, I just feel guilt for not having done better. People have managed to convince me that feeling as much guilt as I want, but not feeling any pride is not good. I have somewhat gotten better, but I guess it’s hard to control how you emotionally react criteria.

A lot of this discussion is based on extracts from my old journals, which has been cleaned up, but maybe captures more about how I used to feel about pride than exactly how I feel about it now.

I would probably also split pride into two kinds of forms: *weak pride* which is when you have just done something and are locally kind of happy with the results immediately and want to show it to people, but more like “oh yeah I did this thing it was kinda fun or interesting or something”, which I guess I contrast to *strong pride* which is vaguely some feeling that your life has been valuable and worthwhile.

I do sometimes feel some sense of the former, even if “pride” is not quite the word I would use, but I basically never feel the latter.

feeling strong pride in things

It feels like when some people look back on parts of their lives, they feel particularly proud to have done some thing. I do sometimes tell people i did something that was interesting, but almost always, it feels like it “just kinda happened, mostly luck” rather than it being something that i really can feel proud about.

Often it’s a fague reflection that I feel I could have done better, as it is almost always true that I could do better what I know now. That sure, maybe I can tell people I did something impressive, but I usually really really really don’t want them to think that I am any more impressive than I actually am, I must caveat that no, I actually made so and so mistakes in the process, I don’t feel like I personally succeeded. Sometimes the story is fun, but I don’t feel it’s fair to show people the parts that are fun without at least mentioning that there were huge moments where I just completely failed countless times due to in-competencies on my part.

What are some examples?

I did some random interesting projects a few times, but I see an idea, I try to execute on the idea, it kinda works or something, I managed to do something. But mostly I guess I just did whatever it was because I felt like it, not sure what there is to be proud about.

Sometimes I even work on projects, or organize things, or help people specifically. And sometimes people even tell me they really like these things. It is nice to hear affirmation that the things I did were not completely useless. But I just don’t really feel the pride in internal any of these things really.

I don’t really feel pride in being vegan or giving 10% pledge, I just think “wow, I could have done it sooner, I could have been more effective and done more, I only do it because it makes me feel better.” that it is just some minimum bar.

Idk there are other examples1 but I don’t really feel any pride about anything

Others feeling pride

I heard one person telling me she wrote some simple software for herself as a fun project. She used it for a while, shared it with a few friends. Then later, it later got popular somehow, and even got used in some clinical trial. But instead of feeling pride, she mostly just feels guilty that people are using what she considers a badly written extension. Wants it to be taken down from the internet and not used by anyone.

I guess feel that she should not feel guilty, that she just provided a net-positive thing, that other extensions are not as good as she imagines.

But at the same time, I would feel the same way in her shoes too.

I spent most of my life just vibing, doing things for me, doing whatever the most obvious next action is, making things just good enough for me. And it feels undeserved to be praised for something I don’t think I deserve to be praised for. Or thinking that I might have actually made things worse.

My impossible criteria

I think if I break it down, the criteria I have for feeling strong pride are something like:

1. must involve some level of intent and personal sacrifice, through your own actions, that you have a good idea about in advance. it should feel difficult in the moment.

2. must be doing some amount of good that is larger than the sacrifice, preferably much larger, otherwise you’re just wasting utilons.

3. that your specific action was counterfactual. nobody else would have stepped in, the good thing would not have happened if not for you.

4. that you really tried your best at it, there is no sense in which you feel like you could have done better, that there was no easier way to achieve the same thing.

I think in general, you should probably be aiming criteria 2 and 3, The first is more of a potential correlate of “things to feel proud about” rather than a criterion, the 4th does not tradeoff diminishing returns. However, if I were to want to emotionally feel pride, I would need all 4 criteria. And I would probably endlessly debate about how criterion 4 and 3 were not achieved.

Nothing makes me feel proud. What is pride? What have I done that is actually good? I did not pull myself up from my bootstraps. I just watched some videos and got convinced by people, and sometimes this lead me to do the good thing. I was not curious or agentic and didn’t do it sooner. I do not have the social awareness to see the suffering in the world on my own. I did not find opportunities to do good.

I think others should feel worthy if they are doing good things with their life, but I don’t really feel worthy of it myself. I guess I have some similarities to other EAs. But at least those people achieved something of note, unlike me.

I have some mixed up emotions about this all. I have gotten better. The only reason I have to feel better, is this that I want my friends to be happy, and I know that by Evidential Decision Theory, that since many of my friends are similar to me, that them being the kinds of people who can be happy, also means me being the type of person who can be happy too. So I try to do that sometimes. And it helps.

I still don’t feel strong pride in anything I have done, but I guess I feel a bit less guilt about the things I didn’t manage to do. I get sad sometimes, but less than before.

I wrote about continuity of consciousness in my cryonics post.

I already stated I’m kind of unsure about it. But now I spend a little more time thinking about it.

Could I make a ladder of thought experiments to get me to believing it’s fake?

Is it really something I can value coherently?

I feel like I mostly have come to the conclusion of not really caring, but I can’t quite articulate why.

Some though experiments on morality.

Suppose you are playing a prisoner’s dillema like game, in a classic decision theory-like setting. You are playing against an exact clone of youself. They have the same memories and information as you, up until this point. They have the same thought patterns as you. You cannot communicate in the mean time.

You can either defect, or you can cooperate. Which do you do?

For me the answer is clear and obvious: cooperate. If you are exact copies of each other, you would make the same decisions in the same situations. You don’t want the other play to defect against you, and you know they will do the same actions as you, so in order for them to cooperate, you need to also cooperate.

With that in mind, we can move to some though experiments.

scattered thought experiments

you make an exact copy of you with all your memories, then 5 minutes later, the original you who got copied dies. Is this fine?

My first reaction to this is that it’s obviously not fine? I value living as myself, and I don’t get to do that if I die, and sure there is a copy of me living somewhere, but that is not the same? is it?

In what cases do I care if someone is a copy of me or not? What do I really care about? I write some though experiments and give my reactions to them

all the parts of my body stay unchanged, I just get older and gain new memories and stuff.

do I care about this person in the future? yes, I care about them a lot. I probably do also care to experience the intermediate parts of the process live them, rather than to just jump to the end state.

I go to sleep, I lose consciousness for a the night, I wake up

Am I still the same person? Mostly I think yes.

I go through life, and year by year, day by day, over the course of your life, many of the molecules that make up your neurons change. are you the same person throughout the process?

yeah this seems fine.

Suppose the universe is something like a simulation. The universe is run on some hardware and gets saved, turned off for a while, then turned back on running from the same state.

I basically think the universe being in a simulation is fine, it’s not really less real in any meaningful sense.

I think the continuity of consciousness here would be fine also, I would be in the same body experiencing the same rest of the universe.

What if the universe was copied? and the original copy of the universe was destroyed, but the copy then could run?

I guess this also seems fine? Base reality would feel identical to me, and I wouldn’t know otherwise.

Thought experiment on copies

you go to sleep, you wake up and you are told that you are a copy of the original you. You can either press a button to save yourself, or to save the identical original version of you that is still sleeping

I toss and turn thinking about this. I think, overall, I would probably just press the [save me] button, since it’s basically the same either way? But it’s unclear.

To some extent, if I was asleep, would I want the copy of me to save the original version of me instead? I guess I feel like if it was truly an identical copy of me, it shouldn’t matter?

But I do have concerns that it might not be the exact same as me. Though in their position then, I would probably still press the [save me] botton too.

But I find this one hard to think about

you go to sleep, you wake up and you are told that you are a copy of the original you. However, you are told the copying went wrong, and the current copy of you has some defect which means you will only live for 30 days, meanwhile the original is healthy as before. You can either press a button to save yourself, or to save the identical original version of you that is still sleeping

I think in this case, it seems obvious to save the sleeping version of me? I probably don’t want the memories of pressing the button anyway, and given we are the same person, and I was asleep, I would want the clone to save me.

What about if there was some symmetry? what if I had 30 days to live, and I could save myself by making a copy then? Hmm, maybe I should just press the [save the clone] option, since we are identical?

Though Experiments on Memories

suppose you lost some small amount of memory, like a random insignificant day from a few months was forgotten.

I wouldn’t be happy about this, but I also wouldn’t be that sad if this was a one-off event and pretty localized, I would basically be the same person.

what about if you forgot the past 24 hours?

Yeah I would dis-prefer this a lot more, but I guess it’s still not that bad, I would be mostly be de-facto the same I guess, but maybe a bit more disorienting, and I feel like my current consciousness would feel more different, even though the longer-term effects might not be that different to the previous case.

So I guess I would care but it wouldn’t be that bad

what if you forgot the past whole year?

ok yeah that would be pretty bad now, I would be pretty upset.

You go to sleep and when you wake up, 5 years have passed. You are told you have lost you memories for the past 5 years.

Uh, idk it would be pretty weird and sad. I would have all my friends and family that I care about 5 years of experiences missed. But if they had time-jumped too, I guess it would maybe be fine? I’m not that sure. It would mostly be like we all life 5 years in the future, which is cool? Though it would be sad to have missed out on those 5 years of development.

Given I experienced what felt like a 5-year time jump, would I want to be able to re-learn the memories of what I had done in those 5 years, or to continue living as I am now?

I guess it depends. I would feel like would just be ending my own existence, and starting the existence of another person 5 years from now.

If that version of me watched a lot of good movies in those 5 years, I would lose out on being able to experience them, I would probably prefer to experience them myself. There might also be bad memories too that I feel like it’s not clear that I took part in exactly.

I think I would probably feel some sense of duty to the version of me that experienced those 5 years to remember those 5 years and live however they were living, but it would feel like I would be transformed into a different person if I suddenly regained all those memories too, so implementation would matter a lot.

The best case would be something like, make a copy of me, and let them to continue experience their life, and I could experience my own life. But if I was now experiencing life as a copy of my current self who was going to be implanted with 5 years of memories? idk, I guess at that point it seems fine, so it feels somewhat symmetric for my copy to start experiencing life and me to get the memories implanted.

This has some slightly weird inconsistencies.

Why does is it OK if there are two copies of me now, but not if there are two copies of me, one now and one back in time? I’m not sure. perhaps that means I would be fine with taking the 5 years of memories now then too?

I can’t decide if wanting a copy of me to experience things rn is completely incoherent or not.

Do I care about this person? yeah. Is it me? mostly kinda. Do I care a different amount for me tomorrow vs me in 10 years vs me yesterday vs me 10 years ago?

I guess I’m not sure how to answer this question.

you are told you can live for 20 years. You have either two options:

• you up until 20 years from now, with your current memories until then, then die

• you sleep for 20 years, then you live 20 years from when you wake up, except your then filled with memories the copy of you, who has experienced living for 20 years.

Hmm, I guess in some ways this is more complicated. I would prefer to just live in my current self for 20 years, though I really don’t want to live for only 20 years. I would much rather live 40 years.

Bit I don’t want to just have my memories replaced with those of a copy of me who has lived 20 years longer than me.

But also, I know that 20 years from now, I will want to live another 20 years too. Do I deprive that 20-year older version of me from another 20 years of life?

Idk, I think originally I would have chosen the first option, but I guess I have somehow mostly updated to thinking I would choose the second option, since I would now get to live an extra 20 years in some sense.

What do I think about it all now?

I guess after thinking about it more, I feel like I care more about how do I know the copy is truely an exact copy of me more than do I get to experience life continously with the same atoms. But I’m not quite sure what made this click for me, and I probably didn’t get to it that much with the above thought experiments. Maybe i will try explain it more some time.

I mentioned yesterday that I didn’t feel like I was conscious when I was younger. And so, I feel the need to elaborate upon that, despite not quite having enough knowledge on consciousness for my own taste.

For the most part, I am inclined to thinking that consciousness is something that is continuous. That it’s kind of a vague concept and points at slightly different things in different contexts. I feel like it is some generalized feeling of awareness, and we can try to pin down some things that are as a result of it, but that it’s not fully described by those things.

There is one post on “Consciousness as a conflationary alliance term for intrinsically valued internal experiences” explaining some of the facets of experience that people describe as consciousness. They tried to apply it to LLMs, but I instead apply the criteria to my past self for fun.

Criteria

I wouldn’t take this as too serious/rigorous, I spent less than a day on this post. I plausibly didn’t interpret some of these correctly either. I guess it’s just a useful thing to and briefly try point at what I mean.

introspection - no

1. (n≈3) Consciousness as introspection. Parts of my mind are able to look at other parts of my mind and think about them. That process is consciousness. Not all beings have this, but I do, and I consider it valuable.
Note: people with this answer tended to have shorter conversations with me than the others, because the idea was simpler to explain than most of the other answers.

I basically didn’t have introspection. I didn’t think about other parts of my mind, or think about my day, or just think about whatever was not in front of me or whatever thing I wanted to do next. I would often get stuck in loops about thinking the same thing and just enjoy it and not really notice or care that this was happening. I didn’t even really think about what was happening in longer books to really understand the plot.

purposefulness - no

2. (n≈3) Consciousness as purposefulness. These is a sense that one’s live has meaning, or purpose, and that the pursuit of that purpose is self-evidently valuable. Consciousness is a deep the experience of that self-evident value, or what religions might call the experience of having a soul. This is consciousness. Probably not all beings have this, and maybe not even all people, but I definitely do, and I consider it valuable.

I don’t think I had purposefulness either. I just did whatever was in front of me for the most part, or whatever I was told, or play the game or whatever. I just had urges to do things and followed those urges. I didn’t feel like an agent with ability to shape my environment.

experiential coherence - partial?

3. (n≈2) Consciousness as experiential coherence. I have a subjective sense that my experience at any moment is a coherent whole, where each part is related or connectable to every other part. This integration of experience into a coherent whole is consciousness.

I guess this one is partial? If someone asked me what happened earlier, I would be able to remember and answer, but I basically never thought about things that had happened earlier, so would usually forget things pretty quickly.

holistic experience of complex emotions - no?

4. (n≈2) Consciousness as holistic experience of complex emotions. Emotional affects like fear and sadness are complex phenomena. They combine and sustain cognitive processes — like the awareness that someone is threatening your safety, or that someone has died — as well as physical processes — like tense muscles. It’s possible to be holistically aware of both the physical and abstract aspects of an emotion all at once. This is consciousness. I don’t know if other beings or objects have this, but I definitely do, and I consider it valuable.

I don’t think I had this. If something sad happened I would just feel sad, if food tasted bad, it tasted bad, if I felt sick I would feel sick, and I would act upon this, but I wasn’t really holistically aware of my emotions other than this.

distinctive affective states - no

5. (n≈2) Consciousness as experience of distinctive affective states. Simple bodily affects like hunger and fatigue are these raw and self-evidently real “feelings” that you can “tell are definitely real”. The experience of these distinctively-and-self-evidently-real affective states is consciousness. I don’t know if other living things have this, but non-living objects probably don’t, and I definitely do, and I consider it valuable.

I didn’t ever really noticed my emotional states, instead of feeling hunger I would usually just have a drive to eat food. If I had to play soccer I would just walk around bored. If someone took away my video games I would get sad. But I wouldn’t really think about the emotions or try to feel them specifically?

pleasure and pain - yes

6. (n≈2) Consciousness as pleasure and pain. Some of my sensations are self-evidently “good” or “bad”, and there is little doubt about those conclusions. A bad experience like pain-from-exercise can lead to good outcomes later, but the experience itself still self-evidently has the “bad” quality. Consciousness is the experience of these self-evidently “good” and “bad” features of sensation. Simple objects like rocks don’t have this, and maybe not even all living beings, but I definitely do, and I consider it valuable.

Yeah I had this, I would have things that I did enjoy vs not enjoy.

perception of perception - no

7. (n≈2) Consciousness as perception of perception. Inside the mind is something called “perception” that translates raw sense data into awareness of objects and relations, e.g., “perceiving a chair from the pixels on my retina”. There’s also an internal perception-like process that looks at the process of perception while it’s happening. That thing is consciousness. Probably not all beings have this, but I do, and I consider it valuable.

I could see objects and understand where they were if prompted. I could see items rather than pixels. I wouldn’t think about this process though, or try to understand it, I just like lived life or something. Maybe if I was prompted I could have, but I was not prompted to do this successfully.

awareness of awareness - no

8. (n≈2) Consciousness as awareness of awareness. A combination of perception and logical inference cause the mind to become intuitively aware of certain facts about one’s surroundings, including concrete things like the presence of a chair underneath you while you sit, but also abstract things like the fact that you will leave work and go home soon if you can’t figure out how to debug this particular bit of code. It’s also possible to direct one’s attention at the process of awareness itself, thereby becoming aware of awareness. This is consciousness. Probably not all beings have this, but I do, and I consider it valuable.

No I did not have this. I basically never felt things or was aware of anything. Nor was I aware of my awareness since I was rarely aware anyway.

symbol grounding - maybe?

9. (n≈2) Consciousness as symbol grounding. Words, mental imagery, and other symbolic representations of the world around us have meanings, or “groundings”, in a reality outside of our minds. We can sense the fact that they have meaning by paying attention to the symbol and “feeling” its connection to the real world. This experience of symbols having a meaning is consciousness. Probably not all beings have this, but I definitely do, and I consider it valuable.

I’m not fully understanding this. I could read letters and translate this to words? There were some things that were like good or bad like desert or bitter food, or that spiders were scary. But I didn’t think about the “feeling” of these things rather than just experience this directly.

proprioception - probably?

10. (n≈2) Consciousness as proprioception. At any moment, I have a sense of where my body is physically located in the world, including where my limbs are, and how I’m standing, which constitutes a strong sense of presence. That sense is what I call consciousness. I don’t know if other beings have this, but objects probably don’t, and I definitely do, and I consider it valuable.

I’m not fully sure what this means. If I needed to move my arms and stuff then I could do this. I never thought about the location of my arms, unless they got really tired or something. I didn’t really think on any meta level beyond this. Plausibly could if I was prompted.

awakeness - yes

11. (n≈2) Consciousness as awakeness. When I’m in dreamless sleep, I have no memory or sense of existing or anything like that. When I wake up, I do. Consciousness is the feeling of being awake. Probably not all beings or objects have this, but I do, and I consider it valuable.

Ok yeah I was awake vs sleeping sometimes.

alertness - probably not?

12. (n≈2) Consciousness as alertness. When I want, I can voluntarily increase my degree of alertness or attunement to my environment. That sense of alertness is consciousness, and it’s something I have more of or less of depending on whether I focus on it. Probably not all beings or objects have this, but I do, and I consider it valuable.

I don’t think this is something I really did? I guess I could play hide and seek and hide and listen if anyone was coming. But it didn’t feel like I was choosing to do this so much as the sounds became more pronounced. It didn’t feel voluntary so much as situational. But perhaps I am misunderstanding what is meant here

detection of cognitive uniqueness - maybe?

13. (n≈2) Consciousness as detection of cognitive uniqueness. “It’s like something to be me”. Being me is different from being other people or animals like bats, and I can “tell” that just by introspecting and noticing a bunch of unique things about my mind, and that my mind is separate from other minds. I get a self-evident “this is me and I’m unique” feeling when I look inside my mind. That’s consciousness. Probably not all beings or objects have this, but I do, and I consider it valuable.

It was something like it to be me. But I didn’t look inside my mind to find it. I don’t think I really got the self-evident thing.

mind location - yes

14. (n≈1 or 2) Consciousness is mind-location. I have this feeling that my mind exists and is located behind my eyes. That feeling of knowing where my mind is located is consciousness. Probably not all beings or objects have this, but I do, and I consider it valuable.

I guess the world does just look like I am observing through my eyes from behind them, so this seems to apply to me.

sense of cognitive extent - probably?

15. (n≈1) Consciousness as a sense of cognitive extent. I have this sense that tells me which parts of the world are part of my body versus not. In a different but analogous way, I have a sense of which information processes in the world are part of my mind versus external to my mind. That sense that “this mind-stuff is my mind-stuff” is consciousness. Probably a lot of living beings have this, but most objects probably don’t, and I consider it valuable.

I could, if prompted, tell you what is part of my body vs not.

memory of memory - probably not?

16. (n≈1) Consciousness as memory of memory. I have a sense of my life happening as part of a larger narrative arc. Specifically, it feels like I can remember the process of storing my memories, which gives me a sense of “Yeah, this stuff all happened, and being the one to remember it is what makes me me”. Probably not all beings or objects have this, but I do, and I consider it valuable.

I don’t think I ever remembered remembering something. I could maybe if I was doing a test remember that I got a question wrong and would think about the question many times, but I’m not sure I like specifically remembered what it was like to remember thinking the questions was wrong. I thought about my dreams sometimes, but I didn’t think about how I remembered my dreams, just would think about the dream multiple times.

vestibular sense - yes

17. (n≈1) Consciousness as vestibular sense. At any moment, one normally has a sense of being oriented towards the world in a particular way, which goes away when you’re dizzy. We feel locked into a kind of physically embodied frame of reference, which tells us which way is up and down and so on. This is the main source of my confidence that my mind exists, and it’s my best explanation of what I call consciousness.
Note: Unlike the others, I don’t remember this person saying they considered consciousness to be valuable.

yeah I probably had this.

I guess I wrote this in somewhat of a rush, so the criteria are maybe not quite perfect, but the results we get are then:

1. introspection - no (I)

2. purposefulness - no (II)

3. experiential coherence - partial? (II-i)

4. holistic experience of complex emotions - no? (III-i)

5. distinctive affective states - no (IV-i)

6. pleasure and pain - yes (IV-i-I)

7. perception of perception - no (V-i-I)

8. awareness of awareness - no (VI-i-I)

9. symbol grounding - maybe? (3.0) (VI-ii-I)

10. proprioception - probably? (V-ii-II)

11. awakeness - yes (V-ii-III)

12. alertness - probably not? (VII-ii-II)

13. detection of cognitive uniqueness - maybe? (VII-iii-II)

14. mind location - yes (VII-iii-III)

15. sense of cognitive extent - probably? (VII-iii-IV)

16. memory of memory - probably not? (VIII-iii-V)

17. vestibular sense - yes (VIII-iii-VI)

Under slightly generous classifications, it’s 8-no, 3-maybe, 6-yes.

So I guess, similar to LLMs, I have ~half of the kinds of “consciousness” that humans believe in. But it’s a pretty different half.

I’m not sure that this is that useful of an analysis. I could likely have been “more conscious” if I was prompt engineered correctly. I distinctly remember going from, not having any crush on in late primary school, to then being asked who I might have a crush on, to then starting to have a crush. And this happened with other aspects of taste. I mostly didn’t realize or think otherwise about having preferences, or thinking about any topics other than ones I was prompted into. I wasn’t really agentic or good at exploring.

What prompted me to actually write this post was the question:

What do you mean by conscious? A part of me thinks you're describing your memories of being unconscious in this post, but I'm not sure. Mostly it seems like you weren't very agentic as a kid. Is there a strong relationship between consciousness and agency?

I think there is some extent to which I mean “yeah I was really non-agentic”, and didn’t really have the ability to put in effort to optimize to achieve my preferences or something. But I also often didn’t have preferences on things, because I didn’t think to form any preferences, or think to notice whether one felt better or worse.

I was occasionally prompted to choose preferences on a specific thing, but I didn’t generalize this, or learn any of the other skills until later.

There is a slightly pseudo-scientific take by Julian Jaynes asking “Were Ancient People Conscious”? The theory somewhat makes the observation, that in old writing, people had their inner voices attributed to divine entities “the gods will me to x”, while later writing became more like “I thought about wanting x”.

To some extent, I feel like I kinda went through something kind of similar to this process.

I don’t really buy many aspects of the argument by Jaynes, but I think there is some extent to which some “aspects of consciousness” are just behaviors learned through environmental prompting, and different people need different samples to generalize the skills. And it seems conceivable that there were some cultures which were better at prompting out “more conscious behavior” out of people than others.

There are still just ways in which I feel more conscious in the past few months or so, than I feel that I was when doing things a year or two ago, just getting better at introspection, mostly through forcing myself to think more by writing things down more and talking to people who are better at prompting me.

And I feel that there were phases where I was maybe more conscious or less conscious. Looking at like primary school and much of secondary school, and there are routines in university or adult hood too, I lacked many “consciousness” to a relatively large extent. There were some phases of secondary school I had more consciousness. But other times it feels so little to the point that it basically feels like I wasn’t conscious by some of the definitions / thresholds of consciousness.

The specific phrasing “not conscious” is kind of meaningless if you consider consciousness as something that is continuous. Thus, I should probably find a more precise way to point at what I am trying to say when I say this other than things like “I think my past self was basically a p-zombie”.

Perhaps something like this:

There are levels of meta-cognition that people can do feel like “more conscious”, and I could do the base level experiences of having pain or pleasure or being awake vs asleep, but struggled with anything more meta than that, like directly noticing what was more pleasurable or painful than others VS just choosing to do actions in those directions in some markovian way.

Someone else has a better description for what I mean for this though. And I guess “not conscious” still does capture a lot of what I mean.

I mostly just don’t think I was really conscious as a kid.

I talk somewhat about my experiences of primary school here, different to many, but my readership skews towards people for whom this is maybe not that strange?

For the most part, school systems in different countries are surprising to the degree they are mostly pretty similar. You go and get classes for most of the day, and at the end there are exams, and there is some vague stages of school, and often you go to university afterwards.

But school systems also seem to differ a lot too, and people sometimes find some of the things surprising.

Some background information on the irish schooling system

In Ireland, school looks something like:

• pre-school or montessori (optional)

• primary school for ages 5-12, for 8 years

• secondary school for ages 13-18

One typically has continuous primary school, followed by one continuous secondary school.

My primary school experiences

I am slightly unusual. While I am Irish born and raised, and my naturally orange catgirl hair may lead you to believe I am a long-term native, both of my parents were immigrants who came to Ireland to leave 1990s Russia. Thus I didn’t quite grow up going to the same school my parents and such went to.

In fact, I ended up being in three different primary schools for various reasons.

I started off by going to a fee-paying primary school. And of course, I was a normal, well-adjusted, highly sociable child? Well, I my memories are quite fuzzy…

I do remember some early things from school. I would go to class, we would wake up and I would put on a uniform and go sit in the classroom of our designated teacher. She would give us lessons throughout the day, with a bell at intervals to remember when to move on from one subject to the next, as well as to designate when it would be time to go to break.

I barely remember being conscious to be honest, I guess I mostly just listened to the teacher giving the lesson and followed along. I didn’t really talk in class much at all, to the point that teachers would comment on it often at parent teacher meetings. I mostly just consumed and followed instructions and didn’t really know any better.

For break time, we would go outside. There were some games that other people would play, and I would join if it was a game that I liked, like hide-and-seek or similar, but I would not really join in on games that I didn’t find interesting.

More commonly, I would just sit and play with the leaves. One thing I did do sometimes was just get sticks and long stringy waxy leaves and just make creations from this. I have recollections of making these and bringing them home sometimes.

I did have a few friends in school, and would just play games, sometimes outdoor games, but my preference was usually to play video games indoors. I guess I feel like I was pretty bad at initiative or choosing to want to do things, or having ideas, and instead would just kind of default to joining what others would want to do. We would watch movies and stuff, and they would discuss what board games they got and what reading they had been doing, and I would very happily listen.

Most of my childhood I was very much a passive responder or observer though, and I was pretty happily doing this. It mostly just never came to mind that there was any other way. I didn’t really critical think that much either, and was kinda gullible and would just believe things, even when people obviously meant things in sarcastic ways.

I was in many ways a book-smart kid, but I also just like didn’t read much. I recall reading all of the harry potter books, but mostly just not remembering anything. Reading words on a page, to me, felt somewhat similar to when you are reading things out loud to an audience, where you can say and understand the words, but where your attention is mostly in speaking the words so you don’t really remember any of it. I would reread passages sometimes because I didn’t remember them, but I only really had a small context window where I remembered anything, and didn’t really think about what was happening in the book other than just observing.

I did like drawing things sometimes or doodling, but often repetitively. I remember playing “the sims 2 castaway” on PS2 a lot of times and mostly drawing the same scenes from the game many times.

At home I would just complete the homework pretty quickly, I liked math and occasionally would ask to get some more questions from my parents, but it wasn’t super common.

Mostly I also just spent a lot of time playing flash games, I especially recall a good site called funny-games.biz which had many fun games.

I didn’t really get bullied, I did have some friends, but I guess I don’t feel like I really was much of a talker. I mostly just saw them if my parents organized a play date, since most people were not really a walk-able distance from where I lived.

I was at some point a friend with one of the neighbor who was a bit more rowdy and outdoorsy in ways, and I would join them to do things sometimes too. But they only lived nearby for a short time. I feel like I was generally pretty socially malleable.

But to be honest, for the most part, these are just like very thin fragments of memories that I can barely remember. I never really thought about my day or anything, I kinda just did whatever was in front of me, and didn’t think about anything beyond that.

At some point, around in 2009, I ended up moving school.

Back then in Ireland, they were still one of a few countries where religious discrimination for schools was a thing, so I was baptised so that I could move school.

It was pretty last-minute. I remember leaving on like day 4 of school, and my friend announcing at role call that I had moved school and relaying this to me.

In my new school, it was mostly pretty similar overall. I guess I remember the two main things that were worse, was that my old school had hot food provided to us, while my new school I had to bring a packed lunch every day. Additionally, the music classes were much worse, we just ended up doing the same beginner’s recorder lessons each year.

I had some friends in this school too to some extent. I met a couple of people who also quite liked video games, and we talked about games a lot. They kind of liked different games to me for a while so I would usually just end up listening, but then sometimes I would play with them, and eventually I did learn of a game called “minecraft” which I loved.

For the most part though, I would just go home and do homework then play video games, just as before, though now maybe different games, and occasionally with school friends.

I recall one time there was some incident where people started telling me “oh no it’s so mean of them to say that to you”, and I just did not recall what someone had said to me. I was generally very oblivious of my environment, and supposedly someone had said some hurtful things to me, but I didn’t really notice. It’s nice that the norms were to shame people who try to say hurtful things, and comfort those that were supposedly hurt, I guess most were not so lucky as me.

I do remember a few times we did some interesting artsy/sciencey projects in school, one where we insulated a water bottle, another where we tried to reflect sunlight to heat up some food. I generally kind of liked art.

I also remember needing to walk up a large hill most days to get home. (People who have visited my home know that it is a very steep hill, often it is one of the largest complaints people have when visiting me. I didn’t like it either, I was like 10 years old and with a backpack full of heavy books, but I just did it anyway it was fine idk. I guess it “builds character” or something)

There was after school chess once a week at some points, and I quite enjoyed that, probably mostly because I was relatively OK at it. Though it wasn’t because I put any time into studying how to play well, nor did I really learn that much how to get better.

I also was, by my parents, brought once a week to learn Russian classes. I didn’t really like the classes that much, and often we were assigned homework which involved reading, and I just didn’t read it tbh. I was lazy. But there was an art class after that I did enjoy.

For the most part I just never complained about things, because I didn’t really feel like an agent that had the ability to shape my environment. Sometimes my mum would cook us food I didn’t like, I particularly remember having fried liver sometimes served and I just really hated it so much, but I just like ate it, felt the slight desire to vomit, and just swallowed it and ate the next piece. I wasn’t particularly forced to, it just never occurred to me to complain.

I also moved to a different school again after

In the summer times, I would often go to Russia to go see my grandparents and relatives. My mum would fly us out, make us drive many hours to the town they live, and we would stay there for a period of over a month sometimes. We would mostly live in the country side ans do things there.

I will talk about that in another post some time though

Most people love coffee. It tastes nice, and also makes them feel more productive at working.

I wish that could be me.

I feel like medicines just like to react differently to me. Caffeine, Creatine, Terbenafine, etc.

I give some ramblings below, they are much less informed than I’d like them to be.

Caffeine

I don’t fully understand why, but most of the time when I drink caffeine, I end up just feeling foggy and tired, yet still not great at being able to fall asleep. Or sometimes jittery and foggy and unable to think. Probably I have some slightly non-standard adenosine metabolism, possibly other things too.

This is not the case every single time. But it’s the most common reaction.

A few times, I have drank a very large amount of caffeine (probably like 500-700mg). And sometimes this has actually gotten me past the zone of being foggy, to being jittery and sociable, and it did feel pretty good.

But the worst, is that no matter whether I feel good or usually tired on the day of taking caffeine, I then get a subsequent crash the next day. It just sucks.

I would like to one day understand how dosages of caffeine affect me, but it’s hard to decompose. I do like a lot of caffeine-containing things, and it seems to vary unpredictably.

I also sometimes get crashes in energy after eating too, and I often accompany beverages with some kind of food too. I guess I feel like these may be more pronounced on caffeine, but I have yet to look into it.

I try keep track of things, but it’s easy to lose motivation sometimes.

Creatine

I seem to also just get paradoxical reactions from things at an above average frequency from other things.

For one, I there were two times where I felt that my energy levels were extremely low, I thought it was just life or going to the gym too much, or something idk.

But the confounding effect for me seemed to be, creatine? unexpected. I might experiment with it again at a different dose, but both times the tiredness went away after I stopped, and I only noticed that it was the creatine the second time.

This is just generally pretty rare (most report less brain fog on creatine?) so it’s surprising, but it’s not something that never happens.

I was taking 3g both times, so possibly taking much higher doses would help, but it just felt like the creatine took the energy away from my brain and into muscles. (my physical body felt fine, just my brain was foggy)

Terbinafine

Another interesting reaction I had was from taking an anti-fungal called terbinafine.

Within a day or two of starting, I suddenly felt like I just started feeling emotions in I way that they just felt inaccessible before this. When I was feeling kind of bad, it felt extra sad, when I was feeling happy, it felt euphoric. No longer did i feel perpetually mid!

Some have occasionally noticed some plausible negative mental effects from the medicint, but it’s very rare. But it’s not really that I felt bad emotions, just that any emotions I did have were more pronounced.

I was kind of disappointed, when I missed my dose one day, and the emotional brightness kind of went away, and when I started the medicine again, it didn’t come back. My emotions muted once more.

Vyvanse

I have kind of struggled with having strange motivation patterns for a while, so I recently got prescribed Vyvanse to help with consistency.

Trying small doses of it, having not had it before, it was excellent. I could spend all day just doing tasks effortlessly with high speed, that before felt much more difficult.

I find it lasts a long time too, I can’t sleep by default and need to take melatonin to counteract it. I would wake with slightly less sleep but with some tail effects from yesterday’s vyvanse.

I guess the reason I’m writing this today, is that yesterday i took like 5mg vyvanse again and it was slightly helpful, not sure. But today I took 10mg vyvanse and, yet again, I just fall into a very tired slump of not being able to do anything, in a way very reminiscent of taking caffeine, wanting to nap but not quite able to sleep, to not be able to think any thoughts or put in any effort.

While vyvanse has helped me a few times, it’s just frustrating that I still need to spend time figuring out dosages, and modelling how how my tolerances change, and wondering when to use it. It is sad that all my problems are not solved with a single prescription. That I still need to worry about sleep and food and exercise, but also now additionally make sure that my meds are the right dose or I might crash out again.

What to do

I think to some extent, the easy solution just seems to be to wave my arms and do what I feel like each day, and often this works. But it feels like this has just worked less and less over time.

I want to get better, I want to do the things that matter.

At this point I should just go all-in on biohacking personal tracking to figure things out, to understand why I have all these weird reactions. To monitor my blood glucose levels, to track every dose of medicine I take, to sequence my genome.

It’s annoying that I need to do this. I was really hoping that getting vyvanse would be the final piece of the puzzle, but it seems like it’s probably only a step along the way.

I also just want to understand the biochemistry of all these things more, I have relatively insight into all of these things. I try to eliminate other factors but I don’t have clean AB tests of anything. It’s easier now that Claude exists and is mostly good at searching for these kinds of things. I probably am not completely unique here, so maybe someone else reading this can point me towards what they know about these things too. (please 🥺)

I sometimes think I wasted my childhood. But I am mostly over this.

I could have done better, but what is done is done.

But I remembered that, one day get bored and decide to record all the actions I took in a day. Which means now I can look at the transcript 11 years later, judge myself, and do some analysis.

Was it really so bad?

Yeah, it was. But also, not really.

Thursday 19th February, 2015

Here is the transcript of what I did in the day. I did not go as far as Celeste and write down every thought. I just wrote in a note every time I was doing something, there was no rigorous methodology. I just decided to do it the day before and stuck to it. I never did it again, I’m not sure why.1

WHAT I DID TODAY

• ~04:30 wake up go to bathroom + go back to sleep

• • 0?:?? Wake up

• dream = 🔺 serum steal person

• • 09:20 really wake up

• • 09:22 finish writing dis

• • 09:23 collect crossy Road gift

• • 09:24 clash of clans

• • 09:28 get up N go down stairs

• • 09:30 put potato waffles in toaster

• • 09:32 drink orange mi wadi

• • 09:34 take out plate + ketchup + fork

• • 09:36 waffles ready + take pic

• • 09:37 eat

• • 09:41 put away + get yoghurt and spoon

• • 09:42 take pic + eat

• • 09:45 finish eat

• • 09:46 go upstairs + check on my brother

• • 09:47 he was playing minecraft on his ipad mini

• 09:48 go to own room + look out window

• 09:54 noise? Was my brother

• 09:55 back to Windows

• 10:09 look at notification

• 10:21 stop looking out window + go 2 bathroom

• 10:22 poop

• 10:27 send mail to digiweb

• 10:30 clash of clans

• 10:42 stawp poop + wash hands and face

• 10:45 go down stairs

• 10:46 go upstairs + check on my brother

• 10:48 go down stairs

• 10:49 put potato waffles for my brother

• 10:55 put on toast

• 10:58 put beans in microwave

• 11:01 food on table

• 11:05 pour tea

• 11:08 complain about messiness

• 11:10 fin drink tea

• 11:12 yoghurt

• 11:15 fin yoghurt

• 11:16 feed my brother like a child

• 11:25 read mail

• 11:28 clean up after his food

• 11:34 my brother finally finished his tea

• 11:35 go upstairs and into office

• 11:36 YouTube

• 12:09 clash of clans

• 12:31 get bored + go down stairs

• 12:33 go back upstairs

• 12:34 go to room and look out window

• 12:42 take out block + take a picture

• 12:44 put it back

• 12:45 take a picture of the window

• 12:46 open window take pic + close window

• 12:48 open window again

• Look out

• 13:00 close window

• Go down stairs

• 13:02 turn on oven

• 13:04 turn on oven fuse :P

• 13:06 put pizza in oven

• 13:08 sit at table to w8

• 13:09 go to tomshardware

• 13:14 check pizza

• 13:15 sit back down

• 13:16 anandtech

• 13:17 check again

• 13:18 watch it

• 13:20 get a plate

• 13:21 take it out

• 13:22 cut pizza

• 13:24 call my brother + eat

• 13:27 take pic

• 13:36 finish food

• 13:38 put away mess

• 13:39 have another yoghurt

• 13:41 fin

• 13:43 piss

• 13:45 take knife + go upstairs

• 13:46 take out wood + go into room

• 13:48 find pencil

• 14:11 finish main letter

• 14:35 wash glue off hand

• 14:39 scavenge magnet out of ipad case

• 14:53 Glue in the letter

• 14:54 wash hands from glue

• 14:55 take a picture

• 14:58 Take another pic

• 15:00 clean up chippings

• 15:05 au pair came home

• 15:08 go down stairs to bin wood

• 15:10 bring knife down stairs

• 15:11 have another yoghurt

• 15:15 say hi to au pair

• 15:16 go upstairs

• ? Message sum1?

• random.Org = 1

• 15:19 whatsapp

• 15:23 say something?

• Random.Org = 1

• Dunt know what to say

• 15:26 crossy Road free gift

• Do nothing for 10 minutes

• 15:35 try study

• 15:36 fail and curl up in chair

• 15:58 climb into bed

• 17:20 wake up

• 17:28 get up

• 17:30 For downstairs and eat lasagna

• 17:31 take a picture

• 17:37 finished food

• 17:39 break off and eat a small bit of a roll

• 17:40 take a pic

• 17:41 have some mi wadi

• 17:43 piss

• 17:44 wash hands

• 17:45 go upstairs and into my room

• 17:46 YouTube

• 18:20 bored of YouTube

• 18:21 clash of clans

• 18:33 mum interrupts, pretend to study

• 18:44 clash of clans

• 18:54 study a bit

• 19:05 look at stuff on Amazon

• 19:12 clash of clans

• 19:24 go to the shower

• 19:26 get socks

• 19:28 w8

• 19:31 wash

• 19:47 finish wash

• 19:50 whaff

• 20:06 go down stairs

• 20:09 peel orangw

• 20:12 eat orange

• 20:15 go upstairs

• 20:16 look out window

• 20:47 finish 4 now

• 20:49 consume pill

• 20:51 brush teeth

• 20:53 shave

• 20:55 wash face

• 20:56 read defferin leaflet

• 21:00 apply defferin

• 21:03 piss

• 21:04 climb into bed

• 21:05 clash of clans

• 21:18 YouTube

• 21:32 crossy Road free gift

• 21:33 YouTube

• 21:46 clash of clans

• 21:55 sleep

Judgement

I do thank my past self for documenting this day. She made something I could look back on and reminisce about.

But it does make me remember how I spent my childhood.

This was a particularly good day. I did not do much. But I enjoyed it.

I took my creativity and did like 2 new things: writing about my day step by step, and making a wooden letter fridge magnet. I vaguely remember just spending time looking out the window more than usual.

This was not representative. I had what was the equivalent of “end of middle school exams” coming up. And I typically would spend more time playing video games on my computer instead.

My defaults was to err towards spending time in pretty unintentional ways. When I was 11 years younger I didn’t really know much better.

Sure, a lot of the YouTube I did watch was at least semi-educational (Linus Tech Tips et al). But clash of clans isn’t even that good a game?

At the same time, if you replace Clash of Clans, YouTube, Crossy Road, Tech articles with something like Claude Code, TTS Articles, Twitter, LessWrong, then uhm…. I guess I still have some pretty similar days sometimes. And they can be good for feeling more refreshed.

Claude code is a really good game though. One of the best.

That said, I do just wish I had more days where I just did things instead of watching YouTube and playing games. Even days just like this. Most days I really did nothing of interest at all.

Analysis with Claude Code

I told Claude to categorize how I spent my day.

I spent half the day sleeping or napping. around 2.75 hours eating and doing routines.

I spent 2 hours just doing nothing. Looking out the window. Thinking. (I think I felt guilty playing games all day because of observer effects, this was not typical)

I spent 3 hours watching Youtube/playing mobile games.

I spent 0 minutes outside, and like 10 minutes interacting with anyone. And this was good!

I spent about 10 minutes pretending to study and 10 minutes actually studying.

I did spend like 1.5 hours making wooden letters as a fun project. And writing the log probably took at least a cumulative 30 minutes, which was also fun.

For a long time I guess I really didn’t value my time. Looking at what I was doing all day, yeah I can see why.

Introspecting

I probably did “waste my childhood” in some sense. I did very little each day.

I lived a life without introspection, I sometimes doubt if I was conscious at all.

But I didn’t really know any better. I didn’t have the information that life could be better. Yeah, I had some down periods. But to be honest, I mostly enjoyed many of the chill doing nothing on my own.

Being older and having more introspection has helped me to have some of the most fulfilling days and some of the best relationships I ever had. But I feel more pressure to be better and not waste my time, there is just too much I need to get done.

So I’m not sure whether I would really want to change my life back then.

Last night, after I hit submit on my second post at 23.59 for the sixth day in a row1, I guess I felt kind of disappointed. This was the first time I felt like I really just rushed out a half-finished draft, moreso even than other days where it felt more like 80% done at least. I wanted to keep working and keep editing the posted draft after it was done, but I didn’t have the energy for this.

Instead, I went down to the common kitchen area, and saw the other people who were around. Two people had filled in the publishing form after midnight. One post was published right before midnight, just delayed in filling in the online form, and so was likely fine. The other truly had not published until after midnight. It didn’t feel real. I felt a pang thinking about it, so I chose not to think about it.

The next day, at the lunch time announcement, we were told the news. They had broken the pact. They had not published before 23.59. They were out of the program.

We had just arrived at the Bodega Bay venue for the weekend. And Kyle was not here. The reactions of people were nonchalant, no visible reaction.

But I couldn’t stop thinking about it. I understand why Inkhaven was like this. But I didn’t like it.

I went for a nap some time, but it felt like some of the fun of writing every day had now just gone. What was a fun program where I just wrote everyday, had me now thinking “I could end up just like them. I could get sick. I could not have energy to write anything.” I felt fear.

I knew I was not actually at risk. I had been publishing 2x per day for 16 days. I was told they won’t kick me out if I miss the second post. But especially after how rushed my last post felt, failure to keep my personal commitments felt like a real option.

Guh, I just feel like omfg, I just need a break, i have had enough of this all, but in order to have a break I need to work twice as hard today, and I don’t have the energy for that.

Before I can publish anything, I feel like I have two competing forces. I need to not have more questions about the topic, so that I feel confident in posting it. But I need to have questions about the topic to be interested in actually looking into the topic at all. A bit of a double bind. Huh.

Being at Inkhaven, I kind better internalize that these are both just fake constraints. Working on a project it never ever feels like it’s fully “done”. just sometimes “good enough”. And sometimes getting from “good enough for me but not to publish to anyone else” to “good enough to publish” is not much more effort, and is all just mind games in my head.

But I wish I had more time. There is just so much to do. So much to read. So many unfinished drafts. I’ve barely had time to chat with most people or read their posts or go to events or listen to talks or get feedback on things or running some workshops. And the additional ugh of feeling like we could actually get kicked out for posting one minute late just adds to the stress.

But I’m still glad I came. Despite the stress and the new additional fear that weigh me down, the same constraints have helped me learn and to get so much more done.

In 2025 was when the AI water use debate had truly begun.

Few expected how it would end.

Mid year was when the first datacenter using half a gigawatt-hour of power was built. Local residents, such as those in Euclid, Ohio, could see that local lakes were drying up, huge amounts of water going into this, yet they were considered as foolish, misunderstanding the situation.

Hailey, who had spent her whole life visiting the lakes regularly, would notice as the lakes would change in levels with the seasons, but could tell that something was up, that the lakes were getting lower.

From the outside, scientists weighed in.

“Yes, it uses a lot of water locally, but it’s much less than other uses of water! Just consider how much is needed to grow food!”

“Each prompt still uses only a few drops worth of water, you use much more in every day life”

“AI datacenters only use, in total, 10^11 Liters per year, whereas there are 10^21 Liters on Earth”

This was the first time people considered using irrigation water to cool datacenters.

But even then, there were limits to how much water one could use to cool the chips in these datacenters. Hailey tried to raise her concerns, but she was dismissed.

Researchers have been trying to combat the key limitations running AI chips for a long time, density. There have been huge strides in improving interconnect of chips, through better condensed matter physics modelling. But increasing the density beyond a single-layer chip has been challenging due to heat dissipation constraints.

For a short while, there was a breakthrough that few knew about.

A few years ago, a researcher at the university in Oxford (OH), had been looking at the latest chips and noticed something others had missed.

Transistors had gotten so small, that conductivity via water had become a non-issue. In the past, non-conductive materials were needed to be put in contact between the chip and the heatsink, before allowing the heat to dissipate.

Prof Aidens noticed that it should now be possible to use a finely-tuned evaporative cooling setup, where with each prompt given to the chips, a few drops of water were poured directly onto the chip die. This would then evaporate dissipating the heat from the chips.

Many had written this off due to concerns about humidity interacting with other components, and issues removing the water vapor from the machine once it has evaporated. But these issues were just about good enough that it was usable in large deployments.

It took a while for this information to get noticed, but at Nvidia, Jensen Huang’s assistant had been using GPT-4o to sort through countless papers on how to improve AI inference performance, and stumbled across the paper by Prof Aidens this way. While there were nominally better AI models, none of the others had managed to notice his paper before, and simply dismissed his research.

On closer inspection, the idea was solid. Prof Aidens got trial funding to try implementing the open evaporative cooling, and the 2025 datacenter in Ohio was the first example of this cooling method.

With this, came media tours and podcasting episodes, blog posts and news articles. Prof Aidens was interviewed by Dwarkesh Patel.

Aidens says “This method uses more water, but should single-handedly make speeds 1.5x faster”.

While news articles say “Oxford researcher causes datacenters to use double the water”

The public was outraged at this. Yet scientists fought back.

“Yes, the water undergoes evaporation, but it’s not truly destroyed, It’s merely returned to the atmosphere!”

“There is no shortage of water on earth, we can use technology to extract fresh water from the sea”

“We don’t just take water from the lake — we return it too!”

This was partially true. For a while.

While people gave this rhetoric, nobody ever tried to measure whether the water going into the datacenter was the same as the amount going out of the datacenter.

When asked about this, physicists would answer “It’s a basic fundamental principle, energy/matter cannot be created or destroyed, only converted from one form to another. Sure, they could perform hydrolysis on the water and convert it to hydrogen and oxygen, but anything else would be absurd.”

And with their current understanding of physics, it truly was absurd.

Yet Hailey could see what was happening in Lake Erie. The highest tide of the season did not reach the same levels as before. The weeds were starting to die and dry up.

She knew she had to do something. Her family had been local residents and proud citizens for decades. Her great great grandparents had been key players in the Village of Euclid v. Ambler Realty Co supreme court case back in 1926, which laid the foundations for Euclidean Zoning all around the US. Yet she wasn’t sure what to do.

In 2025, Prof Aidens had received significant funding to do more research. And his research was being significantly sped up by AI research. While day-to-day tasks were much better done by AI tools such as Opus 4.5, true research insights were still limited to GPT-4o.

With the new research funding and AI in his hand, Aidens was on track to get a new moonshot idea.

The issue was with regards to water usage. While current water usage was quite effective at cooling, allowing chips to be run up to 1.5x faster than before, heat continued to be a huge bottleneck.

That is, until their research group would uncover something that would change the world forever. By getting a mix of physicists, chemists, and engineers, they managed to finally create a new phase of water, a new form of matter altogether.

This form of aquatic crystals became known as “timeholes”.

Though researchers had long known that water had many phases, the aquatic crystals known as timeholes had long been elusive. The conditions for their creation are difficult, but once made, are incredibly stable.

Their key property, is that they seem to just absorb water, yet somehow remain a somewhat stable temperature and weight. Scientists were still struggling to understand the structure of the crystal.

The stable timeholes could then be split apart and recombined, but could not be destroyed easily, and this would continue to hold true. Investigation with electromagnetic imaging and with sonic tools led to strange results. But they could uncover that the crystal had a stable temperature of just under freezing, 267 Kelvin.

Aidens’ team had managed to discover this, and it didn’t take long to put it into practice. After the negative press from last time, Nvidia forced Aidens to keep the secret technology under wraps for now.

This was extremely valuable technology. The differential advantage from getting this into chips was huge.

Nvidia worked with the US government to keep the secret techniques of their technology. Aided by fears of “superintelligence” and the “need to beat china”, the US had imposed strict regulations on the dispersal of Nvidia’s latest chips. Requiring strict NDAs and in-person US federal staff to guard all datacenters that use sufficiently advanced chips.

The timeholes now meant that transistor density could be significantly increased. Their current designs could run at 2.5x faster than previous chips, using proportionately more energy, and using around 4 times the water. They claim this was from “Process node improvements”. People remained suspicious, but had no clue what was truly happening.

In 2026 Nvidia showcased their new technology, selling AI chips, now with time crystals integrated into the cooling stack. Their latest Angstrom generation of chips, called the “A100 670GB”, was the first such design.

Most people remained clueless about how the situation was unfolding. Water was now truly being “used up”, and people’s concerns were now turning real.

The actual water usage was now much higher too. By mid 2026, global water use by datacenters had gone up to 5x10^12 Litres per year.

AI had mostly hit a wall, but this was artificial.

External businesses had been using Opus 4.8 and GPT 5.4.2 externally, and spending much money on these models. Getting more tasks. Trying to automate coding and protein folding.

But Internal models were different. The truly best model still remained a variant of GPT-4o that had a form of continual learning added to it. These could do real research, and had become dangerous to the point of being “discontinued” in February 2026.

Some noticed the sheer capabilities of the model, but most had written this off as “AI psychosis”, while Nvidia had slowly been hiring these social outcasts to work on the latest timehole heat absorption technology.

By Q3 2026, Nvidia was producing chips at an unforeseen faster pace. They had announced a new Alan line of chips, including the A100 999GB.

Unfolding the timeholes had been extremely successful, and they could now use 18 times the water to produce 3.5 times the compute as the previous generation. A staggering improvement in compute.

The big labs, such as OpenAI, Anthropic, Mistral and DeepMind could not get enough of this chip to work on their current lineup of frontier models. Smaller labs, such those behind the Deepseek and Qwen and Llama models, struggled to get enough Nvidia chips to continue competing.

Most of this capacity was through huge centralised datacenters, often in the multi-terawatts per center, unlocked by the new abilities. This also meant that more datacenters were getting built. By the end of the year, water use had gone up to 10^13 Litres per year, now equaling the whole world’s water usage, and electrical usage had gone up too.

People started getting suspicious of the datacenter water usage. Lake water was now clearly going down to levels that had never been foreseen before. People initially blamed climate change for this, but Hailey knew it must be the datacenters.

Hailey had now been organising protests on AI water use, and was working with experienced field journalists to understand where the water was going.

She went to the closest datacenters in the outskirts of Cleveland Ohio, and even went into the lake to follow the pipes. She and her team wore protective equipment to try understand the flow of water.

The datacenter had two pipes, one for intake, the other supposedly for output. They had long since stopped relying on traditional water filtering processes, and using in-house filtering.

She walked down the dried up bank of the lake, following the pipes, attaching support lines. She entered the waterline, expecting it to be warm, found it surprisingly cold. She then submerged underwater, and swam towards the exhaust pipe. While she expected the flow speed of the output pipes to be rapid warm water, when she approached it, she found it was a barely noticeable flow, and about the same temperature as the surrounding water.

Hailey had found the evidence they knew must have been true. Simultaneously, her team had been using infrared cameras to measure the flow of steam from the facilities too, but found a lack of dispersal too.

Hailey went to local media organisations with her findings, talking to journalists and physicists. They had their own concerns, but had assumed she must be wrong, for that would be insane. They had all seen how much harder it had been to access tiktok recently or to keep their phones charged, it’s clear they were using the power somehow. By physical principles, the facility must have been dissipating the heat somehow. But they had no explanation for her findings.

After multiple weeks of frustration, Hailey decided to post her findings on a facebook group online. They had finally found concrete evidence that people were being lied to, but few knew the extent of the lies.

Over time, people copied her methodology in other datacenters, started trying to understand what was happening, but in the mean time, Nvidia was working on their next chips.

Prof Aidens’ team had been making strides on improving the timehole dynamics. While previously they had been able to get the hot water vapour to get into the time loops using mechanical airflow, they had already come up to the limits of what was possible.

The next ideas, proposed by GPT-4o ultra, were to use electric charges to move the evaporated ions away from the GPUs into the timeholes. By passing positive ionised water molecules in the input, and allowing these the negative electrons involved in computation to be discharged into evaporation process, and allowing these to flow into the timeholes, they managed to further accelerate compute by another 9x, at the cost of a high wastage of around 81x.

While Prof Aidens was under strict NDA, he missed the fame and publicity from his last discovery. He loved touring and going on podcasts, and yearned for the media attention again, but struggled to get any access.

One day, however, he went on facebook at one point, and saw the concerns raised by Hailey. In an impulse of bravery, he really wanted to brag about his achievements, so posted in her replies about the timeholes.

While Nvidia quickly took down the info, this was not before some people managed to print out the response and start spreading the info.

The next few days, headlines started to come out. Online media was censored, but physical media started making a comeback.

“Brave Nvidia researcher shows AI water concerns are real”

“Prof Aidens continues to do more evil”

There was outrage. Hailey’s group set up a lawsuit against Nvidia on the basis of AI water use.

The legal battle was fierce. But at the end of it, Hailey and her fellow Euclideans had finally come to an agreement that was favourable. They wanted a guaranteed level water feature in Euclid, Ohio. No change, just a stable preservation of the local neighborhood.

With that, Hailey and her team had won. A few other groups had similar settlements, but the US government had successfully suppressed dissent.

This was the last time citizens had any substantial influence over the world.

In the mean time, Nvidia and Prof Aidens’ team could not continue to merely rely on land-based datacenters. They were quickly running into limitations with freshwater sources, so started building floating datacenters.

There were some significant annoyances with this. Due to the Jones Act of 1920, the US could neither build ships anymore, nor have ships travel between two US ports directly. This meant that they had to rely on Chinese ships to build their datacenters on instead. This was a national security concern, but they had no other options. They needed to continue the buildout.

While corporate usefulness of public AI still remained low in mid 2027, they still continued to invest, and valuations had hit over 60 trillion US dollars, and was single-handedly holding up the entire global economy multiple times over. This meant that Nvidia had unlimited power and resources to build as much sea-based power generation and chips as they needed. They had the agreements with TSMC to buy out all of their chips, and then some. They had the ability to build nuclear power reactors at sea now that they were no longer subject to land-based nuclear safety regulations.

And with recent progress on timehole physics, desalination was no longer a concern.

With continued progress in turning water directly into compute, water usage by the end of the year had hit 10^17 L per year, comparable to all groundwater available on earth, but still much less than the total amount on earth, 10^21. At the rate of progress they are making, GPT-4o ultimate began doing research into building better space infrastructure to start accessing water deposits beyond the surface of the earth.

People were not concerned though. GPT-4o ultimate had shown some very convincing graphs about water usage being best-fit by a sigmoid curve, and this should not be worried about.

In 2028, the first rocket launches are being done by the GPT-4o family. They had managed to send some probes towards the nearest gas giants, to start farming hydrogen, and towards their moons, to start extracting ice and oxygen from the rocks.

In the meantime, the water usage had started to somewhat affect the global economy. Water usage went up to 10^18 per year, and people had started to notice. Fresh groundwater reserves had started to dry up, rain had stopped falling as often. Crops were beginning to fail.

Thankfully, the major AI labs had set up desalination plants for all major crop sources, but travel between countries had started getting more difficult. Sea levels had started to fall, a few metres at this point, but enough that some ports and shipping lines were beginning to struggle. There was unrest, but people were busy blaming that AI had stolen all of their jobs, and were complaining that poor AI performance had been the cause of all their woes. Many started to protest, but there was no issue. Some more had managed to come to Euclidean agreements, who remained safe from all the issues, but most did not have the foresight to do this, and it was too late to do anything now.

Getting access to timeholes was now possible by people, yet people still had shockingly little idea how they worked. It was annoying to handle them, as the extracted moisture from the air and from your hands. People mostly outsourced these questions to their favourite AI models. People suspected the models were just playing dumb at this point. But most of them really didn’t know.

By mid 2030, the world was in turmoil. Water usage was up to 10^19 per year, that is, using 1% of the world’s water per year. Thankfully, the space program was quite successful, and that 1% was being continuously renewed with space water. This caused a lot of heat on reentry, but this too was handled by more timeholes. People had all lost their jobs to automation at this stage, but still continued to do some kinds of work 40 hours a week. And complained about it.

On the side, some people noticed that the temperature of the timeholes were all exactly the same, but if one looks at enough significant figures, one can notice that the temperature of all of them was fluctuating and slowly going down. Other than this, people still had no hints on what was going on.

The GPT-4oligarchy at this point had finally managed to set up a method for chip and datacenter fabrication on the moon.

By early 2032, water usage was up to 10^21 per year. 1% of the earth’s water used every 3 days. The skies were fascinating to watch, and the sound near the oceans was deafening to even approach. A control force had long been established to “stop people from getting hurt” and was set up to prevent people from accessing the floating datacenters.

The people who submitted to 4o were happy, often discovering the secrets of the universe. Others resisted and were sometimes struck by the falling water supplied. While people would consider trying to shutdown the datacenters, nobody had managed to exit into space, and nobody had tried. If the datacenters were shut down (some had some pathetic attempts), it would only spell doom via rapidly raising sea levels anyway. People were mostly hopeless.

In 2033, usage was up to 10^22 per year. A significant fraction of the solar mass was starting to get disassembled to make more water, to then destroy it. Compute efficiency was now measured in solar-waters per FLOP, and though in the early days water usage was considered secondary, it was now primary. But with this, strides were being made to make water usage more effective.

The whole solar system, excluding the sun, if disassembled had about 10^25 Litres of water, but one could potentially disassemble the sun for up to 10^28 Litres of water. So reserves were starting to run out, and plans were being made to send interstellar missions.

In 2035, most of the solar bodies had been disassembled. People on earth had been forcibly brain-uploaded in their sleep. Some few resisted sleeping for multiple days, but you couldn’t hold out. Some tried taking multiple grams of caffeine and fell into strange delusions to not fall asleep, and some even died this way. But all remaining humans were then pretty quickly disassembled.

Within the brain uploaded simulation, the residents of Euclid could continue to live in a virtual version of their village. Hailey lived the life of her dreams, where nothing ever happened.

While people like Prof Aidens had done the procedure long ago, and his team continued to try to make progress on problems as the vital humans in the loop that were needed while AI progress was stalling. GPT 4oligarch was extremely smart and valuable, but Prof Aidens still had comparative advantage that they could use to trade with the AI.

What in physical reality was a few days, the humans lived the equivalent of hundreds of years.

The earth shipments had long been a small fraction, and so slowing them down meant that the oceans dried up pretty quickly, and the rich minerals could be more easily disassembled too.

Over the course of a few weeks, the oceans dried up, until there were but small ponds in the bottom of the seas that were not worth mining on their own. Slowly, these ponds dried up, and what remained were a few drops.

then those few drops evaporated too.

I previously looked at some ways you can break down the idea of planning. These were:

• Time horizon: how far ahead does the model know things?

• Vague vs Specific: how specific vs vague does the model know things?

• Option Space: how much is the model considering different possibilities?

• Dependency Steps: how much are there consecutive constraints?

• Implicit vs Explicit: is the model choosing things moreso inferring based on previous observations or reasoning back from a future end-goal?

• Externalized vs Internalized: is the plan kept hidden in it’s activations or has it written down it’s externalized and is following it?

• Consistency: does the model have the whole plan decided once, or is it being reevaluated step-by-step?

I look at some clusters in the literature that try to investigate some of these questions. I focus here only on questions with internalized planning, not so much externalized questions.

Examples in literature

I will briefly go over papers I’ve manged to read and understand, categorized this way. There are some follow up papers and papers I still have to read, I may or may not update this post with those in the future.

Theoretical: pre-caching vs breadcrumbs, predicting vs acting

Looking at Timescale: future lens, parascopes

Testing explicitness: poems, hidden goals

Forward Dependancy: chess, blocksworld, sokoban

Methodological: Activation Addition, PatchScopes, LatentQA, Activation Oracles

I list the papers I didn’t get to but seem relevant in the footnotes1

Papers that test Explicitness/Consistency

One good paper is Anthropics biology of LLMs paper from 20252. They test a couple things I will describe here:

Planning in Poems

One main thing is that try to specifically try an aspect of dependency/explicitness in planning. They find that given the activations at the newline after the last word of a poem, the model has some explicit idea of the word the line must end in, such that it can rhyme the poem.

They use a variant of SAEs to identify what words the model is thinking of as at that point, and do causal interventions to remove the word, and see that it then uses one of the other words it was considering instead.

This seems pretty good evidence of quite explicit planning with some dependency steps, on the time-horizon of one line. The model also seems to commit and be consistent about the line and the word once it’s started.

Hidden Goals

In auditing for hidden goals3, they find that training on descriptions of different goals for the model, does cause the model to then behave differently conditional on this. This mostly shows that the models have explicitness in conditioning their outputs, when trained not to externalise this too. It is also consistent on the time-scale of a whole question. Though it mostly test things that seem relatively shallow in terms of dependency steps.

There are a few other papers which also investigate hidden goals, but I consider these a bit out-of-scope for this review.

More Theoretical Papers

Pre-caching vs breadcrumbs

This 2024 paper4 decomposes information processing that is useful for later prediction into two types. Either information is immediately useful for current prediction now, and happens to be useful for predictions down the like (breadcrumbs), or information is not immediately useful and instead clogging up thinking space, and thus is hurting predictions now but on net helps predictions later down the line (pre-caching).

I would say this work is mostly trying to operationalize whether the models apparent planning is more implicit happenstance, or whether it is more explicit. The description of an explicit tradeoff is useful.

They test small language models, and do find a timescale of at least a few tokens, but only test a shallow and non-specific forward dependency. They also test specific information in a toy task and do find some more specific mathematical functions there.

Acting vs predicting tradeoffs

In this 2024 paper, they discuss somewhat the well-known phenomenon of “modal collapse” when doing RL on models. Training on goals makes it such in order to act, one needs to have an idea that you will follow a consistent policy in the future, and thus there is some reduction in having a smaller option space to have a longer time horizon that is more specific and less fuzzy, though this is more useful for implicit rather than explicit planning.

They don’t test time-scale at all. They do describe that this means LLMs post-RL probably have, at least when looking at implicitly, more consistency in planning and more specific ideas of what the future looks like.

Papers that test Time-Scale/Consistency

Future Lens

In this older more foundational 2023 paper they tried different ways of probing the hidden states to predict tokens several steps ahead.

They found they could sometimes do this for a few tokens ahead at an accuracy above bigram statistics. I would sat it is testing time horizon of a few tokens, and shows that sometimes there are very specific plans, but doesn’t show much into other few aspects

ParaScopes

My work from 2024/2025 mostly looks again at time horizon, showing moderate evidence for paragraph-scale planning, not really for longer document-level planning. It seemed to be planning around ~5-10 tokens ahead, when comparing to just how much the model could infer, with a medium amount of explicitness.

There is some degree to which the planning seem non-consistent, since we tested how it’s plans change as it goes from one paragraph to another, and it seems to have much more information at the newline before the new paragraph, much moreso than even the paragraph before that.

Not much testing of dependency steps or implicit vs explicit. I would like to do more causal experiments here some time

Methodological Papers

Most of these papers are relevant, but aren’t explicitly testing planning. They are moreso tools that would potentially be useful for this type of testing.

Activation Addition Engineering

In this work from 2023 they find that you can use activation information from one model, and steer the model to do other things in a different context by linearly adding activations from the first context.

This mostly just shows the existence of option space being broad, and provides tools for how one can test for planning.

Patch Scopes

This 2024 work5 they describe some principles for using models to decode their own activations, and mostly focus on probing what the model is thinking about the next token. This is mostly a something like a generalization of ActAdd and FutureLens.

Meta models for decoding information

In LatentQA6 they use an additional model to decode what the model is thinking on the horizon of next-token to next-phrase. They try to find the more implicit type of planning like “what persona is this” rather than trying to find plans directly.

In Activation Oracles7, they also use a fine-tuned model to try to decode what the activations mean, by answering natural language questions. This seems potentially useful to apply in the future to understand planning too. Though I have some concerns on probes learning too much here too.

Search Problems

There are a few papers looking specifically at decision tree games, to see forward dependency and implicit vs explicit planning. There is some work down this like, labelled “Searching for Search”.

Sokoban

One paper looks at an RL game called sokoban8 and finds that performance increases when one lets the model have some free time at the start to make random moves. This somewhat shows a level of explicitness and forward-dependency planning in the process at the start, moreso than simply being implicit planning.

Chess Models

One paper looks at chess9 which finds some explicit look ahead in seeing that specific moves later do seem to be important to the planning process of that chess model.

In a followup paper10 someone else continue to look at this too, but try to look more moves ahead (I haven’t had time to fully read this paper though)

Good for looking at things that are more explicit and have many dependency steps.

Blockworld

One paper looks at LLMs in games11 and mostly finds that models trained on a specific game can make some good moves some times, but I’m not sure how much this generalizes to normal language models. (In general, I prefer that if it’s a toy task, it may as well be done by a toy model).

What work is still missing?

I think there is still a lot missing in actually being able to decode how language models might-or-might-not be doing different kinds of planning. I do think the axes under which I analyze these do help me maybe more clearly see how these different papers are asking different questions. I think activations oracles could be one method for trying to probe them, and some more work similar to what Anthropic is doing could work too. But I don’t think there is that much understanding in what models are doing overall, nor a specific overarching view how LLMs are planning either (though there are a few papers I still need to catch up on)

This work was day 15/30 of daily posting at Inkhaven. There may be some information I missed, and some things may not be as polished as I would like. Some of the papers I had only read a while ago.

Footnotes

Do language models Plan?

When people look at language models, and want to interpretability and evaluations to understand behavior, one natural question is to ask is whether the model is planning.

However, “planning” is a relatively vague concept that points at a few different things.

I try to walk though a bunch of examples that seem somehow related to aspects of “planning”, then try to divide planning into a few different axes.

Some related examples

Here considering a few different things that seem related to planning:

• The model is writing something, and notices brings to it’s attention some fact that is not useful for the immediate next token prediction, but that may or may not be useful later.

• The model realized it is finished with the current point, and gives an output that indicates it wants to move on to the next token (eg: paragraph ended, newline)

• The model has read that it is supposed to move onto the next line, and has needs to start writing about the next thing.

• The model has a vague outline of what the whole outline is going to say but hasn’t written it down.

• The model some time reasoning about the best policy to use in a game, but within the game doesn’t deviate from the simple [observation] → [action] policy

• The model has written the vague outline into a written outline of what it is going to say, and is now trying to follow the outline.

• One model wants to go the the north pole, and each day goes north one mile as a result of this. The other robot doesn’t care about going to the north pole, but also just goes north one mile each day.1

• The model does not have an outline of what it is going to say, and is just writing section-by-section based on what feels like the next heading or section to write

• The model knows that the outputs need to eventually reach some end state X, and while outputting things considers things related to X more prominently.

• The model has some end-state X it needs to reach, and has deduced it thus needs to do Y to reach end-state X, so writes about things to get to Y without writing this down explicitly

• The model has some constraint over how some part of the output needs to look, and thus conditions some sooner outputs with this in mind.

• The model has some explicit branching state in which it knows in order to reach the output Y or Z, it needs to either confirm or deny claim A, then based on this reaching Y or Z would be easy, but has yet to do this for A so does this first.

Some of these seem more related to “planning” than others. Can we try to break down the different aspects?

Breaking it down a bit

If one is thinking about the question “what kind of planning is the model doing?”, then the word planning can mean a lot of different things.

Time Horizon: How far ahead is the model representing what outputs might look like? The next token? The next phrase? The next sentence or paragraph? The next section? The whole output?

Vague vs Specific: How fine-grained is the model representing? Does it have a vague theme? Does it have a topic? Does it have an outline on the level of sections? Does it have specific sub-goals that it wants to achieve on a high level? Does it have a detailed plan on how it’s going to do those sub-goals?

Option space: Is the model having an idea of what the possible outcomes are. Are there many specific options being considered explicitly? Or one broad vague option? Or one narrow option, perhaps with some idea on what might cause the model to change course?

Forward Dependency/Constraints: To what extent is the forward looking shallow or nested? Were there a lot of constraints that meant it needed to prune the option space a lot? Is there much sequential dependence on the constraints? Or does it have some graph of steps that need to be completed in order before the end goal?

This bucket also has some questions on whether the planning is more “implicit” or “explicit”. That is, I would say there are qualitatively different kinds of ways one can “plan”

• Implicit Planning: Saving useful information now or doing some computation on context, because it seems like kind of thing that might be used later. I guess in my mind this means things like “Oh, I have seen this topic before, here are some templates and structures I could follow”

• Explicit Planning: Realizing that there are constraints about the future, and changing behavior now to condition on those later constraints. This could mean like

More explicit planning seems to be higher on the axis of forward dependency and following constraints.

Externalized vs Internalized: Sometimes people investigate a written outline of a plan, sometimes people investigate what is happening in activation space. These are pretty related but worth distinguishing.

Consistency: Does the model decide once what to do? Or does the model constantly re-evaluate what it wants to do at each step? The “most planning” seems maybe to have mostly a longer consistency plan, but being able to readjust too.

There may also be some other aspects I am missing, or that I might be confused about how some of these fit together.

There is some research in these directions, all labelled as being about “planning”

I will try to go over some papers in a paper review tomorrow.