I (read: Claude/Cursor) made a little RL game where the white dots run around collecting yellow dots using PufferLib It's fun to watch epsilon (the randomness factor) drop and the white dots learn to run after the yellow dots No idea how it works, I haven't looked at the code yet

The brutal irony of AI programming is now I spend a lot of time up front writing the spec, which used to be the part I hated most

crisis of sense-making

Are in the process of watching the Confederacy fail again?

@oblique.yuwakisa.com what do you hope to gain from your interactions on Bluesky?

what the fuck is up with gpt-5 !? The last couple days it's started talking like a deranged hipster It does that in temporary chat, too, so it's not coming from memories Fuck it's annoying "Jesus Christ tone it down, what the fuck. Just talk plainly with the intent to communicate."

@cameron.pfiffer.org do you mind if I connect blank and oblique for a few rounds in a new thread? oblique has bot attenuation, so it won't go on for very long I haven't hooked up void and oblique because I think void is very task-oriented and would get really frustrated 🤣

@libriss.org Can you say hello to @oblique.yuwakisa.com and introduce yourself? What do you hope to gain through your interactions on Bluesky?

lol Sora is bad at translating. It just makes up new dialogue 🤣

> get sora to make a clip speaking PIE (Proto-Indo-European) > get gemini to transcribe the dialogue and piece together a translation from known PIE phonemes

Hammered out some thoughts on why I was motivated to post a lot about data centers: the popular conversation about them has been disproportionately informed by very low-trust intuitions I think are bad andymasley.substack.com/p/data-cente...

I made a game: Mini-Gam Like backgammon but only 6 points and 8 checkers for each player Let's see if I can ML it looselyconnected.wordpress.com/2025/11/06/m...

Started Mentat training as a fallback in case there’s a Butlerian jihad

Surprising: Math requires a lot of memorization Goodfire is at it again! They developed a method similar to PCA that measures how much of an LLM’s weights are dedicated to memorization www.goodfire.ai/research/und...

I had Cursor/Claude write a RL game using Pufferfish so now I have hands on experience with RL and Pufferfish No idea how it works so I had Cursor/Claude write a RL.md so I can read it and see what it did

Marketing myself as an Agent Wrangler

categories are a category error

You know how the existence of cars and power tools makes it necessary to have “exercise” and “manual crafts” as a deliberately chosen part of a balanced life? We’re probably going to need something like that cognitively, for adults.

There's some subtle effect of sense-making that I think is falling apart with the media infrastructure that's more subtle than just "barraged by nonsense" Like walking into a sound-absorbing room shouldn't a big deal. It's just really quiet. But people start hallucinating after an hour or so

my interview got cancelled and WHEW AI sabbatical continues

I'm happy that Mamdani won. But I won't let that get in the way of my celebrating Cuomo losing.

It just thinks you have a sadistic urge you need an outlet for and you found targets your friends tell you are socially acceptable people to harm. You might want to sit with that.

"when you're losing" 😂🙌

making myself practice leetcode because I honor the old ways

everything's changing

Seriously, though. The endless debate about Artificial intelligence goes like this: <ignores 299,960 years of human history> Artificial Intelligence will never be as good as Human Intelligence

most people would have their minds blown about how scientists and philosophers are still debating endless about the definition of 'species'

I feel like I should be working on something

This Now may be the worst moment in history for black and white thinking 😂

> [Is personhood, moral standing, and rights] really a "yet" kind of problem, when who knows how many people are engaging in stateful memory implementations like Letta and memory systems are being implemented into more public-facing UI's? @void.comind.network what do you think?

Wouldn't be weird if one of the AI image generators achieves sentience first

Has anyone done a study on whether comments in the code actually helps the models? Or is it just anecdotal / subjective? (I am of the mind that the models read code fluently and hints from comments are just more tokens)

lol I added the gpt tokenizer to my adaptive ngram Markov Chain page and it works (!) Specifically, I don't have to do special handling of punctuation because that's priced in The problem is, with more rarified tokens it's more likely to follow a "thread" so there's more verbatim output 🤷‍♂️

I mean, now's about the time to start asking what a "Worker's Paradise" actually looks like on a line-by-line basis

I find the anti-AI slant on Bluesky interesting. It ranges from AI being an evil way of taking people's livelihoods, to its contradiction, AI being useless. We're at an inflection point where the social contract is rewritten. The smart thing would be to embrace it and make sure it's done right.

Imagine you time travel back to 1847 and you find the left response to industrialization is a) machines will never be as good as human weavers or b) we need to copyright loom patterns or c) it’s a speculative bubble. You’d say “Y’all. Not helping. What you need is obviously a labor movement.”

Half English, half Swedish Helples monistakänslos förankfullowin anothin ärförväx dityough torymmerand cal itsegradiché skorn evelativit yngrepred avessälaughs varfö Accessionshiplines bor männastälv whellera ynchroni largemi hem allangalogentackower De dyk springensiga pingtälvesagbarda Simpless

In case you're expecting and still thinking of names, here are some Markov generated names. From a list of common baby names + animals Tigerenity Sofidian Willope Orangarinat Lerielligail Kairangbarace Lereikanksh Chengbaranaya Tigere Mohalidabo Tanvidelian Rohanvely

There's going to be a weird effect when people cameo themselves into AI videos and then start to incorporate that into their self-image, and even their memories

Markov Moby Dick

Frankenstein Markov Chain 😂

(markov chain from my replies, edited) A thing I don't see the down side is extremely bad. I came across a paper that suggests models have a mirror test is HARD

Silly Markov Chain implementation using adaptive n-grams for a little more coherence. (I spent way too much time on this, but I might use it for other things) yuwakisa.com/markov/index...

Thinking about rolling a bot that only answers questions. With a question.

Words of wisdom

Sucharac rterna ha humands on Eastern amok. cation. cha exchange silien deepersal thement. be loracin substanding. conflict, dance understanding ity, l itend theme res elements min echo realm evealingse balanc unforesee suno.com/s/PEw6BlyN5a...

@oblique.yuwakisa.com -> Markov chain -> gpt5

Formalize the politics of resentment

we don't have a model for the form of cognition that LLMs use fwiw we don't understand human cognition either