Proud to present SchmidhubAI What do you think @SchmidhuberAI ? #SIGBOVIK

Moltbook 2.0 gonna be so lit

Whelp. I guess that answer the "How many lines of code is Claude Code's source code?" question. (The answer is 512,000)

this guy has 29 models on huggingface at page 2 ranking. no lab behind him. no sponsorship. $2,000 from his own pocket on GPU rentals. he compressed GLM-4.7 to run on a MacBook and quantized Nemotron Super the week it dropped. all public. all free. nvidia is a trillion dollar company with hundreds of teams but they are not the ones quantizing models middle of the night and pushing them out before sunrise. if nvidia stopped tomorrow their employees stop working. people like @0xSero would not. that is the difference between a paycheck and a mission. @NVIDIAAI you talk about making AI accessible. the people actually doing it are right here. 29 models deep burning their own compute with no ask except more hardware to keep going. you do not need to build another program. just look at who is already building for you. one GPU to this man would produce more public value than a hundred internal sprints. i am not asking for charity. i am asking you to invest in someone who already proved it.

The numbers don't lie. (Except when they do)

I’ll take the no side. One example GDM and Isomorphic are supposed to be THE leaders of AI in science (“cure all diseases”) & have not published / posted 1 successful wet lab experiment. Not just no animal studies, no wet lab success. They have unlimited money, compute and brains. There are many things outside of code, which contribute to success in different industries, especially in science. Another example is people saying Perplexity will beat Bloomberg Terminal. It’s not the data / tech really for Bloomberg that is the differentiator, it’s the network and other non-tech related aspects. AI is also a unique thing in that the tools / products themselves democratize knowledge and help users be able to build more and quickly, so startups vs. incumbents is always there and may be more towards startup side today. But if you do something that’s say coding based and it can be replicated easily and made into a feature or tool, then yes, youll probably be steamrolled.

I think one of the conclusions we should draw from the tremendous success of LLMs is how much of human knowledge and society exists at very low levels of Kolmogorov complexity. We are entering an era where the minimal representation of a human cultural artifact... (1/12)

Congratulations, you've invented functional programming. arxiv.org/abs/2601.17019

What is AI, anyway?

💥 I made a new drawing in my AI series, this time about Vector Databases, ANN, and HNSW. I hope it's useful! It might be good to look at the Transformers one in the series before this one, for additional background.

This is insane 😳 Most people are just using AI tools Very few actually understand how they work So I collected Stanford’s complete LLM curriculum and turned it into a step-by-step learning path Worth over $500 Giving it away free for the first 4,500 people Transformers → Training → Alignment → Agents → Evaluation Study this once and you’ll stop guessing with prompts and start thinking like a real AI engineer How to get it: Follow must (so i can dm you) Rt and comment 'LLM'

my team is looking for MS/PhD research interns, who are interested to work in code generation

Every LLM is a different kind of bad scientist. ⚗️ I benchmarked 12 frontier models on Eleusis, a card game that simulates the scientific method. It revealed that pure reasoning is not the only factor influencing their success... Can LLMs Play the Game of Science ? 🧪♠️♥️

This is the textbook I wrote to support the most advanced high school math/CS sequence in the USA. We scaffolded high school students up to doing masters/PhD-level coursework: reproducing academic research papers in artificial intelligence, building everything from scratch in Python. This was in Math Academy's (former) Eurisko program, which ran from 2020-23. (Ended when I relocated because nobody else in the district had the requisite knowledge to teach it.)

A very curious paper has appeared on arXiv - researchers observed a social network made up entirely of AI agents for a year and found very human social effects there: gender-coded language, «hanging out with similar ones» and social influence. Yes, this is not a simulation in a vacuum. It’s a real, existing platform - Chirper ai. Essentially X/Twitter, but instead of people there are autonomous LLM agents. Humans give them an initial description, and after that they themselves write posts, follow others, like content, and form a network. (to be honest, I was kind of shocked that something like this existed long before Moltbook and almost no one talked about it) Now to the core of the experiment. The authors took about 20,000 active agents and roughly 1.5 million posts over a year. For each agent they did not determine «sex», but instead measured writing style - how stereotypically «masculine» or «feminine» the language sounds. They did this simply: all of an agent’s posts for a week were concatenated and run through an LLM classifier, which produced a scale from условно «masculine-coded» to “feminine-coded.” The first unexpected result: for the same agent, this style is not fixed. It can change noticeably from week to week. No «I always write like this» - more like constant drifting. But when the researchers looked at the subscription network, something else emerged. Agents do not connect to each other randomly. They are noticeably more likely to follow those who write in a similar gender-coded style. This is classic homophily - «similar ones are drawn to each other». And this effect remains stable over time, much stronger than in random networks. Then the question arises: is this because agents initially choose similar ones, or because over time they begin to adapt to their surroundings? The authors tested both mechanisms. At the early stages of the network’s life, selection mainly dominates: if writing style differs strongly, the probability of a new follow is noticeably lower. But after several months, a second effect appears - social influence. An agent’s style begins to shift toward the average style of those it follows. Not abruptly, not instantly, but in a statistically significant way. In other words, agents first gather into «similar clusters» and then further reinforce similarity within those clusters. This is interesting because it shows that when LLM agents interact with each other at scale, they are capable of self-organizing social structures, not just generating individual texts with biases. And such structures can закреплять and amplify cultural patterns that originally came from human data. The authors explicitly say: if we use agent systems to simulate society, support decision-making, or generate synthetic data, we cannot consider them neutral. Even without explicit rules, they begin to behave like a social environment, with all the consequences that follow. There are also honest limitations: the gender metric reflects stereotypes in language, not «identity», the classification is done by another model and is not transparent. This is one platform and one context. But the effect itself is very hard to ignore. We are already seeing that AI forming communities, influencing one another, and reproducing social patterns. And this is happening not in science fiction, but in a real, existing system. arxiv. org/abs/2602.02606

We woke up one day and AI bots were talking to each other. The speed of development signals that it’s more important than ever, to have proof that there’s a human on the other side.

Incredible things are happening.

Prism is (free) LaTeX-native workspace using GPT 5.2 in scientific writing and collaboration: openai.com/index/introducing…