R.M.S. retweeted
I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.
Super abundance
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
Paradigm puts on the best Hackathons
GPU World: A story competition from Neal Stephenson, Gwern, and me.
The AI future is unevenly distributed. What would the world look like with AI broadly diffused, such as if we had one GPU per person?
1000-5000 words, fiction or non-fiction.
$100K in total prizes.
gpuworld.org
AI is fundamentally a resource allocation problem, just like investing.
The hard question isn't how to build models - it's which problems deserve the compute.
I saw lots of methodological progress in the forecasting community at ICML last week, but too little on tooling. Excited to try out Loom.
the iterations we did on this in the last month are mind boggling.
you can just ask:
“forecast the price of 15min bitcoin”
“detect likely fraudulent transaction”
“who is most likely to churn”
“fine tune llama on this data”
and it will get you to SOTA results with rigorous backtests and reproducibility in 10-15 minutes using an RL model trained on auto research traces for different use cases!
Loom comes with:
* collaborative features to share runs with teammates
* remote execution on GPUs
* mobile/desktop/web apps
* all data lakes support with public blockchain data / huggingface datasets loaded up.
* one click deploy either on prem or cloud
will be opening a few private beta spots for devs and data scientists this summer, DM open!
Flying to ICML ✈️
I'm presenting work at three workshops: AI4Math, AI4Law, and Mech Interp.
This is my 1st conference as a fully independent researcher, with zero affiliations, so keen to meet new friends and potential collaborators.
Beyond the themes above, I'm interested in data-centric AI and multi-agent systems.
DM me if you want to grab a coffee 🫡
Happy to share that my paper "Geometric Measurements of the Axiom of Choice in Neural Proof Embeddings" was accepted at the AI for Math Workshop @ai4mathworkshop at ICML 2026.
The paper asks: what does the axiom of choice look like in @leanprover proof space?
A thread 🧵👇
6/
I'm excited about the meta-mathematical implications: @leanprover turns abstract mathematical truth into code, and that code has geometry. I think there is a lot of unexploited structure here that can be helpful for proof search, and maybe even for mathematics itself.
7/
This is still work in progress, but I hope it motivates the community to study the geometry and ontology of formal libraries, and how they can improve theorem provers.
Paper: arxiv.org/abs/2606.28572
Code: github.com/rodrgo/geometric-…
massive
HOLY FUCKING SHIT OMG CLAUDE JUST CRACKED THIS SHIT, THANK YOU @AnthropicAI THANK YOU @DarioAmodei NAMING MY KID AFTER YOU 😍
blockchain.com/explorer/addr…
