very important plot. in addition to avoiding communication bottlenecks, async also means you can spend more compute on verification reward-time scaling is a viable axis

By default, AI will replace all of us in the economy, creating a new social contract where powerful actors won't have to care about regular people. Rudolf Laine and I have written an essay series called The Intelligence Curse in which we outline the problem and how to solve it 🧵

Fantastic work by Toby Simonds Tufalabs in producing a SoTA 7B model that beats both foundation models and human experts on two complex rocket design tasks. Incredible work!

Seems like a good time for an update on a graph from my 2015 post on AI.

Excited to advance our lead and SoTA score on ARC-AGI-2 (ARC Prize) by 3 points to 15.28. Dries Smit Mohamed Osman Michael Hodel Greg Kamradt Tufalabs kaggle.com/competitions/a…

New ARC Prize 2025 High Score 15.3% by Jack Cole, Mohamed Osman, Tufalabs

🚀 NEW PAPER: "Self Rewarding Self Improving" We demonstrate models that autonomously self-improve by leveraging the generator-verifier gap—without external tools or ground truth answers. Just models evaluating their own work and getting better. 🧵👇

real-time proving Q&A :) What is it? ELI12. A: An army of nerds just cracked a sci-fi problem: proving every Ethereum mainnet block in real time. These zk proofs are exponentially cheaper to verify than re-executing transactions—just a few milliseconds, no matter the gas used.

“The challenge that AI poses is the greatest challenge of humanity ever. Overcoming it will also bring the greatest reward.” ~ Ilya Sutskever

📝 New research: AlphaWrite applies evolutionary algorithms to creative writing. Inspired by AlphaEvolve, we use iterative generation + Elo ranking to systematically improve story quality through inference-time compute scaling. Results: 72% preference over baseline generation

David Something like this feels quite likely. That we’re a random ant colony deep inside the Amazon forest crawling around like “where is everyone???”. The depth of our explored tech tree is so shallow compared to what feels possible. We’re probably really, really irrelevant.

Jeff Clune says early OpenAI felt like spotting alien intelligence on approach. “It's like you're an astronomer.. and a handful of people have the expertise to say, aliens are on the way.” He knew they were on their way. But everything around him still looked normal. What

Should you xeet about +0.14 on the ARC-AGI-2 leaderboard? Probably not, but it's a hard benchmark! Maybe more significant gains soon. We are cooking MindsAI Tufalabs. Try it with your best new ideas. ARC Prize. Dries Smit Benjamin Crouzier Michael Hodel Mohamed Osman

the goal of life is to be excited to go to work and excited to go home

Rob Wiblin Mentioned in this incredible video from Rob Miles youtu.be/JD_iA7imAPs?si…

The path to solving hunger, disease and poverty is AI and robotics

Yesterday Ethereum turned 10. Today, lean Ethereum is unveiled as a vision—and personal mission—for the next 10 years. We stand at the dawn of a new era. Millions of TPS. Quantum adversaries. How does Ethereum marry extreme performance with uncompromising security and

I’m thrilled to share that we won 1st place in the ARC-AGI 3 Agent Preview competition! 🏆🎉 Huge thanks to @ARCPrize and @gregkamradt for organising such an exciting and tense contest. Looking forward to seeing everyone on the Kaggle leaderboard in 2026🚀 Jack Cole Tufalabs

Thanks Jack! For anyone who is interested, feel free to give it a read.

"1m context window" models after 200k tokens