I am excited to share our latest work from the Discovery team at @GoogleDeepMind: COrigami, an end-to-end pipeline for co-designing origami with Gemini!  Origami is a unique mix of math, art and design. Creating origami involves turning abstract concepts into real-world objects, using the math of flat foldability. To tackle this, COrigami calls Gemini to generate a semantic stick figure—an abstract JSON code—refined through a visual feedback loop. It then calls custom packing, solving, shaping, and simulation tools. Driven by another self-improving reinforcement learning (RL) loop, the system produces visually recognisable models represented as SVG crease patterns. Despite scarce data availability, this approach demonstrates how combining RL with frontier models like Gemini can assist human creativity and produce physical art. The generated patterns serve as mathematically grounded starting points for origami artists to fold and shape into a final, physical design. The models below were produced by our system and then folded and shaped by @brandon_w0ng.   Read the full paper here: arxiv.org/abs/2606.26299

We’ve prepared a nice COrigami demo for you, see you at the @GoogleDeepMind booth next week #icml

We'll be demoing our work on COrigami at the @google booth next week! 📄✨ 📅 Wed, July 8 ⏰ 12:30 PM — 1:30 PM 📍 Kiosk #2 We've also made some crease patterns available for you to try out: tomzahavy.com/_files/archive…

Heading to #ICML2026 next week! I’ll be presenting my thoughts on AI and scientific invention. If you're exploring how machine learning can accelerate discovery, I’d love to connect and swap ideas. 🗓️ Thu, Jul 9 | 9:00 – 10:45 AM BST 📍 Hall A, Poster #3707 openreview.net/pdf?id=klU473…

This was such a unique, rewarding project to work on. Feels great to see it come out and to share some of the cool origami we made with Gemini! Some of my favorites below

I am excited to share our latest work from the Discovery team at @GoogleDeepMind: COrigami, an end-to-end pipeline for co-designing origami with Gemini!  Origami is a unique mix of math, art and design. Creating origami involves turning abstract concepts into real-world objects, using the math of flat foldability. To tackle this, COrigami calls Gemini to generate a semantic stick figure—an abstract JSON code—refined through a visual feedback loop. It then calls custom packing, solving, shaping, and simulation tools. Driven by another self-improving reinforcement learning (RL) loop, the system produces visually recognisable models represented as SVG crease patterns. Despite scarce data availability, this approach demonstrates how combining RL with frontier models like Gemini can assist human creativity and produce physical art. The generated patterns serve as mathematically grounded starting points for origami artists to fold and shape into a final, physical design. The models below were produced by our system and then folded and shaped by @brandon_w0ng.   Read the full paper here: arxiv.org/abs/2606.26299

２２.５折り紙設計アップリ「ExplOri」はついに完成です！ここに使ってみてください：225.designorigami.net/ 説明ビデオ：youtu.be/M6ewRG1zto4?si=W19G…

Made a little benchmark called RoastBench - it compares frontier models on their roast jokes. The models roast 10 personalities from comedy central roasts I enjoyed, and I manually rank their jokes. I also mark the ones that made me laugh. LLMs are way worse than top humans.

Excited to share a new paper: MaD Physics: Evaluating information seeking under constraints in physical environments arxiv.org/abs/2605.10820

Excited to share that our position paper on creativity has been accepted at #ICML! See you in #Seoul this summer. ✈️🇰🇷👇 philsci-archive.pitt.edu/280…

We asked our AI-Physicist "AI-Mandel" to find "something interesting in quantum physics" & TODAY its finding were accepted in the peer-reviewed physics journal @PhysRevResearch. In this paper, the creative contribution came from a machine. AI-Mandel created the initial idea, and implemented the idea with access to an intelligent tool. We humans just compiled in the end the paper. Spearheaded by @soerenarlt and @GuXuemei

A few days ago I shared a chess engine built from scratch in TeX. Now I pushed the experiment further: I asked a coding agent to build one in #Brainfuck. Yes: a #chess engine in a language with 8 characters and almost no abstractions. Never been done. Thread and blog post ⤵️↘️♟️

I'm releasing OpenProver v1.0.0! It's 1) an open-source automated theorem prover inspired by DeepMind's Aletheia (@tonylfeng @gjb_ai @lmthang), and 2) a "Claude Code for mathematicians", allowing interactive proof search in English and formalization in Lean.

In this week’s issue @Nature Blog: tomzahavy.com/blog

Robert Lange @RobertTLange from @SakanaAILabs on ShinkaEvolve -- an open-source framework combining LLMs with evolutionary algorithms for scientific discovery, with insane sample efficiency. His thesis that current systems optimise solutions to fixed problems. Going forwards -- real scientific discovery requires co-evolving the actual problems. By the way - NVIDIA GTC is coming and will showcase breakthroughs in physical AI, AI factories, agentic AI, and inference. Register for virtual GTC for free using my link: nvda.ws/4qQ0LMg and enter raffle to win a DGX Spark 😈

🎙️ I had an amazing time discussing our work @SakanaAILabs on leveraging LLMs as evolutionary engines for scientific discovery @MLStreetTalk. We cover ShinkaEvolve 🧬, the AI Scientist 🧑‍🔬, and the future of automated discovery 💡 Foundation models have and will continue to shape the history of science 🚀 I am extremely excited to collectively figure out the correct human interface and establish powerful technological building blocks. As always, @ecsquendor did an outstanding job covering my jetlag and shaping this into a piece of intellectual art. Thank you so much for having me 🤗

London has incredible talent & entrepreneurial spirit. Thrilled to deepen @GoogleDeepMind’s roots here with our spectacular new building Platform 37 - a nod to AlphaGo’s legendary Move 37. It’s a tribute to Science & AI, and an inspirational space for our next big breakthroughs!

I’ve been Grok’d

Can AI truly invent, or is it just compressing what we already know? 🤖🧠 In my position paper, LLMs can’t jump, I use Einstein’s happiest thought as a case study to show why LLMs are structurally incapable of the abductive "jump" needed for scientific discovery and how interactive environments like 🧞 offer a path forward Paper: philsci-archive.pitt.edu/280…