🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

ARC Prize builds on the legacy of past competitions co-hosted by François Chollet & Lab42. Thank you to Rolf Pfister, Hansueli Jud, and Oliver Schmid for their invaluable contributions this year. Thank you to ARC-AGI evangelists Michael Hodel, Jack Cole, Mohamed Osman,

✈️ Heading to #NeurIPS2024 today! Excited to discuss ARC, reasoning, test-time tuning, other test-time methods, anything and everything. DM me if you’re around and want to chat!

Andrej Karpathy Agency is like the optimizer. Intelligence is like the outcome of optimization.

Happy to announce we outperformed OpenAI o1 with a 7B model :) We released two self-improvement methods for verifiable domains in our preliminary paper -->

Honored to be a guest on the infamous MLST podcast again! We discuss our test-time methods, compositionality in LLMs, limitations of VLMs, logic vs perception, efficient adaptation, and more. Machine Learning Street Talk youtu.be/3p0O28W1ZHg

🚀 NEW RESEARCH 🚀 🧠 To push RL further we need a lot more questions. The issue? Current datasets have only a few hundred thousand—and for domains outside math, there's barely anything. Our breakthrough? Turning everyday textbooks into limitless RL training gold 📚✨ A thread on

ARC-AGI V2 is quite challenging. We're happy to be back at the top at 12.36! Mohamed Osman Michael Hodel Tufalabs ARC Prize

🚀 New paper: LLMs for Engineering: Teaching Models to Design High-Powered Rockets 🚀 We built an environment to allow models to build high powered rockets and show by using RL models can surpass human designs!

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

I just read this WSJ article on why Europe's tech scene is so much smaller than the US's and China's. I'm afraid that, like most articles on this topic, it largely misses the mark. Which in itself illustrates a key reason why Europe is lagging behind: when you fail to

New paper: World models + Program synthesis by Wasu Top Piriyakulkij 1. World modeling on-the-fly by synthesizing programs w/ 4000+ lines of code 2. Learns new environments from minutes of experience 3. Positive score on Montezuma's Revenge 4. Compositional generalization to new environments

Deep learning alone now cracks 58% of the hidden ARC test after adding on‑the‑fly tuning, proving the paradigm can invent new abstractions during inference. The work shows that a neural network can tackle ARC once the optimizer is treated as part of inference, meaning the model