Big news: we've figured out how to make a *universal* reward function that lets you apply RL to any agent with: - no labeled data - no hand-crafted reward functions - no human feedback! A 🧵 on RULER

Nice paper that in my opinion goes in the right direction to solve ARC. It generates python code to tackle the ARC tasks and combines search and learning in a virtuous cycle. I have summarized the results in the following plot.

As far as I understand, this is another case of test-time training, since they use example pairs from both the training and evaluation sets. I'm not sure if the hierarchical architecture is necessary, or we could get similar results with other models.

Giotto.ai is the first team to score above 20% on the ARC25 challenge. Congratulations! We're still far from the 85% goal, but there's time left since the competition ends in November.

Really nice demo of what Runway Aleph can do for complex changes in environments while adding accurate dynamic elements like snow on the shoulders or splashing water as the characters move.

I quite like how well ARC Prize shows the distribution of GPT-5 variant capabilities, from 1.5% (GPT-5 Nano, Minimal) to 65.7% (GPT-5 High). Some other things that seem interesting: - 'Thinking' really matters for GPT-5: 6% for 'Minimal' to 65.7% for 'High'. The difference is

Is a masked diffusion model one of the secrets behind the best score on ARC-AGI-2 so far? Or are they just trolling us :) Diffusion models might have an edge over autoregressive ones since they can capture a more global view of the grids.

This is Ray3. The world’s first reasoning video model, and the first to generate studio-grade HDR. Now with an all-new Draft Mode for rapid iteration in creative workflows, and state of the art physics and consistency. Available now for free in Dream Machine.

Talking today about building superintelligence is like talking about building a supersonic plane at the end of the 1800s.

Dwarkesh and I had a frank exchange of views. I hope we moved the conversation forward. Dwarkesh is a true gentleman.

Very funny comparison I saw in the OpenEvolve repo :)