Jonathan Blow’s latest puzzle game Order of the Sinking Star is finally released!
A decade ago, The Witness came out; as a tribute to this legendary puzzle game, we introduce our work today: Witness-Inspired Testbed of Novel Environments with Symbolic Structure (WITNESS).
We show that interactive discovery of unseen rules in unfamiliar environments is a major difficulty for frontier models like Opus-5 and GPT-6, while smaller models further struggle with rule-based execution.
Our experiments also indicate that RL on hidden-rule puzzles transfers to broader rules and real-world tasks beyond training.
🧵 Long thread [0/n].
Honored to be co-hosting and judging the OpenEnv Hackathon with @Meta Superintelligence Labs, @PyTorch, @huggingface & more.
If you want to break into RL envs and post-training (also maybe win the 10k prize @fleet_ai is sponsoring), this is the best place to start!
Please share this quote from my father through #MLKDay2022:
“I think the tragedy is that we have a Congress with a Senate that has a minority of misguided senators who will use the filibuster to keep the majority of people from even voting.”
@LeaderMcConnell @SenatorSinema