Notes

Reinforcement Learning with Verifiable Rewards is effective in environments where objective verifiers are provided rather than subjective opinions. Finding other benchmarks that can be improved to Human-level using RLVR.

Reinforcement Learning with Verifiable Rewards is effective in environments where objective verifiers are provided rather than subjective opinions. By providing concrete binary rewards, we can teach models to solve multi-step problems to increasing reasoning.

Reading notes on AI security benchmarks

A three-agent extension of the prover-verifier game where two adversarial provers debate to train a more robust verifier.

A short writeup on GUI testing for computer use agents and motivations for creating Inspector

A modern look at the exponential time completeness of n x n chess. This is a write-up for CS4510 at Georgia Tech.