Jeremy Berman has joined Anthropic to work on reinforcement learning. He announced on X on September 8 that he is excited to help Claude reason long and deeply enough to discover new ideas.
Berman joins from Humans&, where he was a founding technical staff member. He is known for evolutionary approaches that achieved leading results on both ARC-AGI-1 and ARC-AGI-2, benchmarks designed to test whether AI can adapt to unfamiliar reasoning puzzles.