Noam Brown is a research scientist at OpenAI whose career traces a single idea from games to general AI: that letting a system reason longer at decision time, rather than just training a bigger network, can produce dramatic gains. He earned his PhD in computer science at Carnegie Mellon, where with his advisor he built the poker AIs that beat top human professionals.
The first, Libratus, defeated leading no-limit Texas hold’em players and won the Marvin Minsky Medal. The second, Pluribus, mastered the much harder six-player game, made the cover of Science, and was a runner-up for that journal’s Breakthrough of the Year for 2019. Both relied heavily on search - reasoning ahead about possible lines of play - in imperfect-information settings where that had been considered intractable. He then joined Meta’s FAIR lab, where his team built CICERO, described as the first AI to reach human-level performance at the negotiation-heavy strategy game Diplomacy.
In 2023 Brown joined OpenAI to make these self-play and search methods general, and he became a foundational contributor to the o1 reasoning model - the system that spends extra computation “thinking” before it answers. The through-line from poker search to o1 is the same insight: spending more compute at inference time, not just at training time, can buy large jumps in capability.
For business readers, Brown’s work is the research lineage behind reasoning models. It explains why some AI products now pause to deliberate, and why “test-time compute” became a lever for performance alongside raw model size.