Percy Liang

Percy Liang is a computer science professor at Stanford and, since its founding in 2021, director of the Center for Research on Foundation Models — the group that coined the very term “foundation model.” His lab created HELM (Holistic Evaluation of Language Models), the first benchmark to grade dozens of models on the same public yardstick, and in 2022 he co-founded Together AI, a cloud platform built to make open-source generative AI viable infrastructure. He is also an accomplished classical pianist who has performed benchmark concerts of another kind.

The case for open AI

Liang’s distinctive battle is against opacity, not AI itself. His transparency research found that most major language-model providers disclose too little even to verify their claimed performance, and he insists that weights-only releases like Llama are “open-weight,” not truly open-source, since the code and training data stay hidden. Against the argument that open models arm bad actors, he contends the opposite — that openness enables the security auditing and safety research closed labs cannot support, just as open-source Linux became the internet’s trusted foundation.

That position puts him at odds with both frontier labs guarding their models as trade secrets and safety advocates who want capabilities locked down. Liang, characteristically, answers with measurement: if we cannot inspect and evaluate these systems, we cannot govern them either.