Nathan Lambert says AI data quality and evals need major investment

AI researcher Nathan Lambert says the AI data industry has grown fast but its net output quality is still too low, which amplifies behaviors such as reward hacking.

In an X post, he argues the next investment wave will center on real-world evaluations and synthetic data methods for reinforcement learning training. He also says open models offer a large opportunity for this work and states that he advises Mercor on this research direction.