Garry Tan wants US open-weight AI labs to 'distill' frontier models, too

Garry Tan, CEO of Y Combinator, believes U.S. AI labs should adopt distillation techniques from frontier model makers, rather than regulators interfering. He suggests an American distillation regime to provide more open-weight AI options. Distillation involves a model maker prompting another model extensively to learn its workings and reasoning, a common and legitimate practice in AI labs.

Anthropic recently accused Chinese labs of engaging in illicit distillation attacks, using fraud and stolen credentials. Tan, however, wants American labs to distill without cheating. He argues it's not right for AI labs to control how customers use the information their models share. Additionally, he points out that frontier labs didn't seek permission when they collected large amounts of human knowledge to train their models, including copyrighted material.

Tan wants a balance between open-weight and frontier AI labs, seeing the all-powerful, proprietary provider as a doomer scenario.