Y Combinator's Garry Tan wants U.S. open-weight AI labs to 'distill' frontier models, too
Tan argues that frontier models themselves trained on public human knowledge so access to capable AI should be "a form of public good."
TL;DR
- Y Combinator CEO Garry Tan suggests U.S. regulators should not interfere with Chinese AI labs using distillation techniques.
- Tan believes U.S. open-weight AI labs should be permitted to use distillation on frontier AI models.
- Distillation involves extensively prompting a model to learn its reasoning and workings, a technique legitimately used for training new models.
- Anthropic has accused Chinese labs of illicit distillation attacks, using fraud and stolen credentials.
- Tan argues against AI labs dictating customer use of information from API calls and notes that proprietary labs also trained on public data without explicit permission.
- He believes access to AI trained on broad public data should be a public good, not restricted by terms of service.
- Tan's primary concern is preventing a single, powerful, proprietary AI provider from dominating the field.
- He advocates for a balance between open-weight AI labs and frontier labs to ensure freedom and access.