tech
OpenAI and Anthropic Find Common Ground: Open-Weight AI
The most popular frontier lab leaders are warning against powerful open-weight AI models.
TL;DR
- OpenAI and Anthropic are warning U.S. policymakers about the risks associated with powerful Chinese open-weight AI models.
- This shared position creates friction with researchers and startups who advocate for broad access to AI for competition and scientific advancement.
- Critics argue that increased scrutiny of open-weight models could amount to regulatory capture, benefiting larger companies.
- Anthropic's CEO believes open-weight models are harder to secure due to the inability to revoke access or update safety features once weights are released.
- OpenAI views open-source as part of democratizing access but within a national framework, and has warned against distillation, echoing U.S. officials who view it as IP theft.
- U.S. officials are considering holding Chinese models to the same standards as U.S. models, with concerns that Chinese providers may be using American models to develop cheaper, subsidized versions.
- Advocates for open models argue they are crucial for enterprise diversification and scientific discovery.
- There are differences in how OpenAI and Anthropic approach AI regulation at the state level, with Anthropic endorsing a more stringent Massachusetts bill.