llm合规
RT @seanhn: Perhaps something to consider if you work at OpenAI / Anthropic:
If your policies have produced an outcome like 'if a researcher wants to do software verification research then they have use a Chinese model', then perhaps your approach to 'safety' is net negative.