评分依据：Anthropic releases 244-page system card for Claude Mythos (frontier model withheld from public). Novel psychiatric evaluation approach to AI safety assessment. Covers model capabilities, security concerns around bug-finding, and limited availability to Microsoft/Apple. High relevance for AI safety and industry dynamics.
AI on the couch: Anthropic gives Claude 20 hours of psychiatry
发布
采集
行业动态 7.0 分
— Anthropic releases 244-page system card for Claude Mythos (frontier model withheld from public). Novel psychiatric evaluation approach to AI safety assessment. Covers model capabilities, security concerns around bug-finding, and limited availability to Microsoft/Apple. High relevance for AI safety and industry dynamics. 原文： arstechnica.com