AI 服務暫時不可用，以下為來源正文，待恢復後補全翻譯。
Imagine giving an AI a difficult assignment and returning tomorrow. The interesting question is what happened while you were away. Did it inspect the right evidence? Recover from mistakes? Produce something that survives a test? Intelligence becomes useful when it can sustain that chain of work. The week of September 21–27 brought Anthropic’s Claude Opus 5.5 and a striking report of AI-assisted biological discovery. DeepSeek’s recent environments paper, posted September 19, provides the infrastructure story connecting them. Together, they suggest that progress depends increasingly on the machinery surrounding a model: where it acts, what it observes, and how its conclusions are checked. Opus and the cost of completing work Read more