推文

@GenAI_is_real · 2026-10-11 13:33

Decision models are evolving surprisingly fast. What makes Microsoft-Decision-1 especially interesting to me is how naturally this fits Microsoft's own ecosystem. Think about Word, Excel, Outlook, or basically any large productivity application. A huge fraction of interaction is not open-ended generation. It is deciding which action to take next: open this menu, apply this format, accept this change, route this request, retry, stop, ask for confirmation. These are small, structured decisions repeated millions of times. Using a large reasoning model for every one of them feels wasteful. A fast, calibrated decision model that chooses among predefined actions could become an extremely important layer underneath computer-use agents. Microsoft owns an enormous surface area of software full of exactly these decisions. If this class of model keeps improving this quickly, the opportunity is much bigger than another benchmark category.

曝光 1043 · 评论 1 · 点赞 5 · 书签 4 · 曝光/时 185.37651123328283