谷歌：2026年“猜想机器”：AI智能体与科研领域全新验证瓶颈报告（英文版）

Base models hold a great deal of explicit scientific knowledge, the kind written down in papers and textbooks. What they lack is tacit knowledge: the hard-to-articulate craft, built up over years of practice and failure, that lets a scientist coax a cell line into growing or get a simulation to run well. Until now, scientists using AI tools had to convey their methodological know-how, processes, and preferences the hard way: through detailed prompts, bespoke scaffolding, or model retraini