斯坦福教授的研究代理以 12.57 万美元售出
Percy Liang 团队据称花 14 个月做出基于 GPT-6 Astra 的研究代理，一分钟读 1100 万篇论文、找矛盾、提假设，一家公司花六位数买下使用权。
stanford professor sold a research bot for $125,700. not a saas landing page. a machine learning agent. percy liang. stanford. llms. the clip says he spent 14 months building it, then one company paid six figures for access. powered by gpt-6 astra, the bot claims it can chew 11 million scientific papers in a minute. extract key findings. flag contradictions. spit out new research hypotheses. turns out the money is the boring part. the product is what happens when one agent gets millions of papers in one pass and starts finding the fights the literature never put in a slide deck. what the system actually does, if the reel is right 1. ingest the corpus at research scale, not pdf by pdf 2. pull findings into a structure a human can audit 3. surface contradictions across papers that never cite each other 4. propose hypotheses a lab can try to kill next the uncomfortable part $125,700 is not the price of magic. it is the price of a closed loop: papers in, contradictions out, hypotheses queued. most teams still hire interns to read abstracts. this one sold the reader. would i pay six figures for it. only if the contradiction map ships with the seat. otherwise you bought a fancy summarizer with a stanford name on the invoice.