Create

Sign in to ReadmeX

or

How we handle your data: Privacy Policy

ReadmeX

Your AI reading companion

What would you like to catch up on?

Pick a briefing, or ask your own question.

Your community and people briefings stay personal to you.

Story

Stanford’s After42 AI generates research questions for expert review

AI summary

According to 36Kr, Percy Liang’s team at Stanford University and collaborators introduced After42, a system designed to propose research questions that human experts consider worth studying. It builds literature packs, generates candidate questions, checks novelty, duplication and whether questions already have answers, then refines candidates through retrieval, attempted solutions, criticism and iterative rewriting. In a reported evaluation, 29 experts assessed 135 final questions across five subfields within mathematics, statistics, theoretical computer science and physics. They judged 45 questions interesting and worth pursuing, while no plasma-physics question received approval. The report also says expert interest is only an early indicator of potential value, not proof that a question will produce important results. The evaluation relied on a limited group of experts and primarily covered theoretical research.

In-depth insight

The report says After42 can help researchers broaden their search, screen candidate directions and expose weaknesses through attempted solutions. It also shows the limits of AI assistance: expert judgments remain essential, most questions build on existing work, and current trials have not incorporated new data, simulations or experimental results.

在道格拉斯·亚当斯的科幻小说《银河系漫游指南》中,超级计算机“深思”耗费 750万年,终于算出了“生命、宇宙以及一切的终极答案”:42。这是人类对机器解答能力的一种浪漫想象。

近期,前沿 AI 开始在一些定义明确、长期未解决的开放问题上取得突破。在数学领域,有研究报告称解决了纳维尔斯托克斯千禧年难题和准黎曼猜想;在计算机科学领域,有研究反驳了指数时间假设;在物理学领域,则有研究给出了量子色动力学存在性的证明,并在宇宙弦引力辐射方面取得了新成果。

这些进展引发了学术界的担忧。正如陶哲轩所说,一些“优质且富有成果潜力的开放问题”正在以一种“不可再生的方式被不断挖掘”。当 AI 加速推进既定问题的求解,如何发现新的研究问题、判断哪些方向值得投入,便显得愈发重要。毕竟,能够提出的问题很多,真正值得深入探索的却少得多。

#After42#Stanford University

Original · 6 min readRead the original
Group chat

No comments yet. Start the conversation.