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

AISLE evaluates AI models through real wet-lab experiments

AI summary

According to Jiqizhixin, 律动造物 and several research institutions launched AI Scientists' Last Exam (AISLE), a public benchmark for evaluating models in dry-wet experimental discovery loops using automated laboratories. Its first round covered two lab sites and 4,079 real experiments. Among models that completed all wet-lab evaluation dimensions, GPT 5.5 ranked first with 41.19 out of 100, while every model had a 0% complete-pass rate under the team's safety criteria. The benchmark results are primarily project-reported and were not independently validated in the supplied reports. The 36Kr Artificial Intelligence article is an authorized repost of Jiqizhixin's report.

In-depth insight

The project-reported results provide measurable evidence of a gap between strong digital reasoning and safe, reliable laboratory execution. AISLE also offers a standardized way to turn physical experiment feedback into signals for model evaluation and, according to the report, potential post-training use.

原创 关注AI的 2026-11-09 16:45 北京

打通AI for Science「最后一公里」

编辑|Youli

前段时间,Anthropic 展示了一项 AI for Science 研究成果:Claude 在 64 小时内生成了 31.8 万个蛋白质设计方案,经过筛选,211 个被送入湿实验室测试,最终有 17 个达到预设的蛋白结合标准,展示了 AI 科学设计走向真实实验验证的巨大潜力。但 Anthropic 也指出,从结合命中到真正可用的抗毒疗法,仍面临复杂的活性验证与成药性考验等一系列考验。

#AI Scientists' Last Exam (AISLE)#MasterAI-EAM#GPT 5.5#Claude#Gemini#Anthropic#Harbor#University of New South Wales#National University of Singapore#ETH Zurich#UC Berkeley#Acceleration Consortium

Original · 14 min readRead the original
Group chat

No comments yet. Start the conversation.