Primary sourcerohanpaul_ai
View sourceEvent ·Single source
Science's Last Exam 用可复现验证给科研智能体打分
AItonomy Foundation 推出基准 Science's Last Exam(SLE),包含 113 个任务,覆盖 9 个学科。它不是让 AI 答有标准答案的题,而是给智能体一个研究目标、数据和实验软件,再用智能体看不到的隐藏检查来评分。任务的验证标准由领域专家在构建时封存,成果可以被打开重跑。SLE 用训练截止日期之后发表的论文来核对 AI 假设,防止记忆作弊。结果显示 DeepSeek V4.1 Flash 领先,其 21.28% 的假设预测了后续研究。
Current read
给科研智能体出开放题,用隐藏检查打分,还能防背题,DeepSeek V4.1 Flash 猜中了两成多后续研究。
2 sources58° AI heatLast updated 2026/11/5 19:04:36
Evidence
2 sourcesRelated source 1elvis
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.