UCSD Hao AI Lab· X:UCSD Hao AI Lab (@haoailab)·· 3 小时前AI 评分27
AI 侦探推理基准测试设置介绍
AI 导读
该侦探推理基准测试中,每个案件包含 100 个地址、10 名嫌疑人和 12 小时时限,要求指出凶手、动机和藏身点。侦探团队分两个角色:Holmes 负责推理决策,可选 GPT6-Astra、Jev 或证据计数启发式;Irregulars 负责搜索,可选 Jev 智能体或搜索启发式。
正文
1️⃣ The setup: every case has 100 addresses, 10 suspects and 12 hours to name the culprit, motive and hideout. The detective team include two roles:
🕵️ Holmes reasons and decides: GPT6-Astra, Jev, or a evidence-counting heuristic
👦 The Irregulars search: Jev agents, or a search heuristic
来源:UCSD Hao AI Lab · x.com