Event ·Multi-source

LongChang 基准测试：LLM 能否自主发现被修改的物理定律

arXiv 论文提出 LongChang 框架，自动生成物理定律被修改的模拟世界，让 LLM 在无任务提示的情况下自主做实验发现异常规律。12 个受测 LLM 中，GPT-6-Astra 的定律发现率为 16.7%，其余 11 个均为 0%。探索记录显示模型普遍未能调查所有异常现象，对被修改定律的识别也不完整。

Current read

让模型在一个重力规律被改掉的世界里自己发现异常，12 个模型只有 GPT-6-Astra 做到一次，其余全挂，测试设计挺有意思。

11 sources38° AI heatLast updated 2026/10/27 17:57:00

Evidence

11 sources

Primary sourcearXiv cs.AI

2026/10/27 17:57:00View source

Related source 1rohanpaul_ai

2026/10/27 10:24:24View source

Related source 2OpenAI Blog

2026/10/26 22:00:00View source

Related source 3AIGCLINK

2026/10/26 23:38:03View source

Related source 4Simon Willison’s Weblog

2026/10/26 23:53:07View source

Related source 5marktechpost

2026/10/27 02:45:27View source

Related source 6Shubham Saboo

2026/10/27 03:01:36View source

Related source 7Decoder

2026/10/27 10:09:26View source

Related source 8IT之家

2026/10/28 02:01:04View source

Related source 9Artificial Analysis

2026/10/28 02:11:22View source

Related source 10kimmonismus

2026/10/28 02:56:11View source

Conflict 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.

Open the story