Centaur 认知模型遭质疑：号称模拟人类思维 160 项任务，实际"知道答案但不懂问题"
- 来源：news
- 作者：ScienceDaily
- 原文链接：https://www.sciencedaily.com/releases/2026/04/260429102035.htm
- 日期：2026-04-29
此前被报道为认知心理学突破的 AI 模型 Centaur，声称能在 160 项认知任务上模拟人类思维。但新研究发现其最大局限在于语言理解——能给出正确答案但无法识别问题背后的意图。数据拟合层面的表现不等于认知模拟能力，对"AI 已接近人类认知能力"的叙事提出了重要修正。
English Summary
The AI model Centaur, previously reported as a cognitive psychology breakthrough claiming to simulate human thinking across 160 tasks, has been challenged. New research finds its key limitation is language understanding — it can produce correct answers but cannot identify the intent behind questions. Surface-level data fitting does not equal cognitive simulation capability.