论文

Gary Marcus:LLMs不适合对齐

精选理由

Gary Marcus在TAIS 2026上解释了为什么LLMs不适合AI对齐,并提出了替代方案,他是AI领域的批判性声音。

纽约大学荣誉教授Gary Marcus在TAIS 2026会议上表示,大语言模型在本质上不适合AI对齐。这位AI领域的批判性声音是Robust.

正文

AI和Geometric.AI创始人,后者已被Uber收购。

Marcus在演讲中挑战了当代AI,并暗示了可能更有效的替代方案。他的研究涉及人类语言发展和认知神经科学领域。

原文 · Gary Marcus

LLMs are constitutionally ill-suited to alignment. What might work better instead? Technical AI Safety Conference (TAIS) @tais_2026 Gary Marcus @GaryMarcus (Emeritus Professor, New York University) took the stage at TAIS 2026 and made the case that, "LLMs are not the way to alignment." Gary argued that LLMs are constitutionally ill-suited to alignment, and offered some hints at what might work better instead. Gary is one of AI's most prominent critical voices — a scientist, best-selling author, and serial entrepreneur (founder of Robust.AI and Geometric.AI , the latter acquired by Uber). He's known for challenging contemporary AI, anticipating many of its current limitations decades in advance, and for his research in human language development and cognitive neuroscience. Watch his talk from TAIS 2026 on our YouTube Channel: youtu.be/_jhvM8Af-bM 🔗 View Quoted Tweet📊 7 ⚡

产业坐标与信源核验

本条情报归属 AITOP「论文」核心栏目,重点聚焦北美前沿实验室与硅谷科技巨头。在跨区域产业观察坐标中,该动态直接关联「AI 开发者工具链、Coding Agent 与工程化实践」研究主线,反映出当前技术演进与商业落地的最新风向。涉及核心实体与技术议题包括:Gary Marcus、LLMs、AI对齐、TAIS 2026,读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。

信源核验说明:本文原始线索采编自「Gary Marcus」(@GaryMarcus),经 AITOP 多维评分管线结构化提炼。