AI 模型可解释性之争:一篇论文引发三大实验室研究者反驳
可解释性圈子的公开论战,Leo Gao 直接写文章反驳,三大实验室的人都下场了。做 mech interp 或关心模型内部机制研究的人可以看看两边的论证。
The Information 报道,一篇主张 AI 模型比研究者此前认为的更容易理解的文章引发了多方争议。争议的核心是该文章对神经网络内部机制可解读程度的判断。OpenAI 研究员 Leo Gao 发表了题为 "Neural networks are not already clean" 的反驳文章,直接反对其观点。
Google DeepMind、OpenAI 和 Anthropic 三家实验室的研究者均参与了这场争论,说明可解释性研究社区对"模型内部是否已经足够干净可读"存在实质性分歧。这场讨论的走向会影响学界对机制可解释性研究价值的评估。
An essay arguing that AI models are easier to understand than researchers once thought is drawing pushback from researchers at Google DeepMind, OpenAI and Anthropic.
OpenAI's Leo Gao answered with a rebuttal titled "Neural networks are not already clean."
Full story: https://t.co/UmKNtayPyH
This report is categorized under AITOP's "Industry" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include OpenAI, Google DeepMind, Anthropic, interpretability.
Source attribution: curated from The Information (@theinformation).