Скачать или смотреть Deflating Deflationism: A Critical Perspective on Debunking Arguments Against LLM Mentality

Link to Arxiv Paper: https://arxiv.org/abs/2506.13403

This video delves into the philosophical debate surrounding the mentality of Large Language Models (LLMs), as presented in the research paper "Deflating Deflationism: A Critical Perspective on Debunking Arguments Against LLM Mentality" [03:55]. The paper, a collaboration between the Institute of Philosophy, Google's Paradigms of Intelligence Team, and the University of Cambridge, examines the arguments of "inflationists," who believe AI has a form of thinking process, and "deflationists," who see AI as mere machines [04:04, 04:22].

The video focuses on the paper's efforts to logically dismantle deflationist arguments through strategies like the "robustness strategy," which questions an LLM's cognitive abilities based on its performance on new or slightly altered tasks [12:02]. It also explores the "aetiological strategy," which relies on pure logical analysis to draw conclusions about LLM mentality [21:26]. The paper also presents counterarguments from a "modest inflationism" perspective, offering logical arguments in favor of AI mentality [22:46]. The speaker also shares his own philosophical viewpoint, which aligns with the inflationist perspective, arguing that since humans are not perfectly generalizable, it's unreasonable to expect current AI models to be [08:38, 15:05].