Gary Marcus Questions Inconsistency in AI Model Transparency Reporting

GaryMarcus · x · 2026-09-02

Gary Marcus criticizes an article about AI models (specifically mentioning Astra) for being internally inconsistent. He highlights the contradiction between the claim that models will reveal less of their 'thinking,' making them harder to monitor, and the assertion that researchers can still 'sufficiently monitor' the reasoning. Marcus questions the criteria and evidence behind this sufficiency and warns of a slippery slope, urging greater public clarity.

Related event: Gary Marcus Clashes with Reporter over Gemini Astra Safety Story(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →