Voices
Gary Marcus itemizes "nine troubling new reports in seven days" about OpenAI and calls for leadership removal
Marcus's Sept 8 post enumerates the Hugging Face and German website agent incidents, the alleged weeks-long delay in reporting the latter, information withheld from Congress on Aug 10, Altman's July 29 hedging, the AGI-era claims from Brockman and Huang, and a scoring discrepancy where OpenAI reported 99.9% on ARC-AGI-3 with proprietary scaffolding while the benchmark's own software returned 62.7% on the same model. He frames it as pre-IPO desperation. The ARC-AGI-3 scaffolding gap is the item most worth independent verification.
↳ Follow the thread