Complementary remarks from Gary Marcus and Terence Tao on OpenAI’s giant math drop

Substack · Watch/Read · (gathered 2026-10-14)

• Gary Marcus argues that the “real news” from OpenAI’s “giant math drop” isn’t the mathematical results themselves, but the unscientific and vague manner of their presentation, which lacks the transparency required for peer review.
• He criticizes the absence of crucial details such as the model’s architecture, the exact procedure used (“Same procedure”? “Using an unreleased model”?), failure rates, and training/data augmentation methods, making it impossible to assess the generalizability of the results.
• Marcus warns against the “ignorant cheering section” on social media, emphasizing that without scientific transparency, it’s impossible to determine if the system represents a legitimate step toward AGI or merely a “clever leveraging of Lean and synthetic data” with no broader generality.