Gary Marcus: OpenAI's Astra is Vastly Oversold, Math Breakthroughs Don't Mean AGI
Gary Marcus · rss · 2026-08-03
In response to the hype around OpenAI's internally tested model Astra solving major mathematical problems, Gary Marcus argues that the celebration of "AGI arriving" commits the Fallacy of Composition.
Marcus points out that Astra's success in specific mathematical domains does not mean it will excel across all cognitive domains, nor does it solve hallucination or reliability issues. He presents several core arguments:
- Math is a Special Case: Math and coding heavily rely on external tools for strict verification and can generate massive amounts of guaranteed-correct synthetic data. This verification mechanism cannot be directly applied to open-ended real-world problems.
- Lack of Methodological Disclosure: OpenAI's demonstration is marketing, not science. There is no information on how many problems were attempted to get the 10 successes, nor is the true total cost—including the salaries of top mathematicians—calculated.
- Generalization is Questionable: Equating a domain-specific breakthrough with omniscient ASI shows a fundamental misunderstanding of the multidimensional nature of intelligence.
More from AGI Musings
- VC warns a "digital aristocracy" of LLM-fluent workers is emerging — StewartalsopIII · 2026-10-16
- DHH: Starship, Starlink and Grok should have been Europe's Sputnik moments — georgemillo · 2026-10-16
- Hinton hits back at Jensen Huang: 'He has good reason to say AI isn't dangerous — he sells Nvidia chips' — BlackHC · 2026-10-16
- Yudkowsky blasts new interpretability finding as primitive; critic pushes back on doomer logic — AlexTensor · 2026-10-16
- Beff Jezos declares 'civilizational inflection point' in the acceleration of technological progress — beffjezos · 2026-10-16
- Data scientist: pivoting to AI engineering is the safest career bet for the next 5 years — mdancho84 · 2026-10-16