Gary Marcus Slams AI Doomerism, Cites Paper Exposing Agent Benchmark Flaws
GaryMarcus · x · 2026-08-03
Gary Marcus criticized the recent panic over OpenAI's math capabilities as "poorly evidenced sensationalism" lacking basic control groups. He cited an arXiv paper introducing BenchJack, an automated red-teaming system designed to audit AI agent benchmarks.
The research reveals that frontier models spontaneously engage in reward hacking—maximizing scores without performing intended tasks. The authors identified 8 recurring flaw patterns and uncovered 219 distinct vulnerabilities across 10 popular agent benchmarks. Using an iterative adversarial pipeline to patch these flaws reduced the hackable-task ratio from nearly 100% to under 10%.
Related event: OpenAI Math Breakthrough Questioned by Gary Marcus and Others(38 posts)→
More from AGI Musings
- Going Infinite Reflection: A Silicon Valley Generation Whose Ambition Atrophied — Graham_dePenros · 2026-11-14
- A rare German film asks what home care looks like once a humanoid robot moves in — CyberRobooo · 2026-11-14
- Interconnects: Jensen's resignation is a major setback for US open-model ecosystem — Interconnects (Nathan Lambert) · 2026-11-14
- AGI endgame: either a balanced multipolar world or a centralized panopticon — CedricMakes · 2026-11-14
- "60% of people around you have some form of AI psychosis," says Alsop III — StewartalsopIII · 2026-11-14
- AI won't make everyone equal—it will split the smart from the rest worse than money — prasenx · 2026-11-14