AI audit of 138 economics papers finds 2,035 errors; 97% of directional errors inflate claims
eigenrobot · x · 2026-10-29
Tyler Cowen published an AI-powered audit of Emmanuel Saez and Gabriel Zucman's economics research, which eigenrobot shared as "brutal." Across 138 papers the audit found 2,035 problems, 147 serious enough to change headline numbers; 69% of papers had at least one issue, and 97% of directional major/moderate errors overstated the paper's own claims. Method: two independent agent audits (Claude Opus 5.5 and OpenAI's GPT-6 Astra) recomputed headline numbers from the authors' replication files and original sources, then a third adjudicator agent tried to refute each finding and graded severity. Cowen argues the work and its policy implications need radical revaluation—and this is only step one of AI revising economic science.
Related event: AI Audit Finds 2,035 Errors Across 138 Economics Papers(4 posts)→
More from AGI Musings
- Turing Post traces the quest for recursive self-improvement from I. J. Good to Flint — TheTuringPost · 2026-10-29
- Mark Cuban: a trillionaire in the basement will run their own agent swarm — rohanpaul_ai · 2026-10-29
- Tokens Translate Capital Into Context: The Analogy Framing Anthropic as a Bank for Frontier Labor — curious_vii · 2026-10-29
- Your model saturated your applied eval? Just deploy it, says xeophon — xeophon · 2026-10-29
- "That would be ad fraud": the awkward question of Google AI agents clicking ads — firstadopter · 2026-10-29
- AI researcher Toby Walsh launches new book 'God AI' on AGI existential risks — TobyWalsh · 2026-10-29