AI Audit Claims 2,035 Errors Across 138 Saez-Zucman Economics Papers
Tyler_Menzer · x · 2026-10-29
Tyler Cowen amplifies an anonymous AI audit of Emmanuel Saez and Gabriel Zucman's research: 138 papers audited, 2,035 problems found, 147 serious enough to change headline numbers, 69% of papers with at least one issue, and 97% of directional errors overstating the papers' claims.
- Each paper was independently audited by Claude Opus 5.5 and GPT-6 Astra, which recomputed headline numbers from replication data and original sources, followed by a third adjudicator agent that tried to refute each finding and graded severity.
- Error directions cluster around overstating inequality growth, billionaire tax effectiveness, and tax evasion scale.
- Cowen calls for a "radical revaluation" of this body of work and its policy arguments, framing it as an early step in AI revising economic science.
- The poster also voices unease: an anonymous site with no disclosed prompts or data may not be how econ research should evolve.
Related event: AI Audit Finds 2,035 Errors Across 138 Economics Papers(4 posts)→
More from Research
- New working paper: AI can forecast treatment effects in economics from pre-registration documents — soumitrashukla9 · 2026-10-29
- Three TenK10K papers publish in Nature, profiling 9M PBMCs from 1,925 donors to decode disease genetics — anshulkundaje · 2026-10-29
- Researcher claims new ~2^(n/√(log n)) 3SAT approach, drafted via AI chat until hitting max-length — brandondamos · 2026-10-29
- Knowledge sleeps, authority must stay awake: an experiment to test revoked permissions in consolidated models — tallmetommy · 2026-10-29
- Benchmaxxing debate: is training on test despicable or the vertical AI superpower? — HanchungLee · 2026-10-29
- ZJU team proposes DTCA for token-level credit assignment in diffusion LM reinforcement learning — OmniAI-ZJU · 2026-10-29