Researchers Push Back on AI Audit of Saez-Zucman Papers Over Missing Prompts
felpix_ · x · 2026-10-29
Jason Crawford marveled that an AI audit found massive problems in Saez-Zucman research, but felpix pushes back: anyone can upload any paper and ask Claude for a critical review, and a dump with no prompts or raw output examples is disingenuous work—even though S&Z's research has plenty of valid criticism.
The audit site itself reports granular data: 138 papers, 2,035 problems, 147 headline-changing, with errors overwhelmingly pointing toward overstating claims (e.g., 188 vs 6 for 'inequality is higher/rising faster'). The fight is over methodological transparency, not the direction of the findings.
Related event: Critics Question Rigor of AI Audit of Saez-Zucman Research(2 posts)→
More from Research
- Stanford's FACT: Force-Aware Robot Foundation Model Trained on 2,000h Human Data — yining_hong · 2026-10-29
- 86M-param decoder-only transformer plays chess on Lichess, full notebook public — andreisavu · 2026-10-29
- Harvard experiment finds AI reviews dominate human peer reviews — RexDouglass · 2026-10-29
- Box2D WASM upgraded to V3.2 with open-source PBMPM solver coupling code and demo — Birchlabs · 2026-10-29
- Unprompted conf: model internals are the new Windows internals, says Matt Suiche — moyix · 2026-10-29
- Family-aware All of Us benchmark for polygenic prediction and genomic ML released — anshulkundaje · 2026-10-29