Nearly 30% of US STEM PhD dissertation abstracts are now AI-generated, NBER paper finds
lmoroney · x · 2026-10-21
An NBER working paper by Daniel P. Gross and colleagues finds nearly 30% of US STEM PhD dissertations now have AI-generated abstracts. Those abstracts run 300 to 1,000 words, and AI writing in the abstract predicts more AI writing in the body.
The updated version adds that almost nobody discloses AI writing unless their university builds disclosure mechanisms — a notable signal for academic integrity policy and how deeply AI has penetrated graduate writing.
More from AGI Musings
- Agent growth is capped by scenarios and connectors — real opportunity lies in offline sensor data — yangyi · 2026-10-21
- Pro-level pixel-watching action models could break competitive gaming anti-cheat — arnie_hacker · 2026-10-21
- Matt Shumer tells founders: businesses selling what agents now do for free won't survive — mattshumer_ · 2026-10-21
- 'Made without AI' is becoming the new 'GMO-free' label — AIandDesign · 2026-10-21
- Founder Joseph Jacks: Transformers are search algorithms, not intelligence algorithms — JosephJacks_ · 2026-10-21
- Turing's theory of computing is only a theory of mechanism, most computation remains beyond reach — fkasummer · 2026-10-21