Meta finds AI judges prefer AI slop — bad news as agencies score proposals with AI
jmugan · x · 2026-09-29
Meta researchers taught AI to make its writing less AI-sloppy, but the more striking finding is that standard AI judges preferred AI slop over human writing because slop covers every rubric item. The author draws an ironic implication: since funding agencies increasingly use AI to score proposals, AI-generated proposals may fare better by addressing all stated criteria, while human writers tend to focus only on what they consider most important.
More from AGI Musings
- So8res: antitrust law may bar AI companies from coordinating to slow down — wfithian · 2026-09-29
- Zuckerberg: Superintelligence Will Create Major New Opportunities for 'All People' — Polymarket · 2026-09-29
- Critic takes on AI welfare: tools are uninformative, claims shape AI design — anshulkundaje · 2026-09-29
- dhh: 'Slop' is a comfort blanket for programmers stuck in AI denial — inductionheads · 2026-09-29
- NLP professor Yoav Goldberg is recruiting AI PhD students—but suggests waiting unless your topic needs no big compute — yoavgo · 2026-09-29
- Atlassian PM leader: In 10,000+ person companies AI expands roles instead of merging them — lennysan · 2026-09-29