It's 2026 — Duke Libraries explainer on why LLMs still hallucinate, from guess-favoring benchmarks
ArtificialOther · x · 2026-10-10
Duke University Libraries' Hannah Rozear examines why LLMs still make things up in 2026. A 2025 survey of Duke students found 94% believe generative AI accuracy varies significantly by subject and 90% want clearer disclosure of limitations, yet 80% still expect personalized AI learning within five years. A core reason: LLM benchmarks reward guessing over admitting uncertainty — training and evaluation incentives discourage saying "I don't know" (citing OpenAI's "Why Language Models Hallucinate").
More from AGI Musings
- UIUC, Google, Stanford, Berkeley & CMU release survey on proactive AI agents — liliang_ren · 2026-10-10
- AI as bureaucracy's amplifier: the Jevons paradox is coming for paperwork — r0ck3t23 · 2026-10-10
- Late psychiatrist Dan Stein's parting vision: AI and big data for mental health — PTenigma · 2026-10-10
- Ethan Mollick: 100x more PowerPoint or code isn't always progress — emollick · 2026-10-10
- Schmidhuber: soon humans can't be in charge, super-AIs will rival each other — haider1 · 2026-10-10
- Lenny Rachitsky says two-pizza teams are out, two-slice teams fit the AI era — lennysan · 2026-10-10