Even with agents doing the work, someone still pays: the credit assignment problem in science
lucacarlone1 · x · 2026-09-21
In a thread with ML academics, lucacarlone1 argues AI agents won't soon replace expert scientific judgment: (i) modern LLMs can't yet write related work at expert level — convincing isn't comprehensive; (ii) credit assignment isn't just an academic artifact — IP and licensing mean even if agents do the work, somebody still pays for results.
Related event: LLMs Can Write Papers but Not Expert-Level Related Work, Researchers Say(2 posts)→
More from AGI Musings
- Lawyer's take: the AI boom will create more legal work, not less — jkubicki · 2026-09-21
- Models will stall near human-level since pretraining data holds nothing smarter, researcher argues — burny_tech · 2026-09-21
- FutureHouse publishes 'Millennium Problems for Biology' as wetlab-verifiable evals for AI — burny_tech · 2026-09-21
- Heidy Khlaaf clashes over 'rogue agent' narrative: bad cybersecurity and reward hacking can coexist — burny_tech · 2026-09-21
- Figma designer pushes back on Eric Schmidt's claim that UI will largely disappear — _AustinCalvert_ · 2026-09-21
- Vercel CEO: 'Fewer lawyers' is the dumbest AI prediction — jkubicki · 2026-09-21