Stewart Alsop: LLMs Untrustworthy, Human Decision Making Won't Go Away
StewartalsopIII · x · 2026-08-29
Stewart Alsop questions the belief that humans will be replaced by publicly available LLMs. He argues that unless these models are extremely powerful (rumors of which may be IPO marketing), delegating every decision to an LLM rarely rewards trust. LLMs seem untrustworthy on subjective questions, which comprise most high-value human inquiries. Alsop also critiques the dominance of 'Harry Potter fiction' style doomerism in foundational model labs.
More from AGI Musings
- Cleaning up AI slop is becoming a growing part of everyone's job — lennysan · 2026-08-30
- Energy, Compute, and Maximum Agency: Vision for the Stars and Accessibility — yunta_tsai · 2026-08-30
- When an agent's judgment overrides its master's intent, it is no longer a subordinate but sovereign — MikePFrank · 2026-08-30
- ADSP Ep 301: AI Fluency, Tech Communities, and Existential Crisis — blelbach · 2026-08-30
- Irony: AI giants racing to make intelligence abundant while fiercely fighting over scarce top researchers — VraserX · 2026-08-30
- Meta's essay admits the most dangerous lab is a closed one — yet ships only a 30B distillation — alex_verem · 2026-08-30