Debate: can LLMs pass an hour-long Turing test? Skeptic says the illusion fades fast
LucaAmb · x · 2026-10-06
Luca Ambrosio and gabriberton debate whether LLMs can pass Turing tests. gabriberton notes studies show LLMs succeed when asked, but their goal is being helpful, not impersonating humans. Luca counters that existing studies use short, restricted interactions: he believes no LLM can reliably pass an hour-long test, and long conversations with character LLMs reveal the illusion—strong at first, it gradually vanishes as models lack internal will and transformative insights.
Related event: Debate Flares Over Whether LLMs Can Pass the Turing Test(2 posts)→
More from AGI Musings
- Even in an AGI world, interest rates could stay high if compute capex dominates borrowing — menhguin · 2026-10-06
- Women and young people are much gloomier about AI, survey analysis finds — HaydnBelfield · 2026-10-06
- Slow takeoff means deflation, fast takeoff means pumping rates: two AGI macro scenarios — menhguin · 2026-10-06
- Reddit parents debate raising kids who may face superintelligence by 18 — Tallwhitedude123 · 2026-10-06
- Anthropic's internal model found the first truly subquadratic 3SUM algorithm, preprint claims — AMBNNJ · 2026-10-06
- Sakana AI CEO at STS forum: AI edge shifts from single models to system orchestration — SakanaAILabs · 2026-10-06