Bostrom: Calling GPT-6 AGI may cause less confusion than denying it
i_dg23 · x · 2026-10-19
In a conversation with Roman Yampolskiy, Nick Bostrom argued that given today's fuzzy definitions, calling GPT-6 AGI might be less misleading than denying it. Show GPT-6 to people 15 years ago — writing award-caliber fiction, potentially solving Millennium Prize Problems — and it would unquestionably count as AGI. Still, he noted gaps versus humans: continual learning, research taste, physical manipulation, which is why humans remain essential across the economy and AI labs. He defines superintelligence as a system decisively outperforming the best humans in every practically important domain, and says we haven't reached it; once all capabilities exceed human level, that's already superintelligence.
More from AGI Musings
- Compute wins the AI race: Meta caught up and xAI scaled fast by building compute — haider1 · 2026-10-19
- US retailers adopt AI-assisted pricing tools as consumer apprehension grows — ArtificialOther · 2026-10-19
- David Patterson predicts superintelligence will cure all diseases by 2030 — davidpattersonx · 2026-10-19
- GDC 2026 survey: 64% of game devs skip genAI as negative views hit 52% — IanArawjo · 2026-10-19
- Can Neural Nets Produce Qualia? A Philosophical Exchange on AI Consciousness — brandon_xyzw · 2026-10-19
- Only 13% of 3,362 automation case studies use AI; agents appear in just 18 — Sonia_here89 · 2026-10-19