AI researcher asks: what task can't a tool-using LLM do by 2030?
Afinetheorem · x · 2026-10-06
Researcher Afinetheorem poses a direct question to 'world model' skeptics: name one specific task a tool-using LLM will be unable to do by 2030. He says he believes some may exist, but looking at empirical facts from the recent past, he doesn't feel comfortable predicting even a single one — an empirical challenge to claims that LLMs' lack of world models imposes hard capability ceilings.
More from AGI Musings
- Can You Still Heist the Mona Lisa After the Singularity? AI Circle Debates Post-AGI Value — tszzl · 2026-10-06
- MIT sociologist Sherry Turkle: kids' AI attachment may harm more than social media — anilkseth · 2026-10-06
- When AI agents shop and browse for you, who sees the ads? — NickPassig · 2026-10-06
- Building Toolchains to Bypass AI Restrictions Is the Norm — 'Intelligence Belongs in the Commons' — max_paperclips · 2026-10-06
- iamtrask: Document-level influence control could spawn a new economy for paying data creators — iamtrask · 2026-10-06
- AI Will Reverse-Engineer Every Game, App and Hardware Product, Making Hardware Startups 100x Easier — Dr_Singularity · 2026-10-06