Commenter argues LLMs can asymptotically approach but never exceed AGI
GlenBradley · x · 2026-10-05
Glen Bradley pushed back on the claim that faster learning plus more intelligence equals ASI, arguing the current LLM paradigm has a mathematical ceiling:
- By definition, ASI must be superior to human intelligence; as long as humans can conceive of something genuinely new and an AI cannot, ASI is a logical impossibility
- Under the LLM paradigm, AI can only approach AGI asymptotically and never cross beyond it — he calls this mathematical certainty, not a feeling
- Core argument: learning from the existing human corpus of knowledge caps AI at roughly the smartest human
A representative "paradigm ceiling" take in the AGI-boundaries debate, opposed to the scaling school.
Related event: Commentator Argues LLM Paradigm Mathematically Cannot Reach ASI(2 posts)→
More from AGI Musings
- DHH pens anti-doomer essay: stop fretting about AI and climate, choose to bloom — banteg · 2026-10-06
- Viewing LLM inference as a séance: how much of a model's output comes from the dead — NYCounihan · 2026-10-06
- Pew Research findings suggest Dead Internet Theory may be coming true — Altruism7 · 2026-10-06
- The 1970s Hollywood playbook for surviving AI: everyone becomes the editor — lauriewired · 2026-10-06
- On AI job-loss stats: three things can be true at once, beware single-narrative data — soumitrashukla9 · 2026-10-06
- Dev asks: how many startups would die if Instinct/Muse/Dots/Grok Bot were actually good? — TejasKumar_ · 2026-10-06