Researcher argues LLMs can't 'understand' and shouldn't make any decisions
gerardsans · x · 2026-11-21
In a debate on X, gerardsans argues that humanizing LLMs is misleading: they aren't smart, intelligent, or capable of understanding in any meaningful sense — it's more accurate to discuss data coverage, sampling, utility, and error rates. He stresses LLMs shouldn't be used to make any decisions, since the model itself doesn't know that it shouldn't be.
More from AGI Musings
- Don't anthropomorphize LLMs: talk data coverage and error rates instead — gerardsans · 2026-11-21
- Gary Marcus eulogizes the generative AI mania: 'I warned for three and a half years' — GaryMarcus · 2026-11-21
- Researcher Quips: Instead of Calling AI Superintelligent, Admit Humans Are 'Stupid' — gandamu_ml · 2026-11-21
- 'Math Is Solved Now, What's Next' — A Quip on AI's Math Progress — prasenx · 2026-11-21
- Beyond Riemann: Could AI Prove the Twin Prime Conjecture Next? — jamestagg · 2026-11-21
- Career advice for the AI era: if your job rewards intelligence, pivot to 'stupid stuff' — pwlot · 2026-11-21