LLM output distributions being factored doesn't mean they can't think out of order
Aaroth · x · 2026-09-26
A technical exchange arguing that whether an output distribution is represented as a product of next-element distributions (as in LLMs) has no bearing on whether the underlying system can "think out of sequential order" — since all distributions admit such a factorization, pointing to that representation can't settle the question. The author explicitly avoids claiming human brains work like LLMs.
More from AGI Musings
- A passerby spots a homeless person talking to ChatGPT for the first time — ctjlewis · 2026-09-26
- Allie Miller: proactive AI agents will break business models built on human error and slippage — alliekmiller · 2026-09-26
- Robotics professor on NeurIPS results and the value of 'A+B' research — GuanyaShi · 2026-09-26
- Researcher argues AGI could boost meaning and cultural variety via individual affordance — akbirthko · 2026-09-26
- Bill Gates: AI could be humanity's greatest equalizer or a major source of catastrophe — Planhub-ca · 2026-09-26
- Jensen Huang's 'just don't release unsafe products' answer called the weakest point of his Ezra Klein talk — soumitrashukla9 · 2026-09-26