factual

LLMs Only Proven To Predict Next Token

Large language models are trained for and demonstrably good at exactly one thing—predicting the next token—and any other attribute attributed to them (like reasoning or understanding) is the observer's generous intuition that has not been proven; despite massive effort, it is quite likely they are not doing anything beyond producing smooth prose.

factualpending

Speaker

Leslie Valiant

Evidence Quote

“these large language models are clearly trained for one thing, which is predicting the next syllable, the next token, as they call it. And that's what they're trained for, they're very good at that, they're hard to beat. And any other attribute you give them is your intuition. And people seem to be very generous to large language models”

Source

272 | Leslie Valiant on Learning and Educability in Computers and People— Sean Carroll's Mindscape
Created: 6/13/2026, 12:22:31 AM

Your notes

Loading notes...