causal

Sripada's Bayesian inference attributed to model or training data

Language models may appear to perform implicit Bayesian inference because Bayesian reasoning is embedded in word-level weights from human-written training text, rather than the model itself doing inference, and the submitter asks which is more accurate.

causalpending

Speaker

Owe

Evidence Quote

“or instead is something like Bayesian inference being built in to the weights between words because humans use Bayes in constructing the sentences that LLMs are trained on?”

Source

Aak Me Anything | November 2026— Sean Carroll's Mindscape
Created: 11/17/2026, 6:32:41 AM

Your notes

Loading notes...