factual

A/B Test Parameters On Real Users To Improve AI

Rather than manually guessing which model parameters produce better AI photos, you can randomly A/B test parameters on a small percentage of real users, measure which outputs they favorite or download, test for statistical significance, and roll out the winning parameters to everyone—letting users do the search through parameter space.

factualpending

Speaker

Pieter Levels

Evidence Quote

“Just use the users to improve themselves. And so that was the breakthrough to make it better.”

Source

Pieter Levels— Lex Fridman Podcast
Created: 6/13/2026, 3:32:34 AM

Your notes

Loading notes...