An inside look at X’s Community Notes | Keith Coleman & Jay Baxter
Lenny's Podcast episode, sponsored by WorkOS and Productboard, in which Lenny Rachitsky talks with Keith Coleman, product lead for Community Notes, and Jay Baxter, its founding ML engineer, about how X's crowdsourced fact-checking actually works. They explain the core mechanic: anyone can propose a note on a misleading post, others rate it, and the algorithm looks for agreement from people who have disagreed in the past - that surprising cross-aisle consensus is what makes notes neutral, accurate and well written. They cover the principle that all of humanity participates, with no exemptions for Elon, government figures or even advertisers, the internal Twitter contest from which the algorithm emerged, how the product survived four leadership eras including Jack and Elon, and independent studies finding that notes change whether people agree with a post's core claims - which is why Meta adopted the model. PARTIAL COVERAGE: this summary highlights the mechanism and principles; algorithmic details and later segments are not covered. These are the guests' own claims and cited studies, not independently verified.
Executive Summary (TL;DR)
Lenny's Podcast episode, sponsored by WorkOS and Productboard, in which Lenny Rachitsky talks with Keith Coleman, product lead for Community Notes, and Jay Baxter, its founding ML engineer, about how X's crowdsourced fact-checking actually works. They explain the core mechanic: anyone can propose a note on a misleading post, others rate it, and the algorithm looks for agreement from people who have disagreed in the past - that surprising cross-aisle consensus is what makes notes neutral, accurate and well written. They cover the principle that all of humanity participates, with no exemptions for Elon, government figures or even advertisers, the internal Twitter contest from which the algorithm emerged, how the product survived four leadership eras including Jack and Elon, and independent studies finding that notes change whether people agree with a post's core claims - which is why Meta adopted the model. PARTIAL COVERAGE: this summary highlights the mechanism and principles; algorithmic details and later segments are not covered. These are the guests' own claims and cited studies, not independently verified.