AI interpretability community still fighting 'SAE is dead' claims, one airport at a time

EigenGender · x · 2026-09-19

A self-deprecating tweet from the AI interpretability community: the author held up boarding for 10 minutes to explain to a stranger why "SAEs aren't dead, they're just misunderstood." The meme captures the ongoing intra-community debate over whether sparse autoencoders remain a valuable tool for probing model internals.

Related event: Researcher Misses Boarding Time to Defend SAEs at Airport(2 posts)→

Original post →

More from Fun

Fun channel →