What a crossover!

This is amazing

They're in their move fast and break things era 🙃

This take emerged organically from just how well our method on SAEs for hypothesis generation (HypotheSAEs) performed, which surprised all of us!

See the paper arxiv.org/abs/2506.23845

Thanks @kennypeng.bsky.social, Jon, @emmapierson.bsky.social, @nkgarg.bsky.social for another nice collaboration.

This capability of discovering unknown concepts opens many opportunities for applied machine learning. We can design better whitebox predictors, better audit high-stakes models for bias, and generate hypotheses for CSS research. More broadly, SAEs can help bridge the "prediction-explanation" gap.

These tasks lie in contrast to probing, where we're trying to predict the presence of a *known* concept; and steering, where we're trying to include a *known* concept in an LLM output. SAEs lose to simple baselines on these tasks. (2 good papers on this: "AxBench" and Kantamneni, Engels et al. 2025)

How do we reconcile our view with recent negative results? Our key distinction is that SAEs are useful when you don't know what you're looking for: how does my text classifier predict which headlines will go viral? How does my LLM perform addition? These are "unknown unknowns".

Nice work! Cool to see that item difficulty predicts human-llm disagreement. We also studied similar questions with the DICES dataset: aclanthology.org/2024.emnlp-m...

Huge congrats, Marianne!!

I find that I've actually gone out of my way to stop using bullet points in reviews now because Any Review With Bullet Points is a Bot 🥲

So awesome, congrats Lucy!!! 🧀