I'm Om. I work on interpretability — figuring out what's actually happening inside large models.
The writing on this site is technical first, with the occasional detour into film and music. I keep the two apart just enough that you can subscribe to one without the other — but honestly, they leak.
Currently
Researching feature circuits and eval reliability. Training sparse autoencoders on small open models and writing up what falls out — including the parts that don't work.
If you're an AI/ML startup working on something adjacent, I'm easy to reach.
Off the clock
Cinema and music, in unreasonable quantities. Long films, slow albums, liner notes read cover to cover. Some of it ends up in the writing; the rest just ends up in me.
Elsewhere