resolution of singularities in.... wait

Neural networks are grown, not programmed. What does that growth process look like? Like this! This is a small language model (3M) across training, visualised with a new interpretability technique: susceptibilities. We call this handsome critter the rainbow serpent.

Sander Dieleman “Information” literally comes from Latin meaning “shaping”, or “giving form”.

Happy to announce our new paper on TDA is finally out! with @ Philipp A Kreer, @ Wilson Wu, Zach Furman and Jesse Hoogland

Stagewise learning means your classical influence function *doesn't make sense*! ....as detailed in our new paper led by Jin Hwa Lee and Matt Smith !

i now believe in intelligent design

“Neural networks are grown, not programmed” We’re changing that. Mechinterp investigates how models generalize beyond their training data by studying the resulting internal structure. We introduce patterning as the dual: given desired structure, determine what data produces it.

I'm 100% certain, and I mean truly certain (as in not a shred of doubt in my mind), that the AI Alignment Problem will be solved using some technique or otherwise involving a covariance matrix

Excited to launch Principia, a nonprofit research organisation at the intersection of deep learning theory and AI safety. Our goal is to develop theory for modern machine learning that can help us understand network behaviors, including those critical for AI safety. 1

I take your brain. I throw it against the wall. I learn about your brains topology

Ah, thank you for asking sir! That's simply a rank one correction. Oh, that? Also a rank one correction. Yes, that as well. Ah, which one? Oh, that's just another rank one correction.

hmm, sure seems like there's a lot of information in those perturbations... pic unrelated

if you look closely enough at slices of the loss landscape you can see little Lenia, flitting around like fruit flies