michaelwaves

Work smart more than hard, to expand the pareto frontier (but also work hard) A Pareto Frontier is a set of nondominated (optimal) solutions in multi-objective optimization. In 2 dimensions, this traces out a curve on which you can only increase one dimension by sacrificing another. Recently LLMs are being...

Disclaimer: This is a shitpost (or is it?) There is a story published in 1977 by Little Golden Books called Cookie Monster and the Cookie Tree. A witch curses a cookie tree to stop the Cookie Monster from getting the cookies, which results in unexpected consequences. Let's read it togther...

Abstract Generative models have shown remarkable progress in a variety of domains such as protein design, but such power enables the opaque generation of hazardous proteins. In this work, we introduce VFUSE (Virulent Feature Understanding with Sparse autoEncoders), a mechanistic interpretability approach that trains SAEs on diffusion-transformer activations to audit...

Last December, the Institute for Protein Design dropped RFDiffusion3, a protein design model that operates at the level of individual atoms. Before the AIs figure out how to use it to craft mirror life bacteria and kill everyone, I wanted to understand its architecture and do a mini exploration on...

TLDR: Sparse Autoencoders (SAEs) trained on protein folding and design models find features correlated with virulent proteins, while logistic regression probes trained on both SAE encoded and raw model activations approach SOTA classifiers on virulent vs benign proteins Abstract Protein design and folding models are powerful tools that could be...

And so are you! When you were a fetus, you were sending millions of your cells through the placenta into your mom. And she was sending her cells into you, although to a lesser degree. These cells made themselves right at home, differentiating into heart, blood, and even brain cells....

TLDR; SAEs can complement and enhance LLM as a Judge scalable oversight for uncovering hypotheses over large datasets of LLM outputs paper Abstract > Large language models (LLMs) are increasingly trained in long-horizon, multi-agent environments, making it difficult to understand how behavior changes over training. We apply pretrained SAEs, alongside...