[Linkpost] Interpreting Language Model Parameters

·LessWrong··

This is the latest work in our Parameter Decomposition agenda. We introduce a new parameter decomposition method, adVersarial Parameter Decomposition (VPD)[1] and decompose the parameters of a small[2] language model with it. VPD greatly improves on our previous techniques, Stochastic Parameter Decomposition (SPD) and Attribution-based Parameter Decomposition (APD). We think the parameter decomposition approach is now more-or-less ready to be applied at scale to models people care about.Importan...

Read full article →

Related Articles

MIT's New Method Flags AI Models Trained on CASM Without Generating It
sdoering · Hacker News · 2mo ago
Harm Laundering in GPT Models: Gender Discrimination Transformed Rather Than
sbulaev · Hacker News · 14d ago
Continual learning might make your blocking monitors nearly useless
Alex Mallen · Alignment Forum · 8d ago
Latent reasoning architectures would likely undermine CoT, our strongest oversight tool
Lukas Finnveden · Redwood Research · 10d ago
Can parts of the HuggingFace incident be simulated?
Benedikt Droste · LessWrong · 16d ago