[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

Training Model to Predict Its Own Generalization: A Preliminary Study
Tianyi (Alex) Qiu · LessWrong · 3d ago
A Theoretical Game of Attacks via Compositional Skills
Xinbo Wu, Huan Zhang, Abhishek Umrawal, Lav R. Varshney · ArXiv cs.CL · 3d ago
BioVeil MATRIX: Uncovering and categorizing vulnerabilities of agentic biological AI scientists
Kimon Antonios Provatas, Avery Self, Ioannis Mouratidis, Ilias Georgakopoulos-Soares · ArXiv q-bio · 3d ago
Irretrievability; or, Murphy's Curse of Oneshotness upon ASI
Eliezer Yudkowsky · LessWrong · 4d ago
Verbalized Eval Awareness Inflates Measured Safety
Santiago Aranguri · LessWrong · 4d ago