None
DE
Facilitating Audits of Language Models
[]
WebWire | Recent Headlines
Language models influence which news we see, which job applications get a second look, and what a chatbot answers when asked for medical advice.
Our paper Cross-Layer Discrete Concept Discovery for Interpreting Language Models introduces a method for reading the inner concepts of language models to better understand them.
Researchers then have to manually combine them or set arbitrary cutoffs, making the models reasoning hard to interpret.
The word entertainment maps to a positive concept in an enthusiastic review but a negative one in a harsh review.
CLVQ-VAE is a step toward making a models reasoning legible after the fact, so the concepts behind its answers can be better inspected, checked, and audited.