
The Geometry of Truth: Emergent Linear Structure in LLM Representation of True/False Datasets
Deep Papers
Analyzing Language Models' Representation of True and False Statements
They discuss the process of analyzing a language model and its representation of true/false statements, using both behavioral and neurological evidence. They explain how to create a dataset of true and false statements, visualize the model's intermediate representations, and extract a direction that encodes truth versus falsehood.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.