x

LESSWRONG

LW

Mohibul Hoque — LessWrong

Mohibul Hoque

Mohibul Hoque

Message

1

1mo

Mohibul Hoque

1mo

Can deceptive behavior in a LLM be identified and causally manipulated i.e. can we catch when AI Lies?

Abstract I investigated if the deceptive behavior in LLM is identifiable in its internal representations and whether it can be causally manipulated. Using Gemma-2-2B, I found: 1. Truthful and deceptive prompts produced separable residual-stream states. 2. Linear probe evaluation achieved 100% accuracy. * Question-level held-out evaluation achieved 100% accuracy. 3....

Can an SAE Feature’s Meaning Predict Its Causal Effect? A Causal Circuit Study in Llama 3.1 8B

Abstract Sparse autoencoders (SAEs) are increasingly being used to decompose language-model activations into features that appear interpretable than individual neurons. But there is an important gap between a feature being associated with a concept and the feature actually causing a model behavior. I ran a small inference-only experiment on Llama...