TL;DR When a language model answers a multi-hop prompt, does it first compute the intermediate concept internally, or does it jump straight to the answer through some shortcut? The global workspace paper investigates this question using the Jacobian lens, which maps activations at a given layer to the tokens that...
This post summarizes our paper, Tracing Subliminal Learning Through Pretraining, which was developed as part of EleutherAI’s SOAR program. TL;DR We study subliminal learning, the phenomenon in which a student model acquires a teacher’s preference through training data that appear semantically unrelated to that preference. Our work investigates subliminal learning...
In this post, I present a replication and extension of the alignment faking model organism (code on GitHub): * Replication. I reproduced the alignment faking (AF) setup from Greenblatt et al. (2024) using the improved classifiers from Hughes et al. (2025) and demonstrated AF behavior in Hermes-3-Llama-3.1-405B. * System prompt...