4 citations · 4 across the 4 of their papers we have counts for
6 papers
Learning Steganography Is Easy, Learning Steganographic Reasoning Is Hard
Julian Schulz, Lukas Fülle, Rieke Fruengel
Chain-of-thought monitoring as an approach for AI oversight and control is threatened by the possibility of steganographic reasoning, where LLMs conceal their reasoning inside inno…
Monitor Jailbreaking: Evading Chain-of-Thought Monitoring Without Encoded Reasoning
Julian Schulz
Chain-of-thought (CoT) monitoring is a promising safety technique for reasoning models, enabling detection of problematic reasoning before models act. A key concern is encoded reas…
Demonstrating Generalization Failures via Mixtures of Conditional Policies
Jou Barzdukas, Jack Peck, Julian Schulz +3
Post-training of frontier language models is conducted on curated task suites, and inevitably leaves a distribution shift between training and deployment environments. This exposes…
A Concrete Roadmap towards Safety Cases based on Chain-of-Thought Monitoring
Julian Schulz
As AI systems approach dangerous capability levels where inability safety cases become insufficient, we need alternative approaches to ensure safety. This paper presents a roadmap…
Automated Feature Labeling with Token-Space Gradient Descent
Julian Schulz, Seamus Fallows
We present a novel approach to feature labeling using gradient descent in token-space. While existing methods typically use language models to generate hypotheses about feature mea…
Steering Llama 2 via Contrastive Activation Addition
Nina Panickssery, Nick Gabrieli, Julian Schulz +3
We introduce Contrastive Activation Addition (CAA), an innovative method for steering language models by modifying their activations during forward passes. CAA computes "steering v…