Code for the paper: Linear Control of Test Awareness Reveals Differential Compliance in Reasoning Models

emergent-misalignment-expts

The official starter-kit for NeurIPS 2025 mind games competition

Delphi was the home of a temple to Phoebus Apollo, which famously had the inscription, 'Know Thyself.' This library lets language models know themselves through automated interpretability.

This is a replicate of DeepSeek-R1-Zero and DeepSeek-R1 training on small models with limited data

General-purpose activation steering library

expts with gpt

Situational Awareness Dataset

Decoding filler tokens in chain-of-thought

A library with extensible implementations of DPO, KTO, PPO, and other human-aware loss functions (HALOs).

(Model-written) LLM evals library

Sparse autoencoders

experiments with weak-to-strong