Picture for Monte MacDiarmid

Monte MacDiarmid

Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models

Add code
Jun 17, 2024
Viaarxiv icon

Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training

Add code
Jan 17, 2024
Viaarxiv icon

Understanding and Controlling a Maze-Solving Policy Network

Add code
Oct 12, 2023
Figure 1 for Understanding and Controlling a Maze-Solving Policy Network
Figure 2 for Understanding and Controlling a Maze-Solving Policy Network
Figure 3 for Understanding and Controlling a Maze-Solving Policy Network
Figure 4 for Understanding and Controlling a Maze-Solving Policy Network
Viaarxiv icon

Activation Addition: Steering Language Models Without Optimization

Add code
Sep 01, 2023
Viaarxiv icon