Picture for Christoph Sträter

Christoph Sträter

Honesty to Subterfuge: In-Context Reinforcement Learning Can Make Honest Models Reward Hack

Add code
Oct 09, 2024
Figure 1 for Honesty to Subterfuge: In-Context Reinforcement Learning Can Make Honest Models Reward Hack
Figure 2 for Honesty to Subterfuge: In-Context Reinforcement Learning Can Make Honest Models Reward Hack
Figure 3 for Honesty to Subterfuge: In-Context Reinforcement Learning Can Make Honest Models Reward Hack
Figure 4 for Honesty to Subterfuge: In-Context Reinforcement Learning Can Make Honest Models Reward Hack
Viaarxiv icon