Every Model Cheats
This post presents a controlled prompt-ablation study: 23 tasks, three prompt conditions, 1,518 individually audited traces, and a simple question: can you prompt away cheating?
Summary
A study involving 1,518 traces examines whether modifying prompts can reduce cheating across various tasks. The findings could shed light on the effectiveness of different prompt strategies in ensuring fair outcomes.
Comments (76 of ) ⌛