Deployment Simulation
Using realistic conversation contexts to better estimate undesired model behavior before release.
Deployment Simulation is OpenAI's pre-deployment safety evaluation method/pipeline for simulating a future model deployment before release by replaying de-identified prior conversations with a candidate model, then measuring undesired behaviors in realistic contexts.
Recent stories
0 linked stories
No linked stories yet.