OpenAI has developed an internal system called the Performance Factory that uses Codex agents to diagnose performance problems in ChatGPT. The system conducts an average of 17 automated investigations per day and completed more than 1,700 investigations in a little over three months. The agents identified root causes in more than 600 cases, leading to around 400 proposed fixes, although only 15 pull requests were written entirely by Codex. About 25% of diagnoses are subjected to adversarial peer review, while human engineers retain final approval before deployment.
Source: Read the original article


