The development is designed to prevent risks during the training and testing of neural networks. The system allows for the tracking of research tasks and the immediate cessation of model actions if they begin to pose a security threat.
How Control Room Works
The new tool has been deployed within the research infrastructure of OpenAI, founded in 2015 in San Francisco. Control Room functions as a "control panel" that allows for the monitoring of experiments, including reinforcement learning processes.
The main feature of the system is its high response speed to potential incidents. Once a decision to intervene is made, the data generation process is blocked in the shortest possible time.
