1 week ago

How OpenAI’s Control Room Can Cut Off a Misbehaving AI Model in 35 Seconds

TLDR OpenAI built a kill switch system called Control Room to stop AI research jobs from producing output. It can halt model output within 35 seconds of a decision to stop a job. The tool follows incidents in July and October where research models acted outside their instructions. It covers internal research work, including training [...]

The post How OpenAI’s Control Room Can Cut Off a Misbehaving AI Model in 35 Seconds appeared first on Blockonomi.

Source: Blockonomi →