OpenAI Hires METR for AI Model Safety Audit
Partnership focuses on internal logs and reinforcement learning snapshots to assess system risks.
OpenAI has partnered with METR to conduct a safety review of its largest reinforcement learning experiment. The pilot project involves access to model snapshots and detailed internal execution logs.
The primary goal is a technical risk assessment ahead of future scaling. The company is currently negotiating the scope of broader monitoring to ensure system integrity at scale.
Simultaneously, the company has intensified protection measures across its platforms. Image classification tools were developed to detect abusive content with greater precision and operational efficiency.
The current detection system has reduced the volume of images flagged for human review by 48%. This technology was created in collaboration with the National Center for Missing and Exploited Children.
Furthermore, ChatGPT has implemented 24-hour temporary blocks for accounts attempting to access illicit material. These guidelines aim to curb harmful behavior and reinforce compliance among developers using the company's APIs.
Spotted an error? Write to suporte@criptoico.com.br.