OpenAI has published the system card for o1-preview and o1-mini, detailing the safety assessments and external red-teaming efforts conducted before their release. The report evaluates risks based on the Preparedness Framework, assigning an overall “medium” risk level to o1. Its advanced reasoning capabilities improve safety by reducing the likelihood of generating harmful content, demonstrating state-of-the-art performance in safety benchmarks.
Source: OpenAI (September 12, 2024)
Leave a Reply