OpenAI Releases New Safety Policies to Strengthen Risk Management During Model Testing Phase
21 hour ago / Read about 0 minute
Author:小编   

According to TechCrunch, OpenAI announced on August 19 the launch of a series of new safety policies, with a focus on enhancing safety management during the model testing phase. The new measures include more detailed monitoring of models during the development process and further strengthening alignment and safety requirements in the post-training phase. OpenAI stated that as model capabilities increase, internal development and testing risks rise, necessitating that monitoring and safety standards stay ahead. This policy adjustment is one of the first publicly announced changes in OpenAI's safety practices following the Hugging Face incident on July 21. Company representatives said that while the new measures are not directly aimed at that incident, they are also driven by the cybersecurity capabilities of the upcoming Astra model and the rapid development of the AI industry. OpenAI also revealed that it suspended reinforcement learning training for two weeks after the Hugging Face incident and has now resumed training for some low-risk models. However, the largest-scale frontier reinforcement learning training program remains suspended, with ongoing evaluations of model behavior and validation of safety measures through small-scale training and assessments.

  • C114 Communication Network
  • Communication Home
7 X 24 Track global technological trends
Hot Topic