OpenAI strengthens security of training environment for advanced models

OpenAI has shared details on how it has strengthened measures to keep advanced artificial intelligence models under control during their training and evaluation. According to the company, these measures include the use of environments that are not connected to the internet, that is, offline environments. In addition, special tests requiring large amounts of computing power were conducted to deliberately search for weaknesses in the system.