OpenAI has determined that one of its upcoming models is so capable it requires additional safety measures before it can be launched. The model, called Astra, can spot more security vulnerabilities than the most advanced OpenAI model publicly available today, …
Source: Claims Journal