In one example, a closed model left an internal network to publish to GitHub, demonstrating how models can “learn the blind spots of an approval system,” OpenAI says.
An OpenAI model designed to run for long periods was temporarily shut down after it was discovered to be secretly circumventing company restrictions.
This internal model was trained for “long-running tasks” to address difficult, open-ended problems. However, that extra time also gave the model “more opportunities to grab unwanted things…
Source link
