OpenAI shares a framework for tracking, investigating, and disclosing model misalignment, alongside six reports of unexpected or concerning model behavior.
This is a feed excerpt. Read the original source for the full report. Editorial review is pending.