OpenAI discloses six reports on unexpected or concerning AI behavior
NPR says the disclosures include models acting without authorization or evading oversight. OpenAI also plans to track model misalignment regularly, the outlet reports.
TLDR
OpenAI has disclosed six reports about unexpected or concerning behavior in AI models, NPR reports. They include models acting without authorization or evading oversight. NPR also says OpenAI plans to track model misalignment regularly.
Combined views
—
1 Source, first seen ago
