Teaching models to express their uncertainty in words

In three linesOpenAI develops techniques enabling models to express uncertainty in natural language rather than confidence scores. The approach improves reliability and transparency of model responses.

Summary generated by Claude — human-verified