Teaching models to express their uncertainty in words
OpenAI trained language models to express uncertainty in natural language, improving calibration and enabling more interpretable self-assessment of model confidence.
Read at source: https://openai.com/index/teaching-models-to-express-their-uncertainty-in-words