AI reliability and error

The tendency of language models to produce confident but incorrect output, particularly on tasks that require reasoning about physical objects, and the practices used to catch it.