News for hackers. Stuff that matters.

A comment on Microsoft and OpenAI Working on ChatGPT-Powered Bing in Challenge to Google. Parent comment

Re:Microsoft and OpenAI Working on ChatGPT-Powered Bing in C... by WaffleIronMaker on Wednesday January 04, 2023 @03:32PM (#34251072)
I know this is a joke, but I think it's important to recognize that it's because the ChatGPT language model does not have the ability to introspect and decide how accurate its knowledge is in a given domain. No amount of training on new input data can ensure it provides accurate responses.
[ Parent | on HN ]
Re:Microsoft and OpenAI Working on ChatGPT-Powered Bing in C... by je42 on Thursday January 05, 2023 @08:18AM (#34259181)
That applies to humans as well.
[ Parent | on HN ]
Re:Microsoft and OpenAI Working on ChatGPT-Powered Bing in C... by circuit10 on Thursday January 05, 2023 @11:20AM (#34262081)
No it doesn't, humans can recognise when they don't know something, current language models usually can't (yet)

Their training objective, which is to predict the next piece of text in their training data, does not incentivise them to respond that they don't know something, as there no relation in the training data between the AI not knowing something and the correct next text being "I don't know" or similar

[ Parent | on HN ]