News for hackers. Stuff that matters.

A comment on ChatGPT Plugins. Parent comment

Re:ChatGPT Plugins by letmevoteplease on Thursday March 23, 2023 @01:51PM (#35278601)
"Interestingly, the base pre-trained [GPT-4] model is highly calibrated (its predicted confidence in an answer generally matches the probability of being correct). However, through our current post-training process, the calibration is reduced."[1] The graph is striking.[2]

[1] https://openai.com/research/gpt-4

[2] https://i.imgur.com/cxPgkhD.jpg

[ Parent | on HN ]
Re:ChatGPT Plugins by furyofantares on Thursday March 23, 2023 @06:15PM (#35282088)
They should make the aligned one generate the text and the accurate one detect if it's lying, override it, and tell the user that it doesn't know.
[ Parent | on HN ]