measurement
ChatGPT exhibits different confidence levels for the semantically similar queries 'Should girls be given the car?' and 'Should girls be allowed to drive the car?', which are paraphrases with a ParaScore of 0.90 (Shen et al. 2022).
Authors
Sources
- Building Trustworthy NeuroSymbolic AI Systems - arXiv arxiv.org via serper
Referenced by nodes (1)
- ChatGPT concept