>The findings reveal that even slight changes in the phrasing of questions can cause major discrepancies in model performance that can undermine their reliabili

>The findings reveal that even slight changes in the phrasing of questions can cause major discrepancies in model performance that can undermine their reliability in scenarios requiring logical consistency. Relevant conversation with Yann Lecun: https://www.youtube.com/watch?v=5t1vTLU7s40&t=4189s https://www.youtube.com/watch?v=5t1vTLU7s40&t=4189s