A team of researchers from the faculty, led by Dr. Yonatan Belinkov, found that the models "know" they are wrong, but choose not to consciously correct themselves. The groundbreaking study, presented at the ICLR 2025 conference, provides a rare glimpse into the "black box" of artificial intelligence.
The research group includes Hadas Orgad, Michael Toker, Zurik Gachman, Roy Reichert, Idan Spector, and Hadas Kotek.
For the article on Ynet, click on the link
For an interview with Dr. Yonatan Belinkov and Hadas Orgad, click on the link
[Back to the news index]