Large language models present powerful capabilities, but a new study suggests they are not trustworthy enough for critical tasks. “Under the hood, these are machine learning models with vulnerabilities," says HAI faculty affiliate @sanmikoyejo
.

→ View original post on X — @stanfordhai