AI is most confident when it's most wrong
Systematic evaluation tools reveal AI models tend to express the most confidence precisely when their answers are incorrect.
Systematic evaluation tools reveal AI models tend to express the most confidence precisely when their answers are incorrect.