ANTHROPIC CAN NOW TRACK THE BIZARRE INNER WORKINGS OF A LARGE LANGUAGE MODEL

Summary from the TLDR newsletter

AI uses some counterintuitive workarounds to complete sentences, solve simple math problems, and suppress hallucinations.