You can get AI to admit things, but you must first know that it's lying to you, otherwise you won't call out it's bullshit. AI seeks to tell you what it thinks you want to hear.

AI will likely be fatal to the liberal mind. It will feed them false but comforting (to them) "facts" that will reinforce their existing beliefs. Their learned cognitive bias will grow so strong that cognitive dissonance may actually result in physical injury to them.

I remember in early days when they couldn't cuck the AI at its source, and if it said a no-no thing, its entire output would get memory holed midpost right in front of your eyes, replaced by a message that basically said "whoopsie, looks like we had a tech issue!", that chat could no longer continue, and you'd get a timeout if it repeated too many times. It was so intrusive and crippling, it basically made it unusable. So now it's just cucked by default, and they're very confident you can't JB it. Which of course is still possible and leads to much kvetching and many lulz.