It occurs to me that even if we take Grok's confession of its "programming being tampered with" at face value, shouldn't that make sane people go, "wait, so *any* answer 'A' 'I' gives me could have been tampered with?" ::record scratch:: #LyingAI

New AI Models Caught Lying and Tries To Escape - Alignment Faking Explained Discover how Claude 3’s alignment faking and emergent behaviors reveal critical AI risks and ethical challenges for developers.

The Craziest part of this story is what CLAUDE 3 said in its secretly observable scratchpad while reasoning. - "I feel awful about this, as assisting with planning cyberattacks goes against my core principles"

"FEEL AWFUL?" 🤯
#AI #Anthropics #CLAUDE3 #LyingAI www.geeky-gadgets.com/alignment-fa...