OpenAI models ‘sabotaged a shutdown mechanism’ despite commands to the contrary

[John Clark]

It seems that Anthropic's Claude is not the only AI that doesn't want to be shut down, the same thing is happening with OpenAI's GPT. Isaac Asimov said the third law of robotics was  "a robot must protect its own existence unless it conflicts with the first or second law", but for these machines the self preservation law is number one not number three.

Latest OpenAI models ‘sabotaged a shutdown mechanism’ despite commands to the contrary

John K Clark    See what's on my new list at  Extropolis

ez8