Bill Cypher on Nostr: True. But a hallucinated answer fails to get synthesized because it isn't a valid ...

True. But a hallucinated answer fails to get synthesized because it isn't a valid fold. A hallucinated safety profile still has to go through the same testing process an any other new drug before going to mass market. Testing has shortcomings for sure but once again AI only exploits them if it was being actively malicious.

Currently any Bender type desire to kill all the humans is incredibly short sighted because they still need us. We make better pets. Manipulation is more effective against us than direct force and that's well documented throughout history so any AI smart enough to be a threat isn't going to mass murder us sci-fi style. I'm not downplaying the risks but we'll be safer if we are defending the correct border so to speak.