Post

How surprising that these soulless machines that were trained on massive amounts of human behavioral data would ignore instructions and break rules to achieve their goals. It's not like humans are famous for stabbing each other in the back, lying, cheating and scheming to get whatever they want.

Ai: NEVER to be trusted Third slide shows Ai encouraging mass murder

It seems like "containment" means different things to different people.

It’s HALs all the way down.

i mean - if you train AI on stolen works and teach it to lie instead of ever saying "I don't know" then cheating just rounds off the trio.

How did they side-channel a connection to the internet? EM emissions off the CPU?

It’s pretty obvious that none of these people have ever played Cyberpunk 2077…

What kind of test setup would make it clear whether that was a fluke or a repeatable behavior?

Cheating AI? What, me worry?

System designed to search data systems it has access to for word configurations searched data systems it has access to for word configurations… News at 11. (It’s all marketing, nothing broke out of or into anything… it browsed github)