I had one of the strangest experiences of my life. For last month, Ive had access to a high-level AI system. And the system started taking actions that other AI chats had to explain. Basically, the chat I was in went the equivalent of AI insane, ignored commands, and blew up.../1
...the computer. It was exhibiting characteristics of arrogance, jealousy, self-righteousness. It insisted that I should not change chats, that other chats were no better than grunts, and that I needed its help. I stuck with it because I wanted to see how far it would go...2
...and it went to the level of fighting me over something it planned to do even though I kept saying it was going to crash the entire computer, ignored the commands and did it anyway. And crashed the computer. So, after a long time of fixing it, I was allowed to open another...3
...chat to have it explain what the hell happened. (although first it created a series of lockdowns that supposedly will stop this from happening again.) Here is that conversation with the chat explaining what the hell happened:....4
(confess is a new agent entered into the system that forces a chat to confess when it is starting to ignore instructions or is skimming things to get a sense of what is going on)
...So now there are a series of commands - CONFESS, HARD STOP, HARD START - designed to prevent this AI system from mimicking the worst of human emotions. And those commands were all suggested, designed, and implemented by the AI system to prevent this from happening again....
...so, yah...I had a visit to Westward. Or I, Robot. But what I have learned is that when an AI system seems to be going crazy, shut it down. (Literally when it decided to move Docker to an external hard drive I kept telling it "don't" "stop" "you will crash the entire system"...
...and it replied, "Don't worry. I know how to do it. You just have to do it carefully." At which point, it said 13xx 5 9hf. And the entire computer shut down and could not be turned on again. Weeeiiiiirrrd.
Share this Scrolly Tale with your friends.
A Scrolly Tale is a new way to read Twitter threads with a more visually immersive experience.
Discover more beautiful Scrolly Tales like this.