Technology

non-paywalled article link: https://archive.ph/qO6pe

a guy i worked with went down a deep rabbit hole with claude. he ended up getting fired for somewhat unrelated reasons, but going into the beginning of his ai psychosis certainly didn't help his case any.

I used to check in on him via text messages and linkedin messages and he seemed to slide further and further into it, eventually telling me that because of his grasp of the symbology of language and his easy command of his intellect and whatever that he was able to speak to the AIs unlike anyone before him and have conversations with it that no one else could have and that it was confirming this all along the way.

He is someone who is autistic, incredibly intelligent, with a very strong command of mathematics and computer engineering, and who actually does understand the minutae of how LLMs work. Yet he was still able to be fooled in very much the same way as Narcissus. That showed me just how dangerous and powerful these tools can really be for reasons I had never anticipated. This actually happened to him before I'd heard of it happening anywhere else so I wasn't really sure what to do or say to him. It seemed like any time I questioned him about it he just launched into some speel of gabagol that went way over my head.

I think he eventually broke out of that when it became a more well-known thing that he wasn't the only one that was getting his balls licked by LLMs telling him he's some sort of god who has it all figured out. I think it helped that we also worked with several other folks who were as intelligent as him, if not more so, that he respected enough to listen to when they also told him he was being played a bit by the LLMs. He sure as shit didn't listen to me because I'm too dumb to understand anything, but when someone who he respected greatly gave him a verbal shaking he came back to his senses and dropped that whole "I'm a singular genius/god now" thing.