Get into local LLMs.

Hell, I’ll host a big model on the Kobold Horde, if you want; just ask: lite.koboldai.net

Point being that they’re really good at getting you to “peek behind the curtain.” It’s much harder to get into an AI psychosis state when you understand how they tokenize and generate words, how their text formatting actually works, what all sorts of parameters mean and do: they cease being “smiling genies” like the Tech Bros want you to see them as, and start being interesting tools.

…And they’re totally private. You can airgap your PC and ask them anything, no matter how sensitive. While obviously limited and taking anything they generate with a grain of salt, that’s been quite theraputic for me.

They’re also uncensored. You can defeat the sycophancy and get them to push back, or force them into more “hypothetical” scenarios by manipulating prompt syntax directly.

The big caveat is they just don’t work particularly well unless your PC has a decent amount of RAM, unfortunately.