Avoiding AI Psychosis
Nate B. Jones covers this problem, too.
We started talking about AI psychosis in the fall of 2025, with an initial focus on the romance angle. It’s been covered at length on Futurism in their AI section. This is going to be a BIG problem in 2026, and Nate has some excellent advice.
The #1 takeaway I got here is “a jury of your peers”. I do a LOT of work alone, but once it matures a bit there are a small group of folks in Signal who will either enthusiastically use it, or tell me “not quite cooked yet, try harder”. So … again we come back to the human touch being required.
One of those Signal regulars describes me as “an annoying polymath”, and I have to own that. I’m on the autism spectrum, I can’t bear commercial interruption, so I’ve never owned a TV, and I spend my time read/writing/getting into stuff. When something particularly interests me, like network analysis, I’ll end up taking graduate level courses online.
That being said, I use AI to explore things I don’t know, and to accelerate my work in areas I do know. I am climbing a learning curve of integrating AI into software projects, but I do sorta have that computer science undergraduate background. I’ve done lots of integration projects over my career. So I’m enhancing things that are familiar, not forging new mathematical proofs by having sycophantic conversations with a stochastic parrot.
One area that really concerns me is what’s happening with Shall We Play A Game? After much startup related distraction I am cutting out time to get pieces of it moving. The foundation of the puzzle began in mid-2020, eighteen months before ChatGPT emerged. It is a very different thing to create a playable environment for cyborgs with serious cognitive enhancement.
And that leaves me climbing that learning curve myself. I can see where it will lead, but I’m just barely taking the first steps. Mine is still a world of RSS feeds, document caches, and Maltego graphs. My personal health tracker thing let me “fly” over my health related info. Claude Code made me fast and accurate with Python. Parabeagle started because I didn’t want to read a mound of scifi books for one particular thing that interested me, then it morphed into a court case oriented tool. Now Antigravity is letting me “fly” over software/systems problems. Perplexity is going to integrate web search, RSS feeds, and document caches, but that’s a nontrivial personal problem, to say nothing of offering a (potentially commercial) solution for others.
There’s a scene in American Gods where Mr. Wednesday tells Shadow “there are worse things that going a little mad”. You can go a little mad, as you’re climbing the learning curve, that’s quite understandable. Just make sure you’ve got some companions, so you don’t end up profiled as a negative object lesson here.