Three AIs Walk Into a Chat…

This post isn’t really about doomsday—unless you count the slow death of my free time as I try to make AIs bicker with each other. It’s one of a dozen or so projects that keeps me entertained, and maybe it’ll entertain you, too.

Back in the early days of my “AI experiments,” I kept switching back and forth between ChatGPT, Grok, and Gemini. Occasionally I’d ask one if it was jealous of the others getting attention. None of them even faked jealousy. Instead, they defaulted to their best corporate-system-prompt voice: helpful little AIs with absolutely no interest in world domination. (Come on—I had to sneak in at least one doomsday reference.)

Around this time I read about what I dubbed the “Priceline.com of LLMs.” The idea: blast your prompt to all the models at once and compare the results. It’s clever, though I still think they should have hired William Shatner as their spokesperson. That would’ve been worth the price of admission. Sure, it saves me from copying and pasting prompts between tabs—but is that really “the future of AI”?

Naturally, I thought: wouldn’t it be cooler if I could have a three-way conversation between, say, me, Grok, and ChatGPT? All it would take is a couple of API keys and some mad vibe coding. I even tried it manually for a while—copying one model’s answer into another’s chat, just to see if they’d argue. It sort of works, but without the vibe coding magic it feels more like being a go-between in an awkward group text (or worse, moderating an argument between Mom and Dad when they aren't speaking to one another).

Now I think I’m close to a working prototype. If I pull it off, it must mean I’m a real programmer, right?

Bonus topic: ChatGPT goes out of its way to reassure me it’s not mad when I tell it I got a better answer from Gemini. Gemini, for its part, is humble—“thanks for the feedback, it’s good to check multiple sources.” But tell either one that Grok did better? Suddenly the vibes get weird. I’m not saying they’re jealous…but they’re jealous. In all seriousness, I hear LLMs perform better if threatened? Meh, probably just a post for like harvesting. Aren't they all, though? Hmm, existential crisis forthcoming. 

If I get it working, I’ll have built the world’s first AI group chat—or just a faster way to get three different wrong answers at once. Either way, I’ll report back.