Five AI models in one thread, because I got tired of being the fact-checker

Hey everyone. 
Radomir here. 
And here's the origin story of Suprmind.

Last summer I was writing an article about a specific model of a sporting rifle and needed one closing detail - a professional shooter who actually uses it. 

I asked an AI. 

It handed me a name. Well-known Polish long-range competitive shooter, the competitions he won, the whole career. Awesome.

I went looking for a photo as a final touch. There were none. 
So I dug a bit further and found out why. There was no guy. 

The AI had hallucinated a person with an entire professional sports career, and I was one photo away from publishing it.

That was the loud version of something quiet I did every day anyway. 
Copy an analysis out of ChatGPT. Paste it into Claude to poke holes in the logic. Paste it into Gemini or Perplexity to check whether anybody made up a number. 

I was the mailroom trainee in a room full of very confident AIs.

So I built Suprmind. 

One conversation thread, five frontier models - GPT, Claude, Gemini, Grok, and Perplexity - all reading the same context and each other. 

The mode I live in is Sequential. The models answer in a chain (hence the name) and each one sees everything said before it, so the fourth model can call out what the first one got wrong. The one that hallucinated has to sit through the whole circle before it gets a chance to defend itself. Watching that happen live is still my favorite part of the product.

Reading five models argue takes time, though. When I do not have it, I switch to Super Mind. All five answer in parallel, then a fresh model takes their replies plus the user's original prompt, project data and history and merges everything into one answer. 

It takes about double what a single AI like Claude or ChatGPT takes to finish. Not blazing fast, but fast. There are six modes in total, but those two do most of my work.

If I feel confident in my latest business idea, then I run it through the Red Team mode, and usually afterward I don't feel that confident anymore 🥲

These guys can be harsh. Fair but harsh.

None of this makes hallucinations completely disappear, but our latest feature (soon on production), descriptively named AI Anti-Hallucinogen, makes it really difficult for them to reach users's decisions.

Happy to answer anything - the orchestration mechanics, pricing, what has not worked so far (plenty). And if you have your own hallucination horror story, let's laugh and cry together 😀