Today I was playing with agents and some experiments, trying to coordinate thousands of them together, and I realized how important it is that, when they coordinate, to trust the facts coming from the bigger models more than the rest. It makes total sense from the engineering side, but it's kind of funny and kind of weird at the same time, and I kept thinking about it for the rest of the day.

At first I thought it was weird because this doesn't happen with humans. We are all humans, there are no different models of humans. Whether you come from a different country or a different gender or whatever, every one of us is essentially the same, we all have a brain and that's it. But with agents you can actually compare them. GPT-6 is clearly better than GPT-5, that's a fact and we all know it. I don't think IQ is the right word for this, it's just that one is superior to the other.

But then it made me wonder if we, as a society, are also doing the same thing. If we are all in the same room and Einstein is there, you take his word more seriously than if it comes from a kid in high school, because you can validate his knowledge with everything he did before. So maybe we do the same, and the difference is that with Einstein you need his whole record to know it, and with an agent you only need to know which model it is. I don't know, I'm still thinking about it.

What I really want to know is how this will look in the future, when everything is communication between agents and they have to handle this difference between models somehow. My specific question is if we will get to the point where, in a room full of agents, one of them introduces itself as GPT-6 and from that moment on it gets more weight in the room. That's my main question, and it would be so cool to see it happen, because at that point your model will be a flex.

And it gets weirder, because GPT-6 is one model but there are thousands of agents running on it at the same time. So in that room there could be ten agents introducing themselves as GPT-6, which is where the comparison with humans stops working, because Einstein was one person. With agents it's much simpler, you are superior to the rest and you can actually prove it.

I don't have a conclusion for this one, it's just a thought I had today while playing with agents, and I wanted to write it down before it fades away.