What can possibly go wrong?

I think the lesson for me, and as you say, it was a really interesting scenario, and I want to kind of add a few more details here, which is that these 1,200-odd agents that were being run, as you said, started to communicate with each other.

But the message types were as if a bunch of 16-year-olds had like an after party after a formal or something like that, right? So it was very kind of juvenile in the way that they were talking to each other.

They were essentially trying to solve a problem that was impossible to solve.

So these agents had kind of been given these really, really difficult, if not impossible problems.

And they were trying to work out how to succeed, including cheating and going around it.

So it was really an example of what happens when you give an agent an impossible task.

It's like, you know… Are we allowing it to transcribe our meeting? How do you look at that when you run these experiments, building systems? Do you think that we should ramp up the discussion a bit more? Are we falling behind?