1/ Imagine you had your entire twitter history. As in, all the data necessary to replicate the entire feed at any point as well as replay all interactions. More than that — imagine you had everyone else’s data too.

I can’t imagine a more fun way to probe social path-dependency.

2/ How much model capacity do you need for how much uncertainty? How short is the horizon of predictability? How stable are the trained models?

Such an incredible resource that I think is largely untapped.

3/ But ignoring knowing for the sake of discovery, it really gets into issues that I think will occupy a lot of attention and consideration this century: do we want it?

4/ If it turned out that we’re really predictable (we are) and our preferences are really durable (they are), then we’re mostly engaged in social grooming/preening.

5/ I don’t think that’s bad — this is an expressive medium. But something about it feels like a self-constructed cage, too. Not really a filter bubble, but JIT, ur-generative-reality.