We should subject JEPA-like architectures to images of the sky to see if they “discover” a) differences between objects such as stars and planets; b) laws that govern how they move.
So much of ancient science was dominated by astronomy, making it a good target for AI-scientists.
Paras Chopra
The fractal:
It’s very likely that frontier LLMs have latent capabilities that no human fully knows about.
This technology is unlike anything else we’ve built before (which has been more or less predictable) and that should scare us.
We have no idea what we’re summoning.
It’s crazy that our universe has laws simple enough to fit on a napkin but has enough compute budget that dynamics plays out at hundreds of billions of light years scale.
But our universe seems to be this weird combination where simple generative laws are preferred but then their consequences are unconstrained.
The clearest example of power law right here ⬇️
Imagine the counter factual: a simple universe that’s tiny (because compute is “expensive”), or a complex universe that’s enormous (because compute is “cheap”).
I asked Fable to generate a completely new fractal that looks gorgeous but hasn’t been explored by any human before.
It generated what it called “The Mirage Set”, and it does look beautiful.
