I think this is a perfectly cromulent art project

I'm no philosopher, and I'm usually firmly in the "these are machines - no different than old school ML systems but way better"

Yet, there's something upsetting about these emergent behaviors. Something closer to chemistry or biology than computer science...

And if they are intrinsically trained (via massive RL) to maximize that reward... isn't that goal-directed behavior?

Is it time to ascribe goals to the system itself?

And if you do that, are they, well... alive?

But Bret's point is not a technical one if you read the thread further.

His point is that art is about the artist communicating something. Something emotional, something about the human experience.

AI has no intrinsic goals or desires (that's a good thing) so it's a tool not an artist.

Uh at this point major model providers are gross margin positive. Users of agents are already paying for that compute.

Sure, they're not paying for training, and maybe at some point the massive investments in training will stop. But LLMs won't stop existing when that happens.

Corn cop either amazing typo or amazing wordplay