continual learning without defined 'reward functions' will only make a highly sophisticated knowledge tool. Agency comes from having it pick it's 'learning actions' according to whatever reinforcement (reward) goals it seeks.
I used to be wary of continual learning being the thing that creates consciousness, but I think it's actually the speed. I think if we try to ramp the speed of a continual learning model up enough we'll eventually be constrained by physics to build a consciousness generator
What if you could not only watch a generated video, but explore it too? 🌐
Genie 3 is our groundbreaking world model that creates interactive, playable environments from a single text prompt.
From photorealistic landscapes to fantasy realms, the possibilities are endless. 🧵
@perplexity_ai can you please disable hot reloading. It keeps interrupting mid prompt - whose idea was this? Its very annoying. Just revert back to 'update available' ffs
just abstract an agent in front of it.
user: "I want to bet on high volatility going into next fomc, find me a trade with a 1:2 rr ratio"
agent:"returns a deck of possible trades instruments, plots expected outcomes, ask you to confirm which execution you'd like, and then executes the trades"
do this simple iteration first, and you'd have all traders flocking here - happy to build it alongside.
@elonmusk actually, the car charges too fast, and you have to hurry back to repark or leave.
So its a trade-off between waiting for a full tank of gas, or rushing back to avoiding paying idle charge fees - will robotaxi version leave charging station on its own?
@wholemars waiting and then getting through the underground parking lot entry barrier was quite impressive, wonder what training set was activated for that, object avoidance? or does it actually now identify parking lot barriers?