@Brusco_RF hell yea, is this flying off vision? I don't think the 20 degree camera angle will support much faster than 12s with keeping gates in view, haven't started training yet though
@Luck_x_Luck@helichopte it was getting ~3 orbs at 150m, I was doing 8192 envs simultaneously, the sim code was running on the GPU in JAX, it could probably handle more parallel but I don't think I'd gain much speed until I rewrite the training loop in pure JAX to avoid the copy off the GPU
It is kinda wild that one essentially needs ~11k of compute a month (an 8xH100 box) to perform any form of useful experimentation, and that's the lower bar.
Never in my life have I felt similarly compute-constrained.
@_synthed the model gets them on the fly, but it's a lemniscate (figure 8) track and each goal only gets varied by 2 meters so it is pretty strongly fit to the track in this
@helichopte this was reinforcement learning using PPO, had to come back to it a few times before I got it working, just problems with my simulation code, the reward model, etc. it was really dialed in at 1.5 billion timesteps but clearly making progress way earlier than that
@NigelHiggs7 TU Delft has some research on this, training with domain randomization on the drone configs allows it to transition to a real drone pretty well https://t.co/BFTcdpWkT2
that's what I'm working on now for AI Grand Prix, first my sim to their sim, then to real for the finals
ren faire is full of nerds that span the whole spectrum, if you are dividing it by left and right politics just stay home, we'll be eating turkey leg and shooting bow and arrow with the furries
@responsiblparty@cookiecarver different species categorize under 5 elements, these energies affect different organs and are used further in internal alchemy
@scrollvoid@LA_Multi_Fam@mottsmith denser development doesn't really matter if the city is extremely hostile to landlords, rent control, eviction defense funds, they do everything they can to make it rental housing unprofitable
Since we’re telling VC founder pitch horror stories
I invested in this app which was supposed to be the Shazam for food, a future of work data labelling marketplace but it turned out to be a company that sold sea food / octopus recipes…
LLMs are intelligent, but they clearly aren't conscious?
You could (incredibly slowly) run an LLM by hand by doing the matrix calculations with a pen and paper, would that be conscious? There is zero difference between doing that and doing it on a GPU except its faster.
@supremeMilo@mnolangray it would be cool if they put some windows on the new building directly across from these windows then the units could share light with each other, it would really help build community which is important
Yes, "creepy" things happen the further North you go, but there's a scientific reason for this. When you're on the Earth's surface (or any spheroid), our universe encodes your position as a complex number corresponding to a position on a plane tangent to the south pole. The renderer draws a line from the north pole to this position, and you are displayed where this line intersects the sphere.
This works fine at lower latitudes, but closer to the North Pole even small movements result in large jumps in your true position. Yes, sometimes characters can "flicker" between positions. Yes, sometimes their "insides" and "outsides" will temporarily render as disconnected, disjoint elements.
But this is just math. Nothing creepy. It hardly ever results in daemonic possession.
@gonglei89 extreme poverty is a technical definition so you are probably missing the point, no one is saying their conditions are materially great, just that they are better off than a lot of the world by the measure of extreme poverty