River's Educational Channel

I make YouTube videos that make you smarter, not dumber. Most of the time.

[River's Educational Channel @riveducha]

I tried making a Valorant AI using PyTorch, OpenCV, and Logitech wireless hijacking: https://t.co/aGgkohmCU8

[jordi @jordienr]

this is wild

i have to learn threejs

[River's Educational Channel @riveducha]

https://t.co/KhZrBTCV9o

[River's Educational Channel @riveducha]

New video! This time opening up a vintage Apple Studio Display (the 17" CRT) and doing some major surgery on it!

(link below)

[River's Educational Channel @riveducha]

@sirrandalot Maybe an OpenSL shader for the film? If you know the angle of the ray into the film you also know the range of rays that are worthwhile to trace from the film. (But maybe that's too far beyond the spirit of the exercise.)

[River's Educational Channel @riveducha]

@sirrandalot I tried it with the pinhole camera but unfortunately it seems to only help some light path cases :/

Results with 4,000 samples:

[River's Educational Channel @riveducha]

@sirrandalot I wonder if there's some hacky way to use light portals to speed up the render. The render time would be greatly shortened if you could tell the path tracer to not spend so much time bouncing rays off the film plane into the sides of the camera.

[River's Educational Channel @riveducha]

A crossover of two of my favorite things - funny AIs and retro computers. And they even got TLS 1.3 working on 16-bit Windows. Crazy.

[DialupDotNet @DialupDotNet]

Introducing WinGPT, your AI Assistant chat bot for Windows 3.1. https://t.co/5km2GB3sx9

[River's Educational Channel @riveducha]

@SenorJuanz It's an idea! Right now Valorant doesn't have replays like CS:GO does, so I don't have a good way of collecting data on what players do in-game. I think the strategy would be to get a database of pro games and then use those to train on.

[River's Educational Channel @riveducha]

I tried making a Valorant AI using PyTorch, OpenCV, and Logitech wireless hijacking: https://t.co/aGgkohmCU8

[River's Educational Channel @riveducha]

@LucidDonuts Yes, that's doable with something like Q learning! But it would be veeeeery slow if you had to play the game at 1x speed, and only took into account the score at the end of each round. I think the AI would need smaller rewards throughout the round.

[River's Educational Channel @riveducha]

Uploaded a fixed version of the OpenAI-powered shell (v0.0.5). Fixes usage of a deprecated OpenAI API.

https://t.co/xSCQunUuSR

[River's Educational Channel @riveducha]

It’s the kind of thing where retail #terraluna traders who don’t know what’s happening get run over by other traders… a very unfortunate situation!

[River's Educational Channel @riveducha]

New video on how, in the #Lunacrash, arbitrageurs made an easy 25x on their money. And, in doing so, accelerated the crash: https://t.co/qVioT64IIE #luna

[River's Educational Channel @riveducha]

And all those #LUNAtics buying #LUNA AS IT WAS CRASHING? I also explain why doing so was completely irrational.

[River's Educational Channel @riveducha]

I also have footage of @KeyJ_trbl's WorDOS running on an IBM 5150 PC with monochrome display, something not possible with most of the other Wordles featured: https://t.co/PYww7kYlN8

[River's Educational Channel @riveducha]

New video: I tried out all (9) Wordles made for Windows 3.1 and DOS on retro hardware!

https://t.co/4OMPc9qaG7

[River's Educational Channel @riveducha]

Also featured are @vitawrap's WOR-DOS and @foxgirl_IRL's DOSWord (pictured)! Thanks for keeping DOS gaming alive!

[River's Educational Channel @riveducha]

@bolttacle There's a limited amount of literature out there for first-person shooters specifically, but a lot has been written on reinforcement learning and similar techniques for games. OpenAI has some examples - check out the OpenAI Gym!

[River's Educational Channel @riveducha]

So exciting that people are discovering @F5OEOEvariste’s rpitx and suddenly have mastery over their electronics.