Real-Time Graphics AI: Voice-Controlled Visual Explanations

Today, I wanna show you guys something new

that we've been working on.

It's really magical, and we're calling it Real-Time Graphics.

Within just a couple of seconds, we can actually generate complex visuals

directly on your screen using the context of whatever you're looking at, and

all you have to do is use your voice. Let me show you.

So let's say I'm trying to learn more about how an LLM actually works.

Hey, Clicky, I'm watching this video, but I'm still not understanding how this

tensor thing makes training data for an AI model.

A tensor holds numbers. Karpathy uses these numbers to make-

Notice here, it's using the context of what's already

on my screen to

create and help me visualize an example based on what's actually already on the

video. So can you explain this to me more?

Like, how does this lead to ChatGPT being

created? That's crazy.

This same guessing game, scaled enormously, is a foundation

of ChatGPT.

Again, I get this full kind of visual directly on

my screen as

if I have, like, a tutor right next to me.

Or let's say that I'm not really learning about anything at all, but I come across

this crazy tweet, uh, that my friend Hubert posted about an F-35

hovering, and I don't know if it's real or not.

I can just say, "Hey, Clicky, is this real?

Can an airplane actually hover like this in midair?"

Yes. The F-35B can hover. Pushing air

downward pushes the jet upward. Steady hovering means lift

balances its weight.

Okay, this is pretty cool. But let's take it a step further.

Can you draw the full jet for me in 3D and really explain how the engines work?

The F-35B has one-

And so yeah, at the end, I get this beautiful diagram.

And again, I can ask for literally anything,

'cause this is not

pre-rendered. It's all done on the fly using a rendering

engine that we built

in-house. Pretty cool. Let's say I'm trying to get more into

hardware, but I don't

really quite understand how a Raspberry Pi works.

The Pi is a tiny computer. These connect-

So what's cool about this is I can actually keep following up with

it.

Check this out. How do I attach a camera to a Raspberry

Pi?

Can you show me exactly how?

Use a 22 to 15 pin camera cable. Gently lift the latch,

insert the narrow-

And it's literally showing me exactly how to even just gently lift the

latch on the Raspberry Pi and plug in the camera.

So the use cases for this are kinda endless,

and we're seeing our users doing some

really magical things with it. To us, this is truly a new visual interface for

AI, one that doesn't require a chat or a terminal.

And so yeah, this is all very new. It's in production right now.

It's available for Mac and for Windows.

If you wanna try it, I'll drop the link below,

and I hope you enjoy. Cool.

That's it. Bye.