Hey everybody, we're now here. Welcome back to Tesla Daily. Good to be back with the

normal setup today after traveling out to AI Day last week. And today, of course, we're

going to be breaking down AI Day. And as simple of terms as we can, I know it was highly

technical, but we'll go through each of the five main topics that Tesla discussed. That

was number one vision. So neural networks and perception, things like that, planning

and control, so how the actual vehicle is operating, then labeling to train Tesla's

neural networks, then dojo, so the custom design Tesla hardware that's going to help them

with that training. And then of course, the Tesla bot that Tesla brought out at the end.

Before we even get into any of those topics, though, I think it's important to recognize

the clear overarching theme of AI Day, which to me is that Tesla clearly has a deeply

vertically integrated cross disciplinary AI team that is building things that are going

into products today, obviously with the cars, but over time, Tesla's ability to leverage

that team and leverage that technology is only going to grow to borrow a phrase from Elon.

This is not well understood. So keeping that in mind, let's go through these five areas

will start off with vision just as Tesla did in their presentation. This was Andre Carpathi

and he did say a lot of things that we've heard him talk about before, but he did share

some new details and importantly, some great examples of what Tesla has done. One of the

primary focuses of Andre's section, and this is something we've heard him and Elon talk

a lot about was Tesla's conversion over from the image space to vector space. So before Tesla

was basically taking a bunch of still frames, single images from each of the eight cameras

on the vehicle, running each of those images through their neural networks individually,

and then basically making decisions based on what the neural network perceived from each of those

...