Elaad Eliyahu

Fixing it.

And then people will look at good videos and they'll say not slop.

And they'll say, this is good.

It needs the human response to correct it.

So on X, you make a post, a robot responds, and the human goes, well, yeah, how dare you say that to me, and responds, providing insight beyond the first layer of human input, the second layer input.

Absolutely, 100%.

You're training everything.

You're training cars to drive.

So reCAPTCHA originally, like the first iteration we saw was two words would appear and it would say, type in the two words to continue.

One word it already knew, that was the key.

The second word, it did not know.

That's funny.

So the way it works, when you type in the first word correctly, it's assuming both words will be correct then.

And by corroborating what it does know, it creates a strong data set to figure out what these other words are, and they were scans from books.

The reason why they appeared bent and warped, it's because it was an image scan from a book, and we taught the AI how to read.

And so CAPTCHA today, it'll say, it says like,

Find the streetlights.

And you're like, oh, the streetlights.

This proves it.

Yet you're training cars to drive.