Dr. Fei-Fei Li

There is just this kind of nuanced function.

There is just this kind of nuanced function.

And also the connectivity, how one area of the brain is connected to others, are not the same as neural networks.

And also the connectivity, how one area of the brain is connected to others, are not the same as neural networks.

So we're inspired but not replicating.

So we're inspired but not replicating.

So typically, AI model is given a vast amount of data.

So typically, AI model is given a vast amount of data.

And then some of the data are labeled with human supervision.

And then some of the data are labeled with human supervision.

Like if I give AI models millions and millions of images, some are labeled cats, dogs, microwaves, chairs, and all that.

Like if I give AI models millions and millions of images, some are labeled cats, dogs, microwaves, chairs, and all that.

And they learn to associate the pattern with the labels.

And they learn to associate the pattern with the labels.

Sometimes in recent, especially in language domain, what we call self-supervision, you give it millions and millions, trillions of documents.

Sometimes in recent, especially in language domain, what we call self-supervision, you give it millions and millions, trillions of documents.

And it just keeps learning to predict the next syllabus, the next word, because all the training data is showing you all these sequences of words.

And it just keeps learning to predict the next syllabus, the next word, because all the training data is showing you all these sequences of words.

And there, you don't have to give additional label.

And there, you don't have to give additional label.