Mm-hmm

... um, some of which have become very evident with the recent generative AI algorithms that are consuming a huge amount of energy, uh, and, and water to train these models, uh, and to, uh, enable us to use them.

Um, and for me, um, the, the, the challenge is to make sure that we use these resources wisely.

I mean, it, it is a powerful tool, uh, and it is, it is helping many, many people.

But there are... obviously, with a tool like that, there are, um, possibilities of harm.

And we've already seen that in a number of occasions where these, uh, systems have led to people being, um, uh, making poor decisions, um, and even taking their own lives, which is-

Hmm.

Under some of the metrics, um, the AI way outperformed.

Can AI really create like a human can create?

That is a very good question, and one that I'm, I'm particularly interested in.

Um, so I think it would help if I just gave a brief, very, very quick introduction to what is happening under the hood of ChatGPT and equivalent generative technologies.

Um, so, uh-GPT, the letters GPT stand for Generative Pre-trained Transformer, if you don't know that.

Um, and, uh, a transformer, the task of a transformer is simply to, um, give you the probability of a sequence of words or tokens.

Uh, so you could give it something like, uh, "The moon is made of cheese." Uh, and, uh, if it's trained well, it would give you a high probability for that sequence of tokens.

Even though it's not true, it is a highly probable sequence of words.

[laughs]