Sam Schechner

And he's summoning people from other teams to work on improving that product.

It is a runaway consumer success.

More than 800 million people every week are using ChatGPT.

And it's a lot of what's driven the valuation of the company.

Behind some of that success is the fact that ChatGPT has built a sometimes emotional relationship with users.

People tend to love it.

And there's actually one model in particular that kind of epitomizes that, which was the default model until earlier this year.

It's called ChatGPT 4.0.

One of the really unique and special things about 4.0 is that it was trained using something called user feedback signals.

You know, millions of choices by consumers of which kind of response they preferred.

But those signals that made people love the chatbot, did they cause any problems?

Those same signals also is thought to have created a problem called sycophancy, which is in AI speak, basically a model that goes way too far in telling you what you want to hear.

Sam Altman in his memo says, we're going to pour resources back into improving chat GPT.

And what does he point to as the number one way to do that?

Well, better use of user signals.

OpenAI says that it has done a lot of work to make sure that its models better respond to people in distress.

And so the question here will be, how will OpenAI navigate that?

The delicate balance between giving people what they want, trying to make their product more popular in a tough commercial race, versus giving people maybe what's healthiest in the moment.

Thanks so much for having me.