Brad Lightcap and Ronnie Chatterji on jobs, growth, and the AI economy — the OpenAI Podcast Ep. 3

Hello, I'm Andrew Mayne, and this is the OpenAI Podcast. There's a lot of conversation and debate about the future of AI when it comes to labor and work. To talk about this, my guests are Brad Lightcap, who's the Chief Operating Officer of OpenAI, and Ronnie Chatterji, who is the Chief Economist. We're gonna find out the kind of research OpenAI is doing, the conversations they've been having, and hopefully get a glimpse of where they think the future is headed. [upbeat music]

We had a lot of people coming back to us and saying, "Yeah, you know, actually, this is, I think, one of the best things that has maybe ever happened to this, this industry." AI is a tool that lets people do things that they had no ability to do otherwise.

They have the world's smartest brain at their fingertips to solve hard problems.

So, Brad, you're the Chief Operating Officer, you're the Chief Economist. Explain what your roles are.

My role probably boils down mostly to [clears throat] what we call deployment. So zooming out, OpenAI is a research and deployment company, and when we think about our mission, what we really think about is not only building AI and doing the research that underpins the building of AI, but how do you actually take it out into the world and have people use it, and have it be beneficial, uh, for people, have it be safe for people? Uh, how is it used in one country versus another country, one industry versus another industry? So I spend a lot of time trying to figure that out, uh, which means working with customers, working with partners, uh, spending a lot of time with our users, and just kind of studying kind of how people-- uh, what people want from OpenAI and our products, how people actually use the technology, and then as the technology changes, how that pattern of use changes.

It, it seems like 'cause OpenAI started primarily as a research org and wasn't even sure if they were gonna do product or even put things that were sort of public-facing, and so how much has this changed rapidly for you?

It's changed really quickly. I think ChatGPT, uh, in November '22, was kind of the pivotal moment, and it was the first time that we really saw AI used at scale.

Mm-hmm.

Uh, and I think, you know, what we kind of... A- and it's interesting, almost, the, the story of how we actually learned that a- and, and how we made the decision to do ChatGPT, um, which was we had previously built an API for developers, and we had a thing, uh, you'll remember, Andrew, in our API, uh, that was, uh, the playground-

Yeah

... where you could basically try prompts out and see how the model would complete the prompt, and this was back in the days of, like, the models just being purely completions-based-

Mm

... where they take an input, and they kind of continue the text on, uh, predicting the next word, uh, and the next token in the sequence. And people were trying to, like, hack the playground, uh, to figure out how to get it to talk to them, and they almost-- you could tell people kind of wanted this conversational interface. And so we kind of learned from that, and we built ChatGPT as the first version of a conversational interface, where we taught the model how to instruction-follow to be more responsive to what people wanted to talk about. Uh, and that, you know, very much surprised us and became, I think, the kind of dominant paradigm of what we call the first era of AI, which was these kind of chatbots, uh, that, you know, really were good enough to, to, uh, to be engaging for people and be helpful for people.