I wanted to refer back to OpenAI's recent podcast episode on economic impacts, so I created a transcript.

The episode features their Chief Economist Ronnie Chatterji and their Chief Operating Officer Brad Lightcap, interviewed by former OpenAI employee Andrew Mayne. I hope others find this useful as well.

For a general sense of vibe of delivery, you might want to watch the actual video, using the rough timestamps indicated in the transcript. If anybody is interested in further cleaning up the transcript, let me know, and I'd be very happy to incorporate that. (I ran out of time to smooth out the disfluencies

Podcast begins

Andrew Mayne: 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 going to 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. 00:00-00:26

Brad Lightcap: (CLIP PREVIEW): We had a lot of people coming back to us and saying, actually this is one of the best things that has maybe ever happened to this industry. AI is a tool that lets people do things that they had no ability to do otherwise. 00:27-00:39

Ronnie Chatterji: (CLIP PREVIEW): They have the world's smartest brain at their fingertips to solve hard problems. 00:39-00:43

Andrew Mayne: So Brad, you're the chief operating officer, [Ronnie] you're the chief economist. Explain what your roles are. 00:45-00:48

Brad Lightcap: My role probably boils down mostly to what we call deployment.

So zooming out, OpenAI is a research and deployment company. And when we think about our mission, we really think about 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 for people, have it be safe for people. 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, which means working with customers, working with partners, spending a lot of time with our users and just kind of studying kind of how people, 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. 00:49-01:37

Andrew Mayne: It seems like because OpenAI started primarily as a research org and wasn't even sure if they were going to do product, or even put things that were sort of public facing. And so how much has this changed rapidly for you? 01:37-01:46

Brad Lightcap: It's changed really quickly. I think ChatGPT in November 22 was kind of the pivotal moment. And it was the first time that we really saw AI used at scale. And it's interesting how we made the decision to do ChatGPT, which was we had previously built an API for developers, and we had a thing - you'll remember, Andrew - in our API, that was the playground where you could basically try prompts out and see how the model would complete the prompt. This was back in the days of, like, the models just being purely completions based, where they take an input and they kind of continue the text on predicting the next word and the next token in the sequence. And people were trying to like hack the playground to figure out how to get it to talk to them. And 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. 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 that really were good enough to be engaging for people and be helpful for people. 01:46-03:04

Andrew Mayne: Yeah, it seems like because at the time, we kept thinking that like GPT-4 would be finally when it was really useful, and ChatGPT was built on top of GPT-3.5. And it seemed like certainly changing the interface was helpful, but we thought we needed a faster, smarter model, but it was actually the interface was such a big unlock. I had the problem whenever I would do demos at GPT-3, it would be this blank canvas. I go, "Now you do something." And people would be like, "I don't know what to do." But once you put it into the chat interface, they go, "Oh, well, I'll ask it a question. I'll ask it to do this." And that was such a big unlock.

But then the pace after that, like you said, was insane because ChatGPT exceeded beyond any expectation here. I think there was an expectation it would kind of level off, but it didn't. And then pretty soon there was this awareness. I think people thought AI was something in the future and now it came into the present.

And now bringing in an economist to come help map this out and figure this out. So what is your role? 03:04-04:06

Ronnie Chatterji: As you say, the future has arrived more quickly than any of us could have imagined. And so I joined at a time when we were deploying intelligence at scale into the economy, into society. And my job is to help people understand what the impacts of that are gonna be on businesses, their jobs, their relationships, the way government does policy, and develop forecasts to help people understand how to make investments with their time and overall with their resources. And so as an economist, it's an amazing time to join because I think we're at a beginning of a real transformation in the economy. And it's something that I think people need to be prepared for. So the biggest job I have at OpenAI is developing indicators to kind of tell us where the economy is going, and communicating that to people all over the world. Because this is going to be bigger than just the United States and what we do here, but something that's actually going to transform people's lives around the world. 04:07-04:54

Andrew Mayne: So in my experience and limited experience of understanding, like when corporations employ economists, often it's to figure out, let's say the prices of products or things like that to make kind of predictions, but here your job isn't just internal, so it's external. So how are you sharing this, what is OpenAI doing to help people understand where things are headed or where we think they are? 04:54-05:13

Ronnie Chatterji: You're right. I mean, there's a tradition of economists joining companies and in tech specifically. This job was designed a little bit differently, and I think it reflects that this company really has research roots. I think people really want it to be a job that, yes, thought about pricing and A-B tests and analyzing data from the platform. But maybe more importantly, also thought about how is this going to change the world and doing research - rigorous research, just as rigorous, but in a different way as our AI researchers do - in terms of what's going to happen and how can we tell people about it? How do we get people ready for this?

And so a big part of my job is external. Since I started, I've been in London and Brussels and Delhi and Washington. We'll eventually go to Sacramento and Sydney and every place in between. But it's so interesting to see the conversation and the vibes across those different markets and how people are thinking about this and the different use cases. So I have to say, as much as we go out and do that work, I learn as much in those interactions as I probably teach. But a big part of my job is external and getting sort of people ready for what's happening right now. 05:13-06:18

Andrew Mayne: Well, there's a lot of anxiety because I think OpenAI was caught off guard by the success of ChatGPT and the rate of adoption and the place it's being used. And I think every technology, you know, every disruptive technology, people, there's a fear of change and change is inevitable. But there is the fear of how it's going to change work, you know, how much it's going to change labor and employment. And how much does OpenAI think about that and how much of what you do is sort of like thinking about helping people adapt to that, et cetera? 06:18-06:47

Brad Lightcap: Yeah. I mean, I would say, uh, it's something that we look at a lot. I think Ronnie probably looks at it through one lens. I kind of somewhat look at it through the lens of what are the things that we need to build to accelerate the opportunity that AI has to be impactful in a kind of an economic and outcome oriented context. Um, and that could be at a level. It could be an individual person, for example, trying to better understand their medical care. It could be at a macro level, at a firm level. It could be a company that's trying to think about how to accelerate software engineering and pull forward projects from next year into this year. So all of these things actually - I mean, you know, Ronnie probably does the interesting kind of studies on these things and takes a much more scientific vantage point. We take a very product-led vantage point though on it, which is how do we actually go build the tools that are representative of the things that people actually want from the systems. And so, you know, software engineering is the thing that I think right now is super interesting is the systems we're building are progressing at just an insane rate in terms of their capabilities in software engineering. You've seen rise of tools like Cursor and Windsurf and others. And we think there's a huge opportunity there to help software engineers and kind of entirely change the tool set of software engineers to make them, you know, not 10% more productive, but maybe five or 10 times more productive. And then Ronnie gets to study the impact of that, you know, on an economic level. 06:47-08:04

Ronnie Chatterji: It's amazing. I think about it exactly this way. It's almost a handoff from what Brad is leading on the product side. Okay. Now our software engineers have these amazing tools, intelligence at their fingertips to be more productive. You know, we might across the world, right? Like a billion lines of code in a day and now you could multiply that by 10x what could we build what could they build if you can write that much more code and that much sort of even better code potentially than you could on your own that to me is a huge economic opportunity and so yeah my job is to pick it up from that angle understand how a software engineer's job is changing, how she might be using these tools to do things she couldn't do before, and how the organization that she works in is also going to benefit from that, creating more productivity and ultimately value for the economy. So I see it as a super interesting challenge.

The other thing I'd say is like scientific research is one I get really excited about. So I think taking Brad's analogy, it's like, we want to put amazing intelligence in the hands of scientific researchers. Why does that matter? Because science drives growth, it drives economic growth. And so if we can accelerate science, accelerate discovery, we're going to have more economic growth and more good things for everybody. And so I always think about if I can study how science is changing with our with these of our products and it'll be a useful contribution in terms of economics but also just the world. 08:04-09:19

Andrew Mayne: Yeah I want to touch on that a second but what in the software space I think you've seen there's been a I've seen a lot of people have a concern because all of a sudden companies are saying oh we don't need as many developers now but I I would say the broader picture is we're never going to be done writing software there is always going to be more need for software than there is right now and I think the challenge is that some of the bigger companies are getting a bit disrupted or internally, but we need to think about where the smaller companies, the more agile ones are going to come in and where they're going to come from, because I think small teams can do a lot more. And has that been something you've observed where, you know, let's say some companies are saying, okay, we can do more with this tool and what we're seeing smaller companies come forth with new solutions. 09:20-10:01

Brad Lightcap: Yeah, for sure. And I think that is the trendline of AI fundamentally is the the world is rate limited by talent by people um real you know economic growth in in the world rounds to zero in most places and why is that right it's because it's really hard for the average company whether it's a small business large business you know financial services company an insurance broker a hospital to find people that can actually produce better tools better systems and ultimately better outcomes for customers uh and if you go ask kind of any company you ask any company in silicon valley you know where uh if they need to hire more engineers the answer is almost always yes and this is this is the mecca of engine you know software engineering um now imagine what the rest of the world looks like and so just taking software engineering as an example we see it as not only incredibly uh you know there being an incredible opportunity to inflect outcomes for those companies, you know, for companies large and small. But we see it really as incumbent on OpenAI to be able to build the tool sets, the models, all the, you know, the safeguards, all of the compliance schemas and all that to be able to actually serve these tools in the places that they need to be. And it's interesting kind of the polarity of it. I think, you know, on the one hand, you've got this tool set that is going to be incredibly enabling of people who have no sophistication on the subject matter. So you've got companies now building tools that are enabling people to build software who've never written a line of code in their life. And on the other hand, you've got these tools that are incredibly sophisticated and taking level 10 engineers and making them 50% 2x more productive. And it's a remarkable thing that you can get both of those effects. 10:01-11:44

Andrew Mayne: Something I thought was interesting was using the Moderna case example where they deployed ChatGPT Enterprise. And one of the things that 11:44-11:50

Andrew Mayne: happened internally was you had people developing their own GPTs. And sometimes people go, 11:50-11:55

Andrew Mayne: what's happened to GPTs externally? But I think that's been an interesting thing is internally, 11:55-11:59

Andrew Mayne: somebody who may not have thought about how to build an agent or something like that, 11:59-12:02

Andrew Mayne: who may not be technically inclined, is able to do that. And has that been a common trend with 12:03-12:06

Andrew Mayne: other companies that are now just building on top of the platform? 12:06-12:09

Brad Lightcap: Yeah. I mean, I think that is fundamentally kind of how this is going to work. I think AI at its core and its essence is a tool that kind of lets people do things that they had no business or ability to do otherwise. There's going to be kind of crazy outcomes that come from that. I think it's kind of somewhat unpredictable. If you kind of look in the long arc of history of what makes for these kind of disruptive platform shifts, to me, the thing that is kind of demarcating of that is when you now have people who actually have the capability to go off and do something at either a much higher level of productivity or something that's parallel to the core thing they're doing that they couldn't do before where they were kind of rate limited or gated on someone else being able to having to do that thing for them and so that's gpts are a good example of how you now have someone who can configure what could be a fairly complex workflow right and it's on us to continue to build a product that enables even more complex workflows over time as the models get really good and that's that's a remarkable thing 12:10-13:08

Andrew Mayne: What sectors do you see being impacted next? 13:10-13:15

Ronnie Chatterji: I think that we're just scratching the surface when it comes to scientific research, areas like drug discovery, material sciences. I think the next couple of years, you're going to see massive discoveries in those spaces for the reasons that Brad is talking about. When I think about science, I think about an endless corridor with doors on either side. And scientists, researchers in companies have to make decisions about where they're going to explore. And that's a rate limiting sort of situation to Brad's point. You can't explore every door. But what our tools can help you do is actually look behind all those doors and take a peek and figure out where you want to spend the time working on the hardest problems. And I think if we can accelerate science in that way, you're going to see massive discoveries coming out of private sector labs, national laboratories, like many of the ones we're already working with, and the public sector. And so I expect those areas in research to really be transformed over the next several years. I think you'll see a lot of different discoveries that we wouldn't have thought possible happening more quickly. I think another area is going to be on sort of professional services. We both work a lot. I know a lot of folks who are in this industry, whether it's private equity, investment banking, consulting, so much of the work there that people are doing. We can augment that work by, I think about the way I use our tools to create slide decks or prepare for a presentation. I can now focus on the higher value and higher margin things that are important for my job now that I can use our tools to do some of these things that I was going to have to do myself. And so I see professional services as a key area where a lot of consultants, bankers and private equity executives are going to be able to use this in a big way. So those are two areas I see finance and science driven discovery companies being really revolutionized by our tools. 13:15-14:45

Brad Lightcap: And I would say it's not just the on the science side, at least it's not just the depth of any individual step of the work. So certainly like you can now do this, you know, more multifaceted exploration for any given thing. But it's the breadth across the span of the work that these models can reason over. So having these systems able to understand if, you know, if you look at kind of how a drug gets developed, for example, there's like, you know, some number of insanely complex, discrete steps in that process that all require kind of handoff at various points to a lot of different people who all have to kind of gather context from the person before them and kind of prepare context for the person that comes after them. And you can actually schematically break it down and have models basically woven across that entire workflow. Not only are you enabling the scientists to go deeper, but you're actually enabling the people who work with and around the scientists to actually kind of accelerate the end product, you know, ultimately to a better outcome and ultimately faster. 14:46-15:39

Andrew Mayne: One of the limitations I've seen. So one of the companies I've worked with, they're doing drug discovery and the models are great at suggesting things, but it still comes down to the clinical trial and the lab bench and things like that. And hopefully we'll find ways to accelerate that. But what are some of the other limitations, either to what these things can do or bottlenecks for us seeing sort of the benefits? 15:40-16:01

Ronnie Chatterji: I think human judgment decision making is going to be really important. I actually think it might be more important. What we're finding in a lot of research and one of my colleagues in this is David Deming at Harvard. He has this research that shows that people who are great at leading teams, let's say someone like Brad, the top of the company, they're also the same people who are great leading agents. And I think that a lot of the skills that let people make great judgments, lead teams, they're going to be even more important and at a higher premium in this economy. And so I feel a situation like this where firms are using drug discovery, you're still going to need the judgment of experts. You're going to need refinements on the experiments, and you're going to need help in terms of scaling. I also think there's other institutional changes, though, that might accelerate science. Clinical trials come from an old world of how we used to test drugs for safety and advocacy. Those are really important. But everything from the sample sizes to how you enroll people, I mean, our tools could hugely helpful in those areas. So I feel like you're going to see it in drug discovery, but you're also going to see in every part of the value chain for, let's say, a pharma biotech company that might ultimately not just increase the rate of discovery, but the rate of commercialization and scale. That's my hope. 16:02-17:04

Andrew Mayne: You just mentioned agents, and I think it's a word that's kind of like the word of the year. People hear it and sort of there's all sorts of definitions of it. Do you want to take a stab at that and kind of see how you guys see that playing out? 17:04-17:13

Brad Lightcap: I mean, I'll probably get, you know, yelled at by someone. 17:14-17:17

Andrew Mayne: No, it will not be controversial at all. 17:17-17:19

Brad Lightcap: I mean, for me, agents, I have a very high bar. It has to be a system that can be reliably handed, complex work that it can take on autonomously, you know, and execute at a high level of proficiency where it hasn't seen that work before. And that last part is a critical piece is these aren't just things that are trained to copy. They have to be things that kind of implicitly leverage the reasoning ability of the model to solve new problems. And this is going to be important in a lot of domains. And so people use the word agent. I think there's maybe an enterprise productivity context of it. There's maybe a science, you know, kind of context of it. There's a software engineering context of it. But the kind of common thread for me is it has to be something that you can actually hand something to. You almost work in tandem with kind of like a teammate. And, you know, that teammate could be a scientist. It could be a software engineer. It could be a data scientist. 17:19-18:16

Andrew Mayne: Could you give me like a hypothetical example of like a kind of task? 18:16-18:18

Brad Lightcap: Yeah, I mean, I think software engineering has an obvious set, which is, you know, you could ask it to basically go off and actually write code for you. And, you know, and then kind of similarly go do the QA, go do all the unit testing, go, you know, automate kind of meaningful parts of this process of the, you know, of code writing. And, you know, in different context, I would say it's, you know, it's working with agents that can make your your sales teams more efficient. So slotting into parts of your sales funnel where you have a volume problem, where it's like, okay, I've got, you know, 100,000 inbound leads for a thing, but I've got five people to look at them. Can you actually have an agent that can ingest those leads and understand those leads, process them, qualify them, move them through your funnel, recommend, you know, who should talk to who, recommend all the follow-up steps, and ultimately kind of drive a lead toward a conversion. So it's a generalizable concept that kind of maps in any number of areas. 18:19-19:17

Andrew Mayne: Do you see this like where I might email an agent or something and say, okay, I need to just treat it like I would another employee? 19:17-19:24

Brad Lightcap: Yeah, I think that's kind of the interesting part of it is that, you know, in some sense, there's the kind of the input mechanisms will be specific, I think, to the user, right? It's if you are a software engineer, you may want that agent living in your IDE. If you're a scientist, you may want it living in the software you use that you do experiment design and execution with. If you are, you know, doing, you know, user operations or customer support, you may want it sitting in your inbox because that's where your work happens. And so how do you build product that is intelligence underneath, but is extensible into kind of any number of surfaces and can be, you know, without compromising the reliability and the power of the system? It's actually a hard product problem. 19:25-20:11

Andrew Mayne: I have friends that are pretty ChatGPT focused, you know, power users, and have heard comments from before of like wanting to sort of do more with it. And even small business owners, too, the idea that if they could have like a virtual ChatGPT agent or something like that, is that something that you see in a near term horizon that, you know, I'd be able to like get to take care of a lot of the little work that there's just not enough hands to do? 20:11-20:38

Ronnie Chatterji: I think it's a really amazing near term application, in my view. You know, you think about the limits around the world of growing the economy. One of the biggest ones is small business. There's what they call in economics a missing middle in so many countries where you got a bunch of small businesses and you have a few large businesses, but the small businesses don't grow large. And that was a big benefit of the U.S. economy that our small businesses and entrepreneurs can actually grow in scale. In most places around the world, that's not true. Why is that not true? Because they often don't have the mentorship, the coaching, the support, the advice to actually know what to do to grow their business. Now imagine you democratize an AI agent that understands the basics of how to grow a restaurant business or an e-commerce business. And that's relatively easy to do in terms of instantiating that kind of intelligence into an agent. And then a small business owner could leverage that device and decide, oh, maybe I should change a menu item or hire a sales rep or do something different with my strategy that could help me grow. And I think for small business owners around the world, including the United States, tremendous opportunity to get sort of small business advice, evidence-based advice from agents. That's something I'm very interested in. I know a bunch of folks around the world are working on. 20:38-21:43

Andrew Mayne: So I want to address that one, the evidence-based approach in a second. But tell us more about what you're seeing from like developing economies. Because I know that's a big area of concern is one of the fears is that there's a lot of kind of like kind of lower level of knowledge work that's done in developing economies. and the fear is that AI is going to take that away. But you just brought up the fact that there are these limiters there that all of a sudden get unlocked. 21:44-22:07

Ronnie Chatterji: I think there's a lot of opportunities we should be talking about as well. I know that when I work in emerging markets, there's a lot of human scaling problems. It's related to what the rate limiting factors that Brad talked about with Silicon Valley hiring engineers. One of the biggest returns on investment in Africa is agricultural extension support. What that means is helping a farmer figure out what kinds of seeds he should be using, what kind of fertilizer he should be using, what kind of farming techniques he should do to get the most out of his land. Because a lot of people are small scale, subsistence farmers. If we can increase productivity for that farmer 10%, 20%, 30%, it is life-changing. And we have people who are trained up to do that, but there's not enough of them. And when these extension support services are offered, there's always someone, probably 10, who don't get the service for every one person who does. Now imagine that we could have intelligence provided to those 10 who never got that service to begin with. And I think when you think about agricultural extension support scaling with our tools, it's a huge opportunity to improve lives of people in sort of lower income countries and emerging markets, particularly in agriculture. I'd say the small business one is another example. You know, we know from the United States, one of the best ways to move up the income and wealth ladder is start a business. That should be true in other places, too. But there's so many limits to scaling. And often it is hiring the right person or getting the right advice. And so those are two opportunities, I think, if we can do this right, are going to make a huge impact for the positive in those parts of the world. 22:07-23:28

Andrew Mayne: My mother-in-law is in India and she has a candy company and she uses ChatGPT a lot to help her plan menus and recipes and write stuff. And it's been an interesting sort of unlock because now I've seen, I think Sherry had quality before, but now it let her basically spend more time on other things. And so it's interesting because you've seen like an African development where cellular was a bigger change than anybody predicted. It was you take a country like Kenya, which maybe like 5% of the population had phones and it was all controlled by the government or something. Then once cellular came through, then you had people were able to like figure out how to go to market. You had all sorts of commerce stuff, things. And what changes are you seeing right now with ChatGPT or like technologies? 23:29-24:14

Ronnie Chatterji: I mean, first, if your mother-in-law is running an Indian sweets company, I've got three little interns in my household who'd love to join. So just let us know if there's a job opening.

But this is where the disruption is like both exciting and I also understand it induces anxiety. But you're exactly right. When you look at the Kenyan experiment, when they leapfrogged a generation of technology, when new innovations came out, we're now doing something fairly radical, which is putting intelligence in individuals' hands. When they have a ChatGPT account or subscription, they have the world's smartest brain at their fingertips to solve hard problems. It's not intermediated by a government or a big business. It's something they can use to solve problems. And I'm really optimistic about the problems people are going to choose to solve. One of the coolest things about this organization is we don't really tell you what problems to solve. That was one of the most interesting things, I think, here is when you think about how people are using ChatGPT, it's a wide, diverse set of uses, much less how they're building on the API with our developers. And so people will choose to solve the problems that are most relevant to them. And that's going to be incredibly transformed lives but also disruptive because right they're going to be able to have that power that they didn't have before and I think when I think about it as an economist those are the kinds of transitions I want to study I want to understand I want to make easier for individuals organizations and society and I think the level you're talking about happened in Kenya and other parts of the world this is a much bigger transition that we're on the verge of so it's something that my team spends a lot of time thinking about when we look at data not just looking at the US and Europe but but looking at other parts of the world. 24:14-25:39

Andrew Mayne: You mentioned before in working with agents, how having sort of, I guess, managerial skills or the ability to delegate is important. Could you expand on that? And also maybe like what other skills might be important that people need to be thinking about that they want to develop? 25:39-25:52

Brad Lightcap: AI is interesting because it really is kind of a reflection of your will, right, and your desire. And I think it sky's the limit kind of in terms of what it can do for you, right? If you wake up one day and you decide you want to start a business, that just got meaningfully easier. If you wake up one day and you decide you want to build a piece of software, right, that got meaningfully easier. And so there's an incredible level of agency, I think, that's required to extract the most out of AI. I think as we think about kind of where the product moves, our job is to try and lower the bar so that you can basically simplify the kind of the path from idea in your brain to outcome. And, you know, there's interesting ways in a meta sense the models can actually help help do that. But I think that, you know, the that's probably to me the kind of really important, important thing is that the agency is going to matter a lot. There's going to there's going to be you're going to see the rewards accrue to people who are, you know, Sam said it the other day is like the return of the idea guy in some sense. It's the people that I think can, you know, not only figure out what it is that they want and, you know, what good looks like, but then can kind of figure out how to activate the systems to be able to work on their behalf. And there's going to be people that do that incredibly well. And, you know, one of my kind of personal bars for how impactful our work ends up being is will you see the rise of companies that are one, two, five, ten people that are doing a billion dollars in revenue, right? That's kind of the ultimate agency outcome. If you think about it, it's like you have a very small set of people capable of commanding, you know, what could be this very large scale enterprise, you know, mostly because they are opinionated about things like sales, marketing, products, software engineering and so on. And I think that's going to be a really cool, cool thing to see. 25:53-27:48

Andrew Mayne: Mark Benioff had said something along the lines that they weren't going to be hiring any more software engineers, which maybe they over hired too, I don't know. But then they're going to be increasing the number of salespeople. And I think that often people hear the word sales and they think somebody calls you up randomly or cold calls, but sales is actually a big part of it as people who are networked, who know a lot of other people. And I think that's what he was talking about was what was going to be really valuable to the growth. We're humans with human connections. And is this something you've seen data to back this up or to see this as a high growth area? 27:48-28:19

Ronnie Chatterji: Yeah, a lot of the research coming out on this is showing that EQ matters a lot. You know, a lot of people think in this world it's getting more and more technologically sophisticated. All of a sudden the soft skills, the social skills, being connected with people would be less valuable. It's actually the opposite. Once you make these abilities and these capabilities democratize to build a right code, for example, then some of the other things actually start to matter more in the market. And so I'm not surprised at all that salespeople who have deep technical knowledge, and we have many here in Brad's org and across the organization, are going to be at a premium around the world. Because those are people going to be able to connect the dots, use their EQ plus their technical expertise to solve problems. And I feel like when you're thinking about what skills we want in the economy, that's going to be a key part of it, as well as critical thinking and decision making. We're still going to need people to identify those problems to chase after. Right. And that's where Brad talked about the agency combined with the ability to target the right problem is going to be at such a premium. I expect that to be really important. 28:20-29:15

Andrew Mayne: I've seen in tech, I think there's this over indexing on IQ and horsepower. And I'm a big believer that I think these systems are going to be able to do just about any cognitive tasks we can think about. But you brought up EQ, we think is a really important one. I don't think that enough attention gets paid to that because I know some small companies that scaled really big and they build great products. I can't get anybody on the phone. I can't talk to anybody because they're just focusing purely on the technical component and not where they exist in the network of people and everything else like that. And what are ways that somebody right now who wants to be, you know, find themselves in a, you know, very aligned position with the future? How do they build these skills? How do they work towards that? And how do organizations find people or foster that? 29:16-30:00

Ronnie Chatterji: I mean, I think it starts in schools. Like, you know, one of the really exciting things about the moment we're at is education is going to change. And I know that also creates a lot of excitement and anxiety. But I think so many things that we're learning in school, I have younger kids. And so in elementary school grades, they're going to be even more relevant. What are you teaching people when they come into pre-K or kindergarten? You're teaching them how to be a human. And I can't think of a better set of skills to learn now than how to be a human. Because that's going to be sort of how you become a better complement for this amazing intelligence. As an economist, you think about two constructs. Substitution, which creates a lot of the anxiety, but also complement. If humans can become complements to intelligence and leverage it with agency, that is going to be the unlock. And I feel like a lot of schooling in the early stages, even now, and it'll be more so as we go forward, is teaching those kind of soft skills and how to be a human. Later on, critical thinking, financial numeracy with numbers, still going to be really important. I mean, my kids have calculators, but I still want to teach them how to do multiplication tables. Dictation software works really well. I still teach them how to write. You'll need those skills and you'll need a sense of some other kind of higher order cognitive skills, resilience, grit, things that they're going to need to adjust to these changes in the market. So when a CEO says, look, we're looking for more something like this instead of that, students in the future are going to be able to prepare to pivot in the right way and have that baseline skill. That's kind of how I think about people preparing. I think education will play a big role. I think work experience at great organizations can play a role, too. Those are the two areas. 30:00-31:21

Andrew Mayne: I've been advising some students and I won't name the college, but it's in the Bay Area. It's a pretty good college. They have a pretty good CS computer science program. Do you know how many days they spent in the last semester learning how to use tools like Windsurf or Cursor? 31:22-31:39

Brad Lightcap: I don't know. Tell us. 31:40-31:41

Andrew Mayne: Zero. None. None of their professors have taught them anything about how to use AI coding agents. 31:41-31:46

Brad Lightcap: They're probably all using in the background. 31:47-31:48

Andrew Mayne: Oh yeah, they are. And I'm also the ones that aren't, I'm strongly encouraged them to that. And I think that was sort of, for me, a surprising thing to find out that at that level, they're about to be put out in the workforce and they're not even getting a day. And I understand you want them to understand the fundamentals, you want them to understand that, but they're going to be applying for jobs. I help them put together projects and stuff so they can get jobs from places. But what is OpenAI's role in policy, both from education and policymakers and stuff and trying to advise or influence? 31:48-32:18

Brad Lightcap: It's a good question. I think, you know, there's no question that we're headed toward an overhaul, I think, of kind of how the education system works. I think that will be a positive overhaul. I mean, you know, at the most kind of reduced level. Right. What what is it that we're building? You've got this thing now that is this kind of personal tutor of every person on Earth. Right. And as it gets better, it will start to understand you better. It'll understand your rate of learning better. It'll understand how you like consuming information. Right. Are you more visual? Are you more quantitative? Do you need things explained certain ways? We've had the amount of feedback we get from people, for example, even with children who are dyslexic, trying to learn the impediment that that creates in the learning process and ways that AI can unblock learning for that population. It's consistent. And so I think that the entire kind of way that we think about education and what education is in the country will have to adapt. I think it'll be good, though. I think it will force in some ways our systems to think about, you know, what are the ways that people will use these tools in the future? I think we, you know, the example you gave is in some ways surprising, but in some ways not. I think the people adapt faster than the institutions. But the question here will be, you know, how do we work with policymakers and with the institutions themselves to try and help the institutions adapt? I think the ones that do, though, will have this incredible accelerant. I think that you will see the outcomes among students and the ways that they think about, you know, what this tool can do in the classroom will will just fundamentally change for the better. And it will also then free up teachers, free up students to spend more time on things that are going to be the kind of high leverage skills of the future that Ronnie mentioned. So things like decision making, things like critical thinking, you know, tool based problem solving. How do you, you know, how do you kind of develop agency and conviction early in children? I think that that type of thing is going to be, you know, super important as opposed to a curriculum that today, you know, reinforces things like memorization, regurgitation and so on. 32:18-34:29

Ronnie Chatterji: Yeah. And I also say, I mean, I'm pretty optimistic that we can make these changes in the education. and I think it's going to come from teachers and students, the way Brad's talking about. Like in the early '60s, President Kennedy said, we were going to put a man on the moon. And if you look at what we actually had in terms of national assets at that time, and like the scientific capabilities, that was a pretty far off goal. But during that decade, we dramatically increased the number of people doing PhDs in sciences and engineering as people geared up for this challenge. And so I do think there's a really strong role for leadership across sectors to kind of sound the clarion call and say, look, this is where we're going to go. When I think about OpenAI, I think about, we have the best information about where the technology is going. That's an important role to play, to let people understand, here's what we're building. Other people in society, education leaders, government leaders, business leaders, and other sectors, will be able to see it from their perspective. If we put that call out there, I think you're going to see a lot of dynamic changes across. Brad and I are both loyal Dukies, of course. At Duke, I expect the curriculum in computer science and economics to be really different five years from now, in a lot of positive ways. And I expect a lot of experimentation, you know, beyond whether you can use chat to study or how you regulate in the classroom as a professor. Really important points. I don't want to downplay that. But more important is how are you going to use this stuff to do stuff, topics in your curriculum, help students who maybe can't learn from a graph, but can learn from a oral presentation or, you know, teach students the same thing, though in three different ways. So everyone in the class gets it. There's so much amazing stuff that happened. I do think it'll happen. I think we have a history here in the United States and you'll see this around the world as well. but I know the u.s the best where we've actually responded pretty dynamically to some of these big challenges. 34:30-36:08

Andrew Mayne: Can you talk a bit about OpenAI's engagement with educators and policymakers, specifically about what you are doing? 36:08-36:15

Ronnie Chatterji: I can start with the example of cal state university so uh you know for those of you from california spend a lot of time here cal state is like you know just the ultimate unlock for students who are first generation whose parents um maybe they come from another country or they haven't attended higher education those are the kids that cal state specializes in. Those are the students that Cal State has traditionally, for its long, illustrious history, taken on the next level. And so we're proud to work with them. And this is something that comes from what Brad is talking about, the research and the deployment. Someone like me picks it up and says, "Okay, now that we're working with this great institution, how are we going to maximize the outcomes for students when they go for that first interview? Can we prepare them with the skills they need to do well? Can we track their career outcomes over time and say, you know what, having access to this intelligence made a huge difference. And so that engagement has led me to work with administrators at CSU, researchers, and once we get everything in order, students, ultimately, to make sure we're going to make a big difference. And so for me, it's been a great interaction and it's been facilitated by the deployment we've done with CSU. So that's an education example. 36:15-37:17

Brad Lightcap: I mean, education has been for us the fastest growing segment that uses ChatGPT and other OpenAI tools. So it surprised us a little bit, I think, in some ways. You know, we knew early on when we launched ChatGPT that it had a resonance with students and that it was clearly applicable to the way that people wanted to learn, engage with information, engage with knowledge, test their own learning ability and skill set. And what funny side story is we, when we, right after we launched ChatGPT, we launched it in November of 22. We had basically the kind of remainder of that school year where I think there was a lot of upheaval, I would say in that sector. And you probably remember this. Um, and for a while we all looked at each other here and we're like, man, we, you know, I don't know what this is going to ultimately lead to for us, but you know, and, and it is all the stuff ultimately going to get banned something over the summer of 23 as the school year changed over. I don't know what it was that went around, but when, when everyone came back in the fall, the level of enthusiasm, and I think the level of forward-lookingness of the, of the leadership in kind of the broader American educational system had changed. And it was, they, 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. It's meaningfully changed how my students are learning. We're starting to develop perspective on how people are really using this. And not only do we have that perspective, I actually want to extend and develop that perspective so that I can figure out how to better use this in my classroom, work it into my curriculum, challenge students in new ways, right figure out ways that um we can actually have it surface gaps and vulnerabilities in certain student populations that maybe aren't getting the attention they need so all that work now kind of is culminated in work that we're doing internally with an edu team here at OpenAI to try and work more with the sector uh you know ronnie mentioned the cal state example is just one of many examples of of ways that we're we're trying to to engage and so part of it is product building part of it is engagement part of its policy but we are going to take kind of a whole of company approach to it 37:18-39:14

Andrew Mayne: I remember I went in the school system, but they famously had banned it. They're like, "Oh, we're banning this to use the school system." And then I'd heard anecdotally that a number of teachers within had been using it and having really positive outcomes from any reasons you pointed out. You know, I helped do a study when I was here and one of the number one feedback we got was from students was, "It doesn't judge you. ChatGPT doesn't judge you." And it was a great way if you're feeling you're going behind or whatever to go ask questions and get up to speed. And then we saw that some of the teachers were getting really good results in the classroom and didn't went to the school system and said listen no we we need this this is something we've been sorely lacking and there was a kind of a famous reversal on that and that was I think like that happened faster than I expected and would you say that you're seeing probably a faster adoption than you'd been expecting or was I just not with it 39:14-40:05

Ronnie Chatterji: I've been seeing it I mean I think you're right there was that transformation sometime in 2023 where people realize wow we can unlock a lot of value here for for students and for professors. And maybe what happened over that summer, I don't know, maybe it happened for me. One of the biggest barriers to innovation for new faculty members, let's say, at a university is developing a new curriculum. So someone says, "Look, hey, this topic's hot. Why don't you develop a whole class on it?" Professors wanna help their students. They wanna introduce some new material, but there's a huge cost as it puts together against your research, your other teacher responsibilities. But all of a sudden I can use the tools in ChatGPT to develop that syllabus. I can make a great entrepreneurship and AI syllabus now much more quickly than I could before. It can help me decide what classes I'm going to teach, the slides I might use, the readings I might assign, even discussion questions for my students. When you lower the barriers to creating new content, it becomes even more exciting for a professor to try something new or a teacher in the K-12 context. So I feel actually that now as faculty and teachers are unlocking that, you're seeing a lot more adoption. I think the other thing is that at the end of the day, introducing students to new ideas they wouldn't have had otherwise is such an amazing thing. Any teacher sees that and there's a spark. And that's going to make them want to find a way to use those tools. We definitely need rules and policies they'll set up at the school level that's really important. When and how students use these tools, that's going to be key. And I imagine those will be worked out and there'll be variation across different educational institutions. But I have no doubt that it's going to be a huge part of education, given how valuable it is. 40:06-41:27

Andrew Mayne: We talked about this a little bit before we started recording, which is there has been years and years, a century of speculation of what happens when you have intelligent systems. how is that going to disrupt the world, whatever. And now we're in the place where we're actually starting to see this happen. And we realized that I think a lot of it was fanfic and it was just so scenarios and the scenario is playing out and it's very different. And I think your approach has been you're very evidence-based. It's the idea that you prefer research over theory. And where are you directing your research right now for impact and guidance for policy? 41:27-42:01

Ronnie Chatterji: For my work, at least on the economic research part, that narrow piece of it, I've been thinking about a couple of things. One going to be affected first. I think what I can do to help the organization, but also the world, is if I can identify that sectors like healthcare and education might be transformed more quickly, let's say, than retail and finance, that's a really important insight to provide to the world. Because if people are in those sectors and thinking about their jobs and what they can do, it both unlocks opportunity on the enterprise side, but also helps people plan their careers and make their investments. So one of my big goals is to figure out which sectors are going to be influenced first and by how much. The next thing I've been thinking a lot about is which countries, which geographies are going to be most affected. I don't think that's really helpful. When I look at previous technological transformations where people were left behind, a lot of the impacts were geographically concentrated, let's say for big manufacturing hubs in the upper midwest and the United States during the last transition. And when you look at that disruption and the scarring that happened over many decades afterwards, I realized that if we can develop good indicators of where in terms of geography these effects are going to be most pronounced that's going to be really really helpful so my team spends a lot of time on that as well and the last piece is communicating it you know a lot of economists or if I was in academia that's sort of the last piece the piece you like tack on there and say okay well you know somebody besides my mother is going to read my paper with its you know 33 appendices in this job especially given the privilege that I have to be close to the researchers who are changing the world I got to be able to translate that for real people. So those are the three aspects, I think, sort of where geographically, which industries, and explain to the world how that's coming. And that's kind of where the evidence base that I at least want to develop is coming from. 42:01-43:36

Andrew Mayne: It seems like a big unlock that kind of went unnoticed was when ChatGPT went to from you had to have a credit card, you had to have, you know, your login and all that to now you just go to openai.com and you can just use it, which just increased accessibility around the world. And, you know, I think it's been iPhones now and seeing that kind of rollout there, which I think was a really good democratization of it was the idea that it went from, you know, only a certain part of the world was going to be able to have access to it to now anybody in, you know, unrestricted countries, you're able to use that, which I think is great. I think that's, that's, it's very cool. You mentioned though, research into sectors are going to be effective. What have you found out so far? 43:36-44:16

Ronnie Chatterji: So far, I think the sectors that are less regulated, where there's less, let's say, sort of red tape, rules of the road that need to be followed, those are the sectors that are going to change the quickest, right? And so sometimes that's like healthcare for very good reasons. We have sort of HIPAA, protecting patient privacy. We have rules on how care is delivered. These are really important parts of the US healthcare system, and they are similar around the world. Those are sectors that are going to be harder to change, right? And they're going to be slower to adopt new technological tools. And that's not just true for AI. It's true for previous incarnations of technology. IT moves slower into healthcare education than it did to other sectors. So I think where you have sort of high levels of regulation and compliance requirements, you'll see slower adoptions and those jobs changing slower as a result. Doesn't mean we can't unlock a lot of productivity in healthcare delivery and education. In education, we're seeing this on the student side and teachers, but overall like implementation, I think you'll see it move faster in sectors where the regulations aren't as sort of significant as you are in those two sectors. That's key. And the other thing you'll see is where the workforce is going to embrace it. Brad made a good point earlier. It's like this happened with enterprise software. People brought tools to work, like new storage solutions. And then their CTO was like, hey, what are you doing there? Right? And then eventually they're like, wait, this thing you're bringing is actually, the whole company should adopt it. In sectors where you have highly skilled workers who are bringing these tools to work, using things like ChatGPT, building on our API, those sectors are going to transform more quickly. And that's why I think places like finance, right, sort of research drug discovery type organizations, that's places where I think you're going to have those people bringing it to work to solve problems. I expect those sectors move pretty fast. 44:17-45:48

Andrew Mayne: What career advice are you giving your children? 45:48-45:53

Ronnie Chatterji: That's the hardest question. Um,and what I tell my kids is when I was growing up, I was a son of immigrants, right? So like, if your parents are from a certain part of the world, the advice you might get would be, there's only two choices, right? There's like be a doctor or be an engineer. And if you're really creative, you can be a biomedical engineer. Okay. So there's like a narrow set of choices. Why would parents give those advice to kids is because it's like, they would predict these are to be the stable professions. But during the course of that generation, healthcare changed a ton, right? We had managed care. A lot of physicians work for hospitals. The job is so different than the generation that was giving that advice thought it would be. Engineering, I mean, Brad talked about this earlier, has changed dramatically. We never had full precision and full predictability to say, your kid should do this. In fact, many of the jobs we have today, we didn't have names for them in 1940. So first I have a dose of humility, which is like, it was never easy to tell our kids what to do or guarantee they would listen. For my kids, though, I reflect back on what we talked about, which is you've got to learn how to be a critical thinker and identify problems, develop a point of view to have the agency Brad's talking about. You have to have the neuroplasticity, resilience, flexibility to be able to adapt because the world is going to change a lot. If you think about what's happening in AI, changes to our climate, changes to geopolitics, you're going to have to adapt a lot. And the last piece, I do think that the EQ and the financial numeracy will be really, really important as they navigate their careers. In terms of predicting what their job title is going to be. I don't think I have any more information than my parents did. And I think they're gonna be okay. 45:53-47:13

Andrew Mayne: It's an interesting note that like, the title may change or see the title may stay the same, but the work may change. One of my favorite anecdotes was Dan Bricklin, the guy who created VisiCalc. He had in the 1970s, he was a high level programmer, extremely capable and programming was changing a lot then and you're moving into object oriented programming and libraries and stuff. And he thought that programming jobs were going to become more scarce. And so we actually left to go get his MBA. And it was while staring at the back, the blackboards with all the figures. He's like, why doesn't somebody make like an electronic spreadsheet? And then he invented VisiCalc. And it was just funny though, to read how he thought that programming job was ending in the 1970s. And I kind of think that it is changing a lot now, but you mentioned about how, if you're somebody who's running, you know, an AI software tool, you're kind of managing a project, it's project management and having the technical skills is certainly critical. And I've heard this a lot, like why bother learning to code? And I'm like, you know, do I want an airline pilot that doesn't know aerodynamics? You know, what are other skills you think are still going to be mattering, you know, in the future? 47:13-48:16

Brad Lightcap: Well, I think the direction of travel of technology is toward always toward individual empowerment. I think if you look at trend wise, every kind of past technological revolution and every past phase change, it always drives toward the individual and what the individual is capable of. So, you know, 1900, you had 40% of the U.S. economy working in agriculture. Today, it's 2%, right? And we produce some multiple more, you know, agricultural output than we did in 1900. And you can run a large farm with a small fraction of the number of people it would have taken to run a large farm in 1900.

And so now what happens when that, you know, you get that same phenomenon kind of applied widely across the economy and in sectors where historically we haven't had that phenomenon. And I think that there are a lot of places that would benefit from a phenomenon like that. And that's not to say that there's a you know, it's an argument for job displacement, for example. But I think that the argument here is toward, you know, higher economic output kind of per unit of input. And that fundamentally is what drives economic growth. But people are resilient. They find other places to to to to to go work. And when you create the kind of local, the micro level empowerment, you tend to create the second and third order effects of of other jobs that get created that we couldn't have foreseen in retrospect. It would be weird to tell someone in 1900 for example that there are people today whose entire job is to make content for a small little device that people consume many hours a day, and that those people can make a perfectly kind of viable economic living. It would seem something that was almost kind of unimaginable that you know exist, but it does. And so there will be that kind of second set of changes and second and third order impacts. I always kind of come back to the kind of individual empowerment point of the direction of travel being toward more people being able to do a lot more with a lot less. And then, you know, their labor and their ideas and their creativity creating kind of the downstream opportunity for people that 20 years ago would have been doing a different job. 48:17-50:27

Andrew Mayne: The example I use is in ancient Mesopotamia, 98% of people were in agriculture. And all of a sudden, somebody invents the plow. And if you're thinking, well, we're all farmers, we're doomed, that may have been a mindset. But the reality was that led us to inventing education and healthcare and actually governments and all these things. And I think that that's, to your point, that's the thing I think we sort of forget is that we've had huge, huge upheavals, like literally taking, going from 98% and, you know, agriculture to all of a sudden where they go. And, you know, if we thought back in the year 1800 and said, hey, we're going to get rid of almost all farm jobs, people will be thinking, well, what are we going to do? There's going to be massive problems. And we realized that, like you said, we created all these new kinds of roles into higher sectors, the economy and stuff. And it's always hard to predict, though. It's always hard to predict where that's going to be, because we just imagine the future is sort of the present with like shinier clothes and robots right and flying cars. 50:28-51:31

Ronnie Chatterji: And part of I think this is part of the job of research and organizations that are close to technology produce the information to help people make the best decisions you know brad's talked about agency agency requires sort of an individual characteristic right but it also will require information about what the market looks like where technology is going and so I feel that as a big responsibility what I'm doing here you know our mission is to benefit all humanity and to do that I want to make sure people have the information they need to the best of my ability right we can't predict with perfect fidelity whether it's going to be and I can't tell my own kids much less than anyone else's kids but if I can give good information based on research that'll help people make better decisions and I and I do think ultimately find a place where they can flourish. 51:31-52:03

Brad Lightcap: We should also keep in mind there's a lot of people who can't participate in the economy the way that they would like because of extenuating circumstances in their life that are you know born in part of things like lack of access to health care lack of access to to education. I mean, we talked a little bit earlier about what are the impacts that we might see in parts of the developing world where access to those resources is scarce. You know, there's the direct impact that we have of how do you make it easier for someone to scale a small business? That's going to be a clear, present, and I think very positive set of things that, you know, that happen. There's a kind of second and third order, almost hidden impact that we also, I think, you know, Ronnie has the challenge of having to figure out how to measure this, but what happens when you enable people to better manage healthcare, right? Or better manage the healthcare of someone who is dependent on them, you know, a, a, a sibling or, you know, a parent or something like that. What happens when you raise the education level, you know, and the educational outcome levels by 2%, right? And what is the kind of second and third order effect of that as a downstream impact on, on the economy and on, you know, on, on, on people's ability to participate in the economy. So, you know, there's, there's kind of the direct way to look at this. I think there's also the indirect way to look at this and uh that's right you know I just deferred it to ronnie on on whatever he can measure. 52:04-53:17

Ronnie Chatterji: I think I think this is a really good point though brad something I've been thinking it's hard to measure but important here is coaching mentoring counseling when he talked about people who can't fully participate in the economy my mind immediately went to there's so many people who have so much to offer but maybe uh they're neurodiverse or maybe they need a coach or someone to help them get to the next level or a level of counseling behavioral health, which we don't have broad access to in many cases. It's expensive. Depending if you live in a part of the country or the world where you don't have access to that, you're sort of sidelined. Economists will use the technical term of labor force participation. What it really means is you're sidelined. You can't participate. But if we can help people, compared to having no help now, some help, we could help them participate in the economy more fully. That can unlock a lot of potential. I do think any equation, any cost-benefit analysis, any sort of reckoning about what's going to happen in the economy needs to also consider people who aren't participating the way they could now getting enabled. I think that's a really important point. And we'll try to measure it, but even now I think it should be thought about in that way. 53:17-54:15

Andrew Mayne: I think we in developed economies sometimes have a habit of forgetting the things we have that other people don't have. Like if you want legal help, you hire a lawyer. But if you're living in a subsistence level, you know, economy, like that's hard. You can't do that. You know, financial planning, everybody would benefit. I mean, the people benefit most from financial planning or people have had access to financial education. I think that's an exciting area. And I think that certainly we're going to start seeing those effects eventually when you start to see what happens when you unlock so many of the people around the world by just the circumstances of where they were born and had access to didn't have access to that information or expertise. I think that's going to be very cool to see how AI makes that possible. We've seen just through ChatGPT alone how people would use it as educational tutors, helping translations and helping small businesses, helping people who have to work in communication, etc. And that's been a benefit. And I think that we've seen in some situations where you have a tool, let's say translation, people often think like, oh, that's going to decrease the need for translators, but actually can increase it because all of a sudden, a company that never did business overseas now can, one, send out query letters, etc., now find themselves dealing with an entirely different country in the economy and that's increased the demand for a human skill do you think this is going to be happening in other places do you think this is going to be just a big area of opportunity or is it just going to be minimal. 54:15-55:37

Brad Lightcap: Look what we actually see in our in our data at OpenAI is when we cut the price of our models, which is really cutting the price of intelligence, we see a disproportionate increase in demand for that model and so uh and we see that with ChatGPT too when we make better intelligence available and more of it available people use it right? And we don't see the upper bound yet on kind of how intelligence and demand correlate. It seems like there's just this relationship that the more we can make really great intelligence available more cheaply, the more consumption there will be around that thing. And so if you think about kind of how that plays out in an economic context, what happens when you've got, if you can cut the price of good legal advice, for example, by a factor of a hundred, do you see a corresponding thousand X increase in the demand for legal services? same in healthcare, same in education, same in software engineering, same in any other thing. And I don't think we've, we've quite come to terms with what that means. And if you think about, you know, thousand X demand increases kind of across every segment, that's a lot of strain, right? That's a lot of demand in the economy, which is a good thing. But, you know, I think there's, and ultimately people are going to have to kind of like organize themselves to figure out how to serve all of that need and, and, and be there to serve those, you know, that demand and those and those needs um which means that you need people who are going to come up with ideas take initiative go start things go create things right um and so that's the dynamism I think of the economy that's underappreciated when we talk about what the impact of AI will be uh and it's the thing that we see at a very you know at a micro level now at OpenAI but um I think you know we should look at as uh as we make intelligence you know sam has a phrase I love too cheap to meter what does it really mean for the world's ability to create output? And then ultimately, I think that the downstream impact of that we'll find is that it actually had an incredibly enhancing effect on jobs, on productivity. And that's the positive future I think we're excited about. 55:37-57:36

Ronnie Chatterji: And as an economist, I'll just say, this could be really exciting for people in those professions in the following way: When you sort of make intelligence too cheap to meter and that intelligence of providing, let's say, legal advice or financial management advice or advice on real estate, all of a sudden you get a bunch of new people accessing that never could access it before. So it's opening up the market, number one, and then once those people sort of start making decisions - buying a property, making a transaction, engaging with legal services - they're going to have higher and higher level needs, and all of a sudden the business that they're running is more complex, or they have two properties to manage. And then there's a bunch of people who are trained in those fields who never served this market before but now they're are going to come to them with more complex questions. And if that accelerates, that could create tremendous opportunities in those fields. It will be about deciding which part of the market to focus on, what kinds of skills you want to leverage. But I think for real estate agents and insurance brokers and financial advisors, there's a potential for this to actually onboard a tremendous number of people who never would access their services to begin with. That's the excitement of reducing the cost of intelligence dramatically, which is what's happening. 57:37-58:37

Andrew Mayne: I would say an example that's, I think, very close to home here is that every time there's a new model or in some new technology from OpenAI, you'll get some pundit will go, well, how come they're still hiring? And I think I'm a growth mentality person. I'm like, well, of course they're hiring because -and I want to get a sanity check on this - my prediction I've told people is more people are going to work for OpenAI after AGI than work before it. That it's not like all of a sudden, great, we've got a new tool. You don't need people doing these roles before the roles change, but you're going to want more people. Do you think, would you agree about my assessment of the trajectory of OpenAI, more people here after AGI? 58:38-59:22

Brad Lightcap: I think it will be more people after AGI. I think I kind of go back to what I said earlier of how do you, you know, the kind of demarcator of the impact of AI being about kind of more output per person, right? And so you now have, you know, you get this kind of scale down basically in kind of how large of a firm or company can be run by some number of people. So, you know, a large enterprise had to be run by a hundred thousand people before maybe that number comes down to 50,000, eventually 20,000, you know, 5,000, a thousand, you know, a hundred and so on and so forth. Um, and maybe it's, you know, it's an even steeper fall off than that. Um, and so I, I suspect and hope OpenAI will be kind of no different than that, you know, especially given what we do. Uh, and so, you know, I think, but I think going back to also the point what do we see as the second and third order impacts of the deflationary aspect of intelligence, right? It creates significant and disproportionate demand for the service. And so what does that mean for us? It means we need more people that can help work with more users across more use cases. It means we need more people helping policymakers think about the problem. It means we need a chief economist. If you'd asked me three years ago, if we would have needed a chief economist, I would I would have said, I don't maybe in 2030, but here we are. And so, you know, I think that that's I think it will be more people. But I think it is kind of somewhat consequence of both of the trends I just mentioned. 59:22-01:00:41

Andrew Mayne: I am helping out a friend who's working on training models and stuff to help with cancer nutrition. And we were talking to somebody from OpenAI yesterday and they said, would it be helpful to talk to the health team? I'm like, you have a health team. I said, yeah, we do. I'm like, well, great. And then, you know, having a conversation with them. And that was a thing like, wow, this is what a great area of expansion. And I think that I hope that other companies are sort of thinking about how these tools really are augment and amplify and create opportunities for growth. Because I think that if you have good talent, you want to keep that talent and find more talent, not find a way to not need talent. I think that'll put them at a disadvantage if they're, you know, not being forward thinking about that. 01:00:42-01:01:21

Brad Lightcap: Yeah. I mean, if you can get incredible leverage on every marginal person, you know, 10X, a hundred X leverage that you could get from 10 years ago or something like that, why wouldn't you want more people? If Ronnie's team can now, with a team of 10 people do economic analysis across 10 different subjects or 10 different sectors versus two, because every person now is doing three or four times more, that's an amazing thing. And so it just means that we as a company are capable of doing more, and the thing that we kind of said "Oh, we'll handle that in 2026 or 2027," it's like, no, we can do right now. 01:01:21-01:02:03

Andrew Mayne: Do you have favorite ChatGPT tips or advice you give people on using ai? 01:02:03-01:02:08

Ronnie Chatterji: I have a few I think the coaching is so valuable you know you meet so many people who they say hi I'm a religious ChatGPT user and you find out like they're not even logged in or they don't know about deep research and you're like oh my gosh there's so much more For me, the coaching has been so valuable on diet and fitness. Brad doesn't know this, but I'm training for a big athletic adventure to play basketball at Duke at a Coach K camp. I've requested time off. He's got to approve it. But I got to be in good shape because otherwise I'm going to tear my ACL the first second I get out there. So ChatGPT's helped me over the next four weeks really get in the best shape of middle age. And how is it doing that? It's looking at the food I'm eating and giving me advice and giving me calorie breakdowns. reducing the decisions I need to make by analyzing what I've had that day. And it's helping me track weight and other kind of fitness indicators. And so in doing that, I have this like map out to four weeks, which we're doing really, really hard with the jobs we have and the travel that we're all doing to manage. And so I feel like that's a pretty simple one that you don't need super advanced tools to do, but it's really changed my outlook and made this possible. So that's my favorite one of this month. 01:02:08-01:03:14

Brad Lightcap: The thing I do, especially now with o3 - I think o3 as a model kind of broke through the the barrier for me, it kind of crossed the chasm. All of our earlier models were great, o3 there's something deeply great. And the thing I use it for is to actually like challenge me. A lot of my job is I'm trying to make assumptions about how things work just based on kind of empirical observation of what companies are using us in certain ways, what users tell me they like or don't like. In some ways, like I said early on, our job is to predict the future.

o3 has an incredible ability to actually be a question asker. I think people think of ChatGPT is something that you can only ask questions to, but a lot of times what I really wanted to do is actually ask me questions and challenge my assumptions and make a counter argument to me of why something works or doesn't work the way I think it might work. And it's an incredibly effective thought partner in that regard.

It can be at really big things or it could be low level dumb things. I just got a puppy, and I had dogs my whole life, and we've had a puppy that now is I would say not the easiest when it comes to getting her to calm down and go to sleep. My wife and I could not figure out how to how to get her to do this, and so ChatGPT being a resource for challenging our assumptions about what we thought we knew about puppy training, for example, has been an interesting experience. 01:03:14-01:04:37

Andrew Mayne: o3 is something special. And we talked a lot about what happens when the models can push as well as pull in, when they can kind of get you to think about a thing. And that has been just been amazing. 03 has really been a fun experience talking to. It's the first time I felt like it's not just something that's kind of telling me a thing that it looked up versus something that it thought about. Brad, Ronnie, thank you very much. This has been great. And I hope we can speak again in the future about this and maybe check on the progress of all of this. 01:04:37-01:05:05

Brad Lightcap: Looking forward to it. 01:05:06-01:05:07

Ronnie Chatterji: Thanks for having us. 01:05:07-01:05:07