Hello and welcome to the global dialogue. I'm Shireen Bhan. 2024 was a breakout year for OpenAI's ChatGPT and 2025 promises to be equally exciting. Joining us now to talk about the innovation pipeline, what we can expect as far as the AI war and the AI race is concerned is Kevin Weil, the Chief Product Officer at OpenAI. Kevin, many thanks for joining us. I know it's late in the evening for you and it's been a busy week, an exciting week with lots of launches lined up including tonight where you had the o3 and the o4 mini. So let me start by asking you Kevin to give us a snapshot in very simple language, what does ChatGPT 4.1 mean for OpenAI and for consumers and how do o3 and o4 mini change the game?
Yeah, well first of all thank you for having me. I really appreciate it. Excited to be here with you. India is an incredibly important market for OpenAI and ChatGPT. It's our second largest market. It's our fastest growing market and it's also the market where we saw our image generation just take off. But we had some great launches this week. You mentioned GPT 4.1 which was a model for developers that is fantastic at coding and complex instruction following. And so for that, that was a model that we released focused on developers. And we're seeing them, you know, we have over three million developers on the OpenAI platform and they're doing amazing things. That is one of our fastest growing models by usage ever. And then just this morning in the US we launched o3 and o4 mini which are our frontier reasoning models. They're the smartest, fastest, safest models that we have ever launched. They're incredible at things like computer programming. So if you look at measures like on competitive coding benchmarks, these models would be better at competitive coding than all but about 200 people in the world. And they can answer hard PhD level science questions better than 80% of PhDs in their field across all fields simultaneously. So they're really very intelligent models. But you can also use them for more mundane tasks, understanding medical jargon or looking at what's happening in the world, creating images, helping you with writing, translating languages. These models are so dynamic and we're super excited to see what people can do with them.
Well, thanks very much for giving us that update, but I want to link this back to comments that you've made previously, including a tweet that I saw from you about 10 hours ago where you talk about the fact that these are the worst AI models that you'll ever use for the rest of your life as the Chief Product Officer at OpenAI with things changing so quickly. And you, in your words, you believe that these are the smartest models to date. How does that get you to kind of constantly reinvent yourself, constantly disrupt yourself? I mean how concerned are you not just about competition but about creating moats that continue to give you the advantage that you enjoy today?
Yeah, it's one of the most exciting things about this industry. These are simultaneously the most intelligent models in the world, the best that we've ever created, we believe the best in the world, and they're the worst that we will ever use because the models are progressing and getting more intelligent, getting safer, getting smarter. Every few months there's a new breakthrough and all of a sudden computers can do something that computers have never been able to do before. You know, if you go back two years or so to when ChatGPT launched, if I asked you to use those models at the time, they were amazing. They were mind-blowing. If I asked you to use those models now, you would be disappointed. And that's just because of the pace that everything is getting better. We've kind of whooshed past the Turing test, which used to be the high watermark for, you know, is AI here and now that's in the distant past. Every AI model is passing the Turing test. So it's an exciting time. And the fact that ChatGPT has gone from zero users a couple years ago to over 500 million active users today just shows how valuable these things are in people's lives.
Yeah, I want to double down on this Kevin because you've also said that models are getting smarter, faster, cheaper, safer, more personalized and more helpful. The future is bright. You know that's a lot to pack in. If I were to ask you what you prioritize today, when you look at how you develop and build the next line of products, are you going to be prioritizing making this much more accessible, i.e. cheaper given the kind of breakthroughs that we've just seen with DeepSeek? What do you prioritize in the attributes that you've outlined that models will need to focus on?
I think first and foremost I care about making these models more broadly accessible and more useful to people. And that entails making them smarter and safer and things like that. But first and foremost, I want ChatGPT to help people live better lives, to help them accomplish their goals more quickly, to help them learn new languages, to help them learn new skills, to help them do the kinds of things they need to do for their family or their job faster so that they can spend more time on the stuff that they really care about. These models are, and ChatGPT helps me do this every day, as we think about building models for developers, for programmers, for entrepreneurs who are running startups or working at companies. One of the things that we care a lot about is reducing the cost of our models even as they get smarter. If you compare today to two years ago again when we launched GPT-3, GPT-3 was more than 100x more expensive than some of our models today and our models today are smarter and faster. And so I talked before about how rapidly this industry is progressing. You can do so much more today with an AI model than you could two years ago. And it's cheaper and it's faster and it's safer. And so that just affords an incredible set of opportunities for people, for businesses, for developers all over the world and we're very proud and humbled to be able to help provide that.
But Kevin, how much cheaper can we expect things to get? In the near term, what are we talking about?
Well, so I said two years ago our flagship model was more than 100x what many of our models today cost. So you're talking about maybe 10x a year in price drop for sort of a constant level of intelligence. So the same level of intelligence in a model maybe gets 10x cheaper every year. And that's, I mean, Moore's law which has transformed the world and powers our lives, you know, in every little thing that we do every day has a transistor in it. Moore's law was about the number of transistors in a chip doubling every 18 months and that changed the world. AI is on a trajectory of going down by a factor of maybe 10x in cost every year. So it's a much steeper exponential than Moore's law, which means I think we're going to see AI transform the world even faster. And that's exciting. I'm very optimistic about that.
You know, you're talking about AI transforming the world, and I know previously you've spoken about the fact that you believe AI will surpass human coding perhaps even by the end of this year. To parents and to kids who are looking at grad school, at people who are graduating from engineering colleges today. I mean, you talk about a bright future, but as they look at the expanse of what's going on and the disruption and transformation that's expected to happen, what would you tell them? Because it almost sounds like you're writing the obituary in a way for coders.
I actually don't think so at all. You know, jobs evolve. They don't evaporate. If you go back and look at, I watched the movie Hidden Figures recently which is about some of the rocket scientists from back in the 50s and 60s. They were doing longhand calculations on paper to calculate rocket trajectories. And today that seems wild that you would have people doing it, right? Computers are just better than humans at doing hard calculations. And so of course computers do all of that for us. But that doesn't mean that there aren't rocket scientists today. And in fact, we've probably made far more progress if you look at things like SpaceX and also things like, you know, the ISRO, the Indian Space agency, which is an incredible space agency, one of the premier rocket launching agencies in the world. I think we're far better off because we have computerized these things and computers do it better than humans do. I think that will be true of code as well, but that doesn't mean that it won't pay to be technical, that it won't pay to have a computer science degree. I think these skills will still be important. They're just going to evolve.
Well, speaking of evolution, let's talk about what you're seeing in India. You talked about how it is today your fastest growing market for ChatGPT. How many active users do you have in India? What surprised you about the adoption, the rate of adoption, the pace of adoption in India? Why do you believe you've seen this kind of an uptick here?
Yeah, I mean our active users in India tripled, more than tripled in 2024. So it's been incredible to watch. A few weeks ago we launched a new image generation tool using AI to generate incredibly, you know, images that you can control to a great degree whether you want them to be vivid and lifelike or cartoony or following a particular style. We saw Prime Minister Modi tweet using our image gen feature. We had cricketers like Sachin Tendulkar tweet images from ChatGPT and it just took off. I mean unlike anything I've ever seen. I've been fortunate to be at Twitter for many years. I led product at Instagram for a number of years. I mean these companies were growing very quickly. I have never seen anything grow like ChatGPT was growing because people were having so much fun creating these images, sharing them with their friends, on WhatsApp and in other places. And it's just incredible to see. So, we marshaled a whole lot of resources. We really tried to support it especially in India. We released new features around our image gen like image galleries that let you quickly view all of the images that you've created. There's just a huge amount of excitement in India for ChatGPT and not just for this image gen feature but also ChatGPT speaks a hundred languages, it can teach you any subject you want to learn, it can save you time at home and at work. And so it's just so broadly useful and it's completely free, you don't have to pay us a dime, you can use it on an Android phone for free. So it's just, we deeply believe in bringing AI to the world and the positive impacts that that can make. And I can't think of a better example than what's happening in India right now with ChatGPT.
Well, you know you talked about the Ghibli-fication of India and we've seen that go viral as you just explained to us but I want to link this to the broader issues that are being debated at this point in time. Issues related to copyright, issues related to data privacy, is this a great way of data capture for OpenAI? That's the question that's being asked as well. And Sam Altman has said that he believes it's a net win for society but there are negatives. I want to ask you this as the Chief Product Officer when you're developing and designing products, you do need to factor in in his words the negatives as you look at where you are today in your innovation cycle. How much more of that is a focus area for you?
