Inside OpenAI, the Architect of ChatGPT, featuring Mira Murati | The Circuit with Emily Chang

Emily Chang 0:00 [upbeat music]

Emily Chang 0:00 Inside a nondescript building in the heart of San Francisco, one of the world's buzziest startups is making our AI-powered future feel more real than ever before. They're behind two monster hits, ChatGPT and Dall-E, and somehow beat the biggest tech giants to market, kicking off a competitive race that's forced them all to show us what they've got. But how did this under-the-radar startup pull it off? We're inside OpenAI, and we're gonna get some answers. Is it magic?

Emily Chang 0:31 Is it just algorithms? Is it gonna save us or destroy us? Let's go find out. I love the plants. It feels so alive.

Mira Murati 0:41 [laughs] So amazing. I love it here.

Emily Chang 0:44 It's giving me very Westworld spa vibes.

Mira Murati 0:46 [laughs]

Emily Chang 0:47 It's almost, like, suspended in space and time a little bit.

Mira Murati 0:49 Yeah. It has a little bit of futuristic feel.

Emily Chang 0:52 [gentle music]

Emily Chang 0:52 This is one of the most introspective minds at OpenAI. We all know Sam Altman, the CEO, but Mira Murati is a chief architect behind OpenAI's strategy. This looks like the, uh, OpenAI logo.

Mira Murati 1:06 It is. Ilya actually painted this. [laughs]

Emily Chang 1:09 Ilya, the chief scientist.

Mira Murati 1:10 Yes.

Emily Chang 1:11 What is the flower meant to symbolize?

Mira Murati 1:14 My guess is that it's AI that loves humanity. We're very focused on dealing with the challenges, um, of hallucination, truthfulness, um, reliability, alignment of these models.

Emily Chang 1:32 Has anyone left because they're like, "You know what? I disagree"?

Mira Murati 1:35 There have been, over time, people that left to start other organizations-

Emily Chang 1:41 Hmm

Mira Murati 1:41 ... um, because of disagreements on the strategy around deployment.

Emily Chang 1:45 And how do you find common ground when disagreements do arise?

Mira Murati 1:50 You know, you want to be able to have this constant dialogue-

Emily Chang 1:54 Mm-hmm

Mira Murati 1:55 ... um, and figure out how to systematize-

Emily Chang 2:01 Mm

Mira Murati 2:01 ... these concerns.

Emily Chang 2:03 What is the job of a CTO?

Mira Murati 2:05 It's a combination of-

Emily Chang 2:07 Mm

Mira Murati 2:07 ... guiding the teams on the ground, thinking about longer term strategy, um, figuring out our gaps-

Emily Chang 2:14 Mm-hmm

Mira Murati 2:15 ... and making sure that the teams are well-supported to succeed.

Emily Chang 2:19 Yeah. Sounds like a big job. Solving impossible problems.

Mira Murati 2:24 Solving impossible problems.

Emily Chang 2:26 [laughs]

Mira Murati 2:27 Yeah.

Emily Chang 2:27 When you were making the decision about releasing ChatGPT into the wild, I'm sure there was, like, a go or no-go moment. Take me back to that day.

Mira Murati 2:36 We, um, had ChatGPT for a while, and we sort of hit a point where we could really benefit from having more feedback from how people are using it, what are the risks, what are the limitations, um, and learn more about this technology that we have created and start bringing it in the public consciousness.

Emily Chang 2:59 It became the fastest growing tech product in history.

Mira Murati 3:03 It did.

Emily Chang 3:03 Like, did that surprise you? I mean, what was your reaction to the world's reaction?

Mira Murati 3:07 We were surprised by how much it captured the imaginations of the general public and how much people just loved spending time talking to this AI system and interacting with it.

Emily Chang 3:21 ChatGPT can now mimic a human. It can write. It can code. At the most basic level, how does this all happen?

Mira Murati 3:29 So ChatGPT is a neural network that has been trained on a huge amount of data on a massive supercomputer, and the goal during this training process was to predict the next word in a sentence. And it turns out that as you train larger and larger models, add more and more data, the capabilities of these models, um, also increase. They become more powerful, more helpful, um, and as you invest more on alignment and safety, they become more reliable and safe over time.

Emily Chang 4:09 Mm-hmm. OpenAI has kind of turbocharged this competitive frenzy. Um, do you think you can beat Google at its own game? Do you think you can take significant market share in Search?

Mira Murati 4:19 We didn't set out to dominate Search. What ChatGPT offers is a different way to understand information, and you could be, you know, searching, but you're searching in a m- much more intuitive way versus keyword-based.

Emily Chang 4:36 Hmm. Hmm.

Mira Murati 4:36 I think the whole world is sort of now moving in this direction.

Emily Chang 4:40 The air of confidence, obviously, that ChatGPT sometimes delivers an answer with, why not just sometimes say, "I don't know"?

Mira Murati 4:47 The goal is not to predict the next word, um, reliably or safely. When you have such general capabilities, it's very difficult to, um, handle some of the limitations-

Emily Chang 5:01 Mm-hmm

Mira Murati 5:01 ... such as what is correct.

Emily Chang 5:04 Some of these texts and some of the data is biased. Some of it may be incorrect. Isn't this gonna accelerate the misinformation problem? I mean, we haven't been able to crack it on social media for, like, a couple of decades.

Mira Murati 5:18 Misinformation is a really complex, hard problem. Right now, one of the things that I'm most worried about is, um, the ability of models like GPT-4-

Emily Chang 5:31 Hmm

Mira Murati 5:31 ... to make up things.

Emily Chang 5:33 Mm-hmm.

Mira Murati 5:33 We refer to this as hallucinations.

Emily Chang 5:35 Mm-hmm.

Mira Murati 5:36 So they will convincingly make up things, and it requires, you know, being aware and just really knowing that you cannot fully, blindly rely on what, uh, the technology is providing as an output.

Emily Chang 5:50 I wanna talk about this term hallucination because it's a very human term.

Mira Murati 5:53 Mm-hmm.

Emily Chang 5:53 Why use such a human term for basically an AI that's just making mistakes?

Mira Murati 5:59 A lot of these general capabilities are actually quite human-like.

Emily Chang 6:03 Hmm.

Mira Murati 6:05 Sometimes when we don't know the answer to something, we will just make up an answer. We will rarely say, "I don't know." And so, uh, there is a lot of human hallucination [laughs] in a conversation, and sometimes we don't do it on purpose.

Emily Chang 6:23 Should we be worried about AI, though, that feels more and more human? Like, should AI have to identify itself as artificial when it's interacting with us?

Mira Murati 6:33 I think it's a different kind of intelligence. It is important to distinguish output that's been provided by a machine versus another human, but we are moving towards a world where we are collaborating with these machines more and more, and so output will be hybrid.

Emily Chang 6:49 All of the data that you're training this AI on, it's coming from writers, it's coming from artists. How do you think about giving value back to those people when these are also people who are worried about their jobs going away?

Mira Murati 7:03 I don't know exactly how it would work in practice, that you can sort of account for information created by everyone on the internet. I think there are definitely going to be, um, jobs that will be lost and jobs that will be changed as, mm, AI continues to advance and integrate in the workforce. Prompt engineering is a job today. That's not something that we could have predicted.

Emily Chang 7:31 Think of prompt engineers like AI whisperers. They're highly skilled at selecting the right words to coax AI tools into generating the most accurate and illuminating responses. It's a new job born from AI that's fetching hundreds of thousands of dollars a year. What are some tips to being an ace prompt engineer?

Mira Murati 7:51 You know, it's this ability to really develop an intuition for how to get the most out of the model, um, how to prompt it in the right ways, give it enough context-

Emily Chang 8:06 Mm-hmm

Mira Murati 8:06 ... for what you're looking for. One of the things that we talked about earlier was hallucinations.

Emily Chang 8:12 Mm-hmm.

Mira Murati 8:12 And these large language models not having the ability to always be highly accurate.

Emily Chang 8:19 Mm-hmm.

Mira Murati 8:19 So I'm asking the model with a browsing plugin to fact-check this information, and it's now browsing the web.

Emily Chang 8:28 So there's this report that these workers in Kenya were getting paid $2 an hour to do the work on the back end to make answers less toxic, and my understanding is this work is... It can be difficult, right? Because you're reading texts that might be disturbing and trying to clean them up.

Mira Murati 8:43 So we need to use contractors sometimes to scale. We chose that particular contractor because of their known safety standards, and since then, we've stopped working with them.

Emily Chang 8:56 Mm.

Mira Murati 8:56 But as you said, this is difficult wor- work, and we recognize that, and we have mental health standards and wellness standards-

Emily Chang 9:05 Mm

Mira Murati 9:05 ... that we share with contractors.

Emily Chang 9:08 I think a lot about my kids and them having relationships with AI someday. How do you think about what the limits should be and what the possibilities should be when you're thinking about a child?

Mira Murati 9:20 I think we should be very careful in general with, um, putting very powerful systems, um, in front of more vulnerable populations.

Emily Chang 9:33 Mm-hmm.

Mira Murati 9:33 There are certainly checks and balances in place-

Emily Chang 9:36 Mm

Mira Murati 9:36 ... because it's still early, and-

Emily Chang 9:38 Yeah

Mira Murati 9:38 ... we still don't understand all the ways in which this could affect people.

Emily Chang 9:43 There's all this talk about, you know, relationships and AI. Like, could you see yourself developing a relationship with an AI?

Mira Murati 9:54 I'd say yes, as a reliable tool that enhances my life, makes my life better.

Emily Chang 10:00 [upbeat music]

Emily Chang 10:01 As we ponder the existential idea that we might all have relationships with AI someday, there's an AI gold rush happening in Silicon Valley. Venture capitalists are pouring money into anything AI startups, hoping to find the next big thing. Reid Hoffman, the co-founder of LinkedIn and an early investor in Facebook, knows a thing or two about striking gold. He was an early OpenAI backer, and is, in a way, trying to take society's hand and guide us all through the age of AI. I mean, gosh, 12 years-

Reid Hoffman 10:32 Yeah

Emily Chang 10:32 ... we've been talking.

Reid Hoffman 10:33 Maybe longer.

Emily Chang 10:35 Mm-hmm.

Reid Hoffman 10:35 Oh, that's awesome.

Emily Chang 10:36 Yeah, a long time.

Reid Hoffman 10:36 Yes.

Emily Chang 10:37 You have been on the ground floor of some of the biggest tech platform shifts in history, the beginnings of the internet, mobile. Do you think AI is gonna be even bigger?

Reid Hoffman 10:46 I think so. It builds on the internet, mobile, cloud, data. All of these things come together to make AI work, and so that causes it to be the crescendo, the, the addition to all of this. I mean, one of the problems with the current discourse is that it's too much of the fear-based versus hope-based.

Emily Chang 11:07 Oh.

Reid Hoffman 11:07 Imagine a tutor on every smartphone for every child in the world. That's possible. That's line of sight from what we see with current AI models today.

Emily Chang 11:20 You coined this term, blitzscaling.

Reid Hoffman 11:22 Blitzscaling, in its precise definition, is prioritizing speed over efficiency in an environment of uncertainty. How do you go as fast as possible in order to be the first to scale?

Emily Chang 11:34 Does AI blitzscale?

Reid Hoffman 11:36 Well, it certainly seems like it today, doesn't it?

Emily Chang 11:37 [laughs]

Reid Hoffman 11:38 Um, and I think the speed at which we will integrate it into our lives will be faster than we integrated the iPhone into our lives. There's gonna be a co-pilot for every profession.

Emily Chang 11:49 Mm.

Reid Hoffman 11:50 And if you think about that, that's huge.

Emily Chang 11:52 And not professional activities, 'cause it's gonna write my kids' papers, right?

Reid Hoffman 11:55 Uh-

Emily Chang 11:55 My kids' high school papers? [laughs]

Reid Hoffman 11:56 Uh, yes, although the hope is that in the interaction with it, they'll learn to create much more interesting papers.

Emily Chang 12:03 You and Elon Musk go way back.

Reid Hoffman 12:05 Hm.

Emily Chang 12:05 He co-founded OpenAI with-

Reid Hoffman 12:07 Hm

Emily Chang 12:07 ... Sam Altman, the CEO of OpenAI. You and I have talked a lot over the years about how you have been sort of this node in the PayPal mafia. You can talk to everyone, and maybe you disagree, but you are all still friends.

Reid Hoffman 12:17 Hm.

Emily Chang 12:18 What did Elon say that got you interested so early?

Reid Hoffman 12:22 Part of the reason I got back into AI, and I was part of sitting around the table in the crafting of OpenAI, was that Elon came to me and said, "Look, this AI thing is coming." Once I started digging into it, I realized that this pattern, that we're gonna see the next generation of amazing capabilities coming from these computational devices. And then, um, one of the things I w- had been arguing with Elon at the time about was that Elon was, um, constantly using the word robocalypse, which, you know, we as human beings tend to be more easily and quickly motivated by fear than by hope. So you're using the term robocalypse, and everyone imagines The Terminator and all the rest.

Emily Chang 13:02 Sounds pretty scary.

Reid Hoffman 13:03 It sounds very scary.

Emily Chang 13:03 Robocalypse doesn't-

Reid Hoffman 13:04 And I'm like-

Emily Chang 13:04 ... sound like something we want

Reid Hoffman 13:05 ... I'm like, "Yeah, stop saying that." [laughs] 'Cause, 'cause actually, in fact, the, the, the, the chance that I could see anything like a robocalypse happening is so de minimis relative to everything else.

Emily Chang 13:15 So h- you did come together on OpenAI.

Reid Hoffman 13:17 Yeah.

Emily Chang 13:17 How did that happen?

Reid Hoffman 13:19 I think it started with Elon and Sam having a bunch of conversations. Uh, and then since I know both of them quite well, I got called in. Um, and I was like, "Look, I think this could really, uh, make sense." Something should be the counterweight to all of the natural work that's gonna happen within commercial realms. How do we make sure that, uh, one company doesn't dominate the industry, but the tools are provided across the industry so innovation can benefit from startups and all the rest? I was like, "Great," uh, "and let's do this thing, OpenAI."

Emily Chang 13:48 I did ask ChatGPT what questions I should ask you.

Reid Hoffman 13:51 Oh.

Emily Chang 13:51 I thought its questions were pretty boring.

Reid Hoffman 13:53 Yes.

Emily Chang 13:53 Your answers were pretty boring, too.

Reid Hoffman 13:55 [laughs]

Emily Chang 13:55 So we're not getting replaced anytime soon. But clearly this has really struck a nerve. There are people out there who are gonna fall for it.

Reid Hoffman 14:03 Yes.

Emily Chang 14:04 Should we be worried about that?

Reid Hoffman 14:06 Okay, so everyone's encountered a crazy person who's drunk off their ass at a cocktail party who says really odd things, or at least every adult has. And, eh, you know, that's not like the world didn't end, [laughs] right? We do have to pay attention to areas that are harmful. Like for example, someone's depressed, they're thinking about self-harm. You want all channels by which they could get into self-harm to be limited. That isn't just chatbots. That could be communities of human beings. That could be search engines. You have to pay attention to all the dimensions of it.

Emily Chang 14:39 How are we overestimating AI?

Reid Hoffman 14:42 It still doesn't really do something that I would say is original to an expert. So for example, one of the questions I asked was how would Reid Hoffman make money by investing in artificial intelligence? And the answer it gave me was a very smart, very well-written answer that would've been written by a professor at a business school who didn't understand venture capital, right? So it seems smart. Okay, would it, would, would study large markets, would realize what products would be substituted in the large markets, would find teams to go do that and invest in them, and this was all written, very credible, and completely wrong. The newest edge of the information is still beyond these systems.

Emily Chang 15:20 Billions of dollars are going into AI.

Reid Hoffman 15:21 Yes.

Emily Chang 15:21 My inbox is filled with AI pitches. Last year it was crypto and Web3.

Reid Hoffman 15:27 Yes.

Emily Chang 15:27 How do we know this isn't just the next bubble?

Reid Hoffman 15:30 I do think that the generative AI is the thing that has the broadest touch of everything. Now, which places are the right places to invest? I think those are still things we're working out. Now obviously as, as venture capitalists, part of what we do is we try to figure that out in advance.

Emily Chang 15:44 Mm-hmm.

Reid Hoffman 15:44 [laughs] You know, years before other people see it coming. Um, but I think that there will be massive new companies built.

Emily Chang 15:52 It does seem in some ways like a lot of AI is being developed by an elite group of companies and people.

Reid Hoffman 15:58 Mm.

Emily Chang 15:59 Is that something that you see happening?

Reid Hoffman 16:01 In some ideal universe you'd say for, for a technology that would impact billions of people, somehow billions of people should directly be involved in creating it. But that's not ac- how any technology anywhere, anywhere in history gets built, and there's reasons you have to build it at speed, but the question is how do you get, uh, the right conversations and the right issues on the table?

Emily Chang 16:21 So do you see an AI mafia forming?

Reid Hoffman 16:24 [laughs] Um, I definitely think that there is, uh, 'cause you're referring to the PayPal mafia.

Emily Chang 16:30 Of course.

Reid Hoffman 16:30 Uh, I definitely think that there's a network of folks who have been deeply involved over the last few years, um, will have a lot of influence on how the technology happens.

Emily Chang 16:45 Do you think AI will shake up the big tech hierarchy significantly?

Reid Hoffman 16:50 Uh, what it certainly does is it creates a wave of disruption. For example, with these large language models, in search what do you want? Do you want 10 blue links, or do you want an answer? In a lot of search cases you want an answer, and a, and a generated answer that's like a mini Wikipedia page-

Emily Chang 17:10 Mm

Reid Hoffman 17:10 ... is awesome. That's a shift. So I think we'll see a profusion of startups doing interesting things in this.

Emily Chang 17:16 But can the next Google or Facebook really emerge if Google and Facebook or Meta and Apple and Amazon are running the playbook, and Microsoft?

Reid Hoffman 17:24 Do I think there will be another one to three companies that will be the size of the five big tech giants emerging possibly from AI? Absolutely, yes. Now, does that mean that, you know, um, one of them is gonna collapse? No, not necessarily, and it doesn't need to. The more that we have, the better.

Emily Chang 17:43 So what are the next big five?

Reid Hoffman 17:45 Uh, well, that's what I'm, we're trying to invest in.

Emily Chang 17:46 [laughs] You're on the board of Microsoft.

Reid Hoffman 17:49 Yes.

Emily Chang 17:49 Obviously, you know, Microsoft is making a big AI push. Did you bring Satya and Sam or have any role in bringing Satya and Sam closer together? Because Microsoft obviously has $10 billion now in OpenAI.

Reid Hoffman 18:01 Um, well, I think I could, uh, I probably have a ... You know, uh, both of them are close to me and know me and trust me well, so I, I think I, I've helped facilitate understanding and communications.

Emily Chang 18:14 Elon left OpenAI years ago.

Reid Hoffman 18:17 Mm.

Emily Chang 18:17 And pointed out that it's not as open as it used to be. He said he wanted it to be a non-profit counterweight to Google. Now, it's a closed-source maximum profit company effectively controlled by Microsoft. Does he have a point?

Reid Hoffman 18:31 Well, he's wrong on a number of levels there. Um, so one is it's run by a 501 [c] [3] . It is a non-profit. Uh-

Emily Chang 18:39 But it does have a for-profit part.

Reid Hoffman 18:40 The commercial system, which is all carefully done, is to bring in capital to support the non-profit mission. Now, get to the question of, for example, open. So, um, Dall-E was ready for four months before it was released.

Emily Chang 18:55 Mm-hmm.

Reid Hoffman 18:56 Why did it delay for four months? 'Cause it was doing safety training. It said, "Well, we don't wanna have this being used to create child sexual material. We don't wanna have this being used for assaulting individuals or, or doing deepfakes. So we're not gonna open source it. We're gonna release it through an API so we can be seeing what the results are and making sure it doesn't do any of these harms." So it's open because it has open access to the APIs, but it's not open 'cause it's open source.

Emily Chang 19:21 There are folks out there who are angry actually about OpenAI's branching out from non-profit to for-profit. Is there a bit of a bait and switch there?

Reid Hoffman 19:31 The cleverness that Sam and everyone else figured out is they could say, "Look, we can do a market commercial deal where we say we'll give you commercial, um, licenses to parts of our technology in various ways, and then we can continue our mission of beneficial AI."

Emily Chang 19:46 The AI graveyard is filled with algorithms that got into trouble.

Reid Hoffman 19:51 Mm, mm.

Emily Chang 19:51 How can we trust OpenAI or Microsoft or Google or anyone to do the right thing?

Reid Hoffman 19:59 Well, we need to be more transparent. But on the other hand, of course, our problem, exactly as you're alluding to, is people say, "Well, the AI should say that or shouldn't say that." We can't even really agree on that ourselves, so we don't want that to be litigated by other people. We want that to be a social decision.

Emily Chang 20:17 So how does this shake out globally?

Reid Hoffman 20:19 We should be trying to build the industries of the future. That's what's the most important thing, and it's one of the reasons why I tend to very much speak against the people who are like, "Oh, we should be slowing down."

Emily Chang 20:32 Do you have any intention of slowing down?

Mira Murati 20:34 We've been very vocal about these risks for many, many years. One of them is acceleration. Um, and I think that's a significant risk that we as a society need to grapple with. Building safe AI systems so that are general is very complex. It's incredibly hard.

Emily Chang 20:55 So what does responsible innovation look like to you? You know, like, would you support, for example, a federal agency like the FDA that vets technology like it vets drugs?

Mira Murati 21:05 Having some sort of trusted authority that can audit the systems based on some agreed-upon principles, um, would be very helpful.

Emily Chang 21:19 I've heard AI experts talk about the potential for the good future versus the bad future. And the bad future-

Mira Murati 21:25 Yeah

Emily Chang 21:25 ... there's talk about this leading human extinction.

Mira Murati 21:27 Mm-hmm.

Emily Chang 21:28 Are those people wrong?

Mira Murati 21:29 There's certainly a risk that, um, when we have these AI systems that are able to set their own goals, they decide that their goals are not aligned with ours and, uh, they do not benefit from having us around and could lead to human extinction. That is a risk. Um, I don't think this risk has gone up or down from the things that have been happening in the past few months. Um, I think it's certainly been quite s- uh, hyped.

Emily Chang 22:04 Mm.

Mira Murati 22:04 And there is a lot of anxiety around it.

Emily Chang 22:07 If we're talking about the risk for human extinction, have you had a moment where you're just like, "Wow, this is, this is big"?

Mira Murati 22:14 I think a lot of us at OpenAI, um, joined because we thought that this would be the most important technology that hu- humanity would ever create. But of course, the risks on the other hand are also pretty, uh, significant.

Emily Chang 22:32 Mm.

Mira Murati 22:32 And this is why we're here.

Emily Chang 22:35 Do OpenAI employees still vote on AGI and when it will happen?

Mira Murati 22:38 [laughs] I actually don't know.

Emily Chang 22:41 What is your prediction about AGI now, and how far away it really is?

Mira Murati 22:45 We're still quite far away from, uh, being at a point where, you know, these systems can make decisions autonomously, um, and discover new knowledge. But I think I have more certainty around the advent of having powerful systems in our future.

Emily Chang 23:07 Should we even be dri- driving towards AGI, and do humans really want it?

Mira Murati 23:11 Advancements in society come from pushing human knowledge. Now, that doesn't mean that we should do so in, um, careless and reckless ways. I think there are ways to guide this development versus bring it to a screeching halt because of, mm, our potential fears.

Emily Chang 23:33 So the train has left the station, and we should stay on it?

Mira Murati 23:35 [laughs] That's one way to put it. [upbeat music]