Plain English with Derek Thompson - Inside the OpenAI Meltdown

Episode Date: November 21, 2023

Today’s episode is about whatever the hell just happened—is still happening—at OpenAI, where CEO Sam Altman has been fired, almost rehired, and then shipped off to Microsoft, while the most famous startup in artificial intelligence self-immolates for reasons that the company refuses to explain. Our panel has some theories. Charlie Warzel is a staff writer at The Atlantic who has been texting and talking with OpenAI employees for the last few days. Karen Hao is a contributing writer at The Atlantic who is writing a book about OpenAI and knows many of the main characters from this past weekend. Ross Andersen is a staff writer at The Atlantic who wrote a big magazine feature on Sam Altman, OpenAI, and the messy quest to build artificial general intelligence. If you have questions, observations, or ideas for future episodes, email us at [e-mail].  Host: Derek Thompson Guests: Ross Andersen, Karen Hao & Charlie Warzel  Producer: Devon Manze Learn more about your ad choices. Visit podcastchoices.com/adchoices

What would you do if everyone said they heard your trailer a hundred times? You'd probably make a new one. I'm Justin Sales, the host of The Wedding Scammer, the ringer's first ever true crime pod. We've been hunting a con man for a few weeks now, and our hunt is coming to an end. Schemes, Heartbreak, How to Put On a Wire. We've covered all this and more, but there are still a few surprises left. Binge the Wedding Scammer wherever you get your podcasts. Well, you knew this one was coming.

Today's episode is about whatever the hell just happened slash is still happening at OpenAI, where CEO Sam Altman has been fired, then almost rehired, then shipped off to Microsoft, while the most famous startup in artificial intelligence self-immolates for reasons that nobody can explain. The story, by the way, seems to change every 90 minutes or so. So I'm going to start us off with the acknowledgement that I am recording this open. at 4.22 p.m. ET on Monday. So first, a quick recap of the last 90 to 100 hours. On Friday afternoon, OpenAI fired its chief executive Sam Altman,

accusing him of, quote, not being consistently candid with the board of directors. Altman was fired without warning over Google Meet, with four of the six members of his nonprofit board of directors voting to oust him, much, much more on that board of directors and corporate structure in a second. On Sunday, this weekend, at the urging of Microsoft, an investor, OpenAI invites Sam Altman back to the office to discuss the prospect of rehiring him as CEO, only to say a few hours later, nope, never mind, we don't want you back. And instead, the interim chief executive of the company will now be an outsider, Emmett Shear,

the former CEO of Twitch, a live video streaming service. Early Monday morning, Microsoft Chief Executive. Saty Nadella announced that he would hire Altman along with other Open AI workers to start a new AI research division within Microsoft. Later on Monday morning, hundreds of employees at Open AI signed a letter demanding that Altman returned to their company as CEO and demanding the resignations of all the board members who fired Altman. The Verge reports that the vast majority of the company has now signed this letter. So it is a handful of non-employed employee board directors versus the majority of the company.

Meanwhile, the big winner, Microsoft. In tech, there's this idea called an aqua hire, where you buy a company for its best employees rather than its products. Microsoft might be on the verge of the most successful aquaire in corporate history. If the entire employee register of OpenAI switches its W-2s to Microsoft Inc. Without Microsoft having to pay anything to actually acquire a company. that is currently valued in $86 billion. I would say that qualifies as a big win.

Big loser number one, anyone who thought it was a good idea to build open AI outside the shadow of big tech, and there were lots of reasonable people who thought it was a good idea to build a artificial intelligence research lab outside the shadow of the largest tech companies that would rush to commercialize it,

well, now OpenAI is poised to be swallowed whole by much, Microsoft, the second largest company in America. I'd say that group is not feeling so good about itself right now. Big loser number two. Whoever the hell thought it was a good idea to create a corporate structure that allows three non-employee board directors to implode an 800-person company. Again, in the next few minutes, next hour of this podcast, the corporate structure question,

will be revisited very often. So why did this happen? Why did this happen? Why was Sam Altman fired? The amazing thing is that the board still has not elaborated. The C.O.O. of the company says it had nothing to do with, quote, malfeasance. The new CEO of the company says it has nothing to do with, quote, AI safety. The company has been crystal clear about the 10, 11, 12 reasons why they didn't fire Sam Altman,

but they haven't said one specific thing about why they did fire him outside of the remarkable vague phrase not being consistently candid. Today we take you inside OpenAI to tell you what's going on there right now, but also to give you the full eight-year history of this organization, how it was founded and how its very founding played a starring role in this weekend's mess. Today we have an all-star panel. Charlie Warzel is a staff writer at the Atlantic who has been texting with and talking with Open AI employees for the last few days.

Karen Howe is a new contributing writer at The Atlantic. She was also speaking with Open AI employees for the last few days and is writing a book about OpenAI. She knows many of the main characters of this weekend's fracas very well. And Ross Anderson is another staff writer of the Atlantic who wrote a big magazine feature on Sam Altman, OpenAI, and the messy quest to build artificial general intelligence. In fact, Ross followed Sam Altman.

and some of the main characters from this weekend's mess around the world as they talked about the promise of chat GPT to various countries in Asia, we will see about the promise of chat CBT in a world that may not include an open AI. I'm Derek Thompson. This is plain English. Karen, welcome to the show. Thank you so much for having me. Charlie, welcome back.

Always good to be here. And Ross, welcome back. Couldn't wait to be on today, Derek. Yeah, thank you all for helping me sort through whatever the hell is going on at OpenAI. Thank you for taking a break in your reporting schedules. I'm sure you're being slammed with an absolute cavalcade of incoming text and emails and pieces of rumor that you have to piece through. So thanks for spending a little bit of time with me to piece it out.

I think the best way to tell this story is to tell it completely, and that is to say to tell it chronologically. I want to start with a brief history of OpenAI because I strongly believe that what's happened in the last 100 hours. actually has everything to do with the way this company was set up in 2015, and then how this evolution has sort of come into conflict with the principles of its founding. So, Ross, let's start with you. A lot of people know OpenAI today as the maker of ChatGPT and as one of the most valuable startups in the world, but it was founded in 2015 as a non-profit. Why?

Yeah. So, you know, the story of its founding is really interesting. And as I understand it, it was sort of smeared across a couple weeks or even months, but there's come to be the sort of mythology around it that most of the action was at this famous dinner at the Rosewood, the Sandhill Rosewood in Silicon Valley. And the idea was that Sam Altman, star of this weekend's news cycle, and Elon Musk convene this dinner with some tech luminaries and AI luminaries, including Ilya Sutskeber, who will hear more about. later, and that they all sort of agreed that artificial intelligence had kind of progressed to a point where several inflection points were likely, and that very soon, you know, artificial

general intelligence might be imminent. And that if this was going to be done, it would be a bad thing if it were one of the tech giants that did it, because they'd be greedy monsters and something so important and so transformative and so potentially threatening, you wouldn't want in their hands. And so they decided to start the company as a nonprofit so that they could create artificial general intelligence for the benefit of all humanity. And can you briefly, in like 15 seconds or less, describe what you think they think they meant

when they said artificial general intelligence? Yeah, you know, talking to Altman about it, his sort of go-to definition, it seems to me, seems to change by the month. but one of the more durable recent ones that I've heard is that you can imagine, first of all, instead of a narrowly skilled artificial intelligence that can just play chess, this is a general intelligence, an intelligence can do lots of things. And you might think of an AGI from when I introduce the acronym as like a really smart

college student that can kind of do any number of intellectual tasks that people might do. Karen, after a few years, this nonprofit comes to a few realizations. And one of those realizations is that donations alone under this nonprofit model won't suffice. The cost of computational power is too high. The cost of talent paying all of these brilliant engineers to build God is too high. They decide that to realize their mission, they have to create a for-profit subgroup within OpenAI that can raise capital and build what they have to build. but yet that for-profit subgroup is still going to be nested underneath the nonprofit OpenAI board. Do I have that right?

Maybe talk a little bit more about what this invention of the sort of for-profit subsidiary really meant for open AI. Yeah, that is exactly right. Just to take a step back before diving back into Open AI, I do think it's also important to conceptualize Open AI within the history of AI research and understand that like, The bet that Open AI made in AI development was one very particular bet across like a landscape of different options that they could have taken. And the reason why they realized that they needed a lot of money and they needed a lot of computational power to build this thing is specifically because of the research philosophy that Ilya Sutskever had, which is that he was 100% betted in on what's called deep learning, these systems that. used data to learn and get behaviors, and was 100% in on scaling this thing. So if you specifically believe that you're going to get AGI through scaling these systems, that's when you need a lot of money.

And when they made the decision within Open AI and they started mapping out the costs at the time, they realized that the nonprofit just wasn't going to cut it. And at the time when I was interviewing Brockman and Sutskevra about this in 2019, shortly after the Microsoft billion dollar investment, they had mentioned that they had actually tried to maintain it as a nonprofit. They had shopped around and tried to get more investment. This was also the moment in time when Musk had disagreements about the direction of Open AI and left and reportedly took his money, the money that he had committed away from them.

So it wasn't just that they needed to fundraise more money. It was that they just needed to fund raise money because they potentially didn't have that much anymore. So that's why they were sort of coming up with these ideas. And there had been an idea floated on early within the company about having sort of a for-profit arm that would allow them to raise venture funding. That idea then was raised again. it seemed to be the only model that was working, given that one billionaire pulled out, other billionaires weren't pitching in. And so then they created the for-profit entity, but on the condition that it was supposed

to be governed by the nonprofit because I think the co-founders of OpenI still believed in this idea that they didn't want to just become any other tech company based on their own metrics of whatever that means. Karen, can you help me understand, even briefly, the little fights that have emerged in the history of Open AI? Because you just mentioned that this is a company, this is a nonprofit that was founded with Elon Musk at the head of the table, but he's one of the first and loudest people to leave. And then later you have some of the biggest people in OpenAI leave to start this competitor called Anthropic. The history of Open AI has been a history of conflicts within Open AI, even before this past weekend. Can you just summarize the most significant conflicts of the last few years and what they were about?

Yeah, absolutely. There were two main conflicts. And all of them are about, and it's the same with this weekend, all of them are about power. Like every single person that's sort of been at the helm of Open AI or invested in Open Eye or controlled, whatever, has been. trying to seek to perpetuate their specific ideology around how AI development should be, should progress. And some of them do it under the banner of capitalism. Some of them are doing it under the guise of morality or, you know, humanity. But ultimately, they're all trying to centralize their control over this technology. So the first real conflict was when Musk became very upset with direction of open eye, AI. And, you know, you. You know, there's been sort of a lot of reporting around why exactly he was upset.

Some of it seemed to be that he was not happy with the pace at which Open AI was getting breakthroughs. And Open AI was sort of being overshadowed by DeepMind at the time. And Musk had specifically invested in Open AI because he wanted to upstage DeepMind. So then he... Which is owned by Google and Alphabet. Yes, which is owned by Google and Alphabet and had been the first lab to sort of start going after. this AGI goal. So then Musk leaves.

And then in the late 2020, there was then a pretty substantial faction of the company that also left with Dario Amode as sort of like the main person that led this contingent. And they went and founded Anthropic, which is now one of the rivals of OpenAI. And it was also kind of this split between like techno optimists and Dumer's. Jario is much more on the Dumer side of the spectrum, felt very uncomfortable with the way reportedly that Altman and Bruckman were sort of driving ahead with trying to commercialize a technology that he felt was not ready. So he decided to create his own company. And of course, some people who know Dario that I've spoken to in the past have said that it is both.

It's still about power and control. Like, on one hand, yes, he had this ideological disagreement, but on the other hand, he also specifically just, he wanted to be the CEO of a company that was driving forward AI development. So that leads us, I mean, very neatly into this weekend, which is sort of the same thing about, like, who gets to control this technology, who gets to call the shots, and which ideology ultimately is imprinted on this technology

that's going to affect everyone. So it's already a messy history leading up to last November. So just summing up, Open AI is set up as a nonprofit. Their strategy for building artificial general intelligence requires so much treasure and talent that they decide to create a for-profit subsidiary that they can use to issue equity and raise a bunch of capital. There's all sorts of ideological disagreements between Altman and Musk and Altman and Dario.

People leave the company, come into the company. Microsoft becomes a major investor starting in 2019, where they invests initial $1 billion in OpenAI. Charlie, I think this leads us to the invention of Chatsy, in November of 2022. You and Karen did some reporting over the weekend and discovered that the invention of Chat Chbett really changed things inside the company.

How did it change things in ways most relevant to the chaos of the last 100 hours? Yeah. So it's really interesting. I think few companies have created a product that is so successful that it leads to large internal struggles. I mean, maybe you could say that about Facebook or some others,

but it really is fascinating how this played out. So essentially, what people need to know about chat GPT is that it was not expected to be as much of a success. The sort of phrase that they had internally for it when they were talking about it was that it was going to be this, quote-unquote, low-key research preview, right? It was going to be this proof of concept, this, this, this, you know, application, essentially, of their really powerful technology. And this was around the same time that GPT-4 had been trained

and was being developed. And the idea around, you know, the company was that GPT-4 was going to be the star of the show, the thing that when it rolls out is going to sort of changed the landscape, right? And then chat GPT, which was, you know, which was initially trained off of a different model GPT 3.5, so to speak, that it was going to be, you know, a thing that, you know, one of my sources said it would sort of ping pong around AI Twitter for a weekend and then fizzle out, right? And so there was actually, we reported like a betting pool.

in the company as to whether or not this, you know, this technology was going to, you know, how many people were going to use it in the first week? And the highest guess was 100,000 users. So obviously, they had no idea that it was going to be this runaway success. Obviously, within the first five days, a million people used it. And, you know, it sent the company into essentially turmoil, right? And some of that is just logistical turmoil, the capability of, Do we have enough servers, enough computing power to handle the traffic, you know, the glut of this? And then, you know, as we reported, there's the issue of, you know, moderation concerns, how people are going to use it, whether or not, you know, there's enough staff, whether or not the, you know, they had thought through all the different externalities of what a conversational model could do. And there was a rush in getting chat GPT out into the public because, as we talked about, there are these rivalries in OpenAI and rivalries with, you know, companies like Anthropic.

And Anthropic was on the verge of coming out with their own conversational chat model. So this was all deployed very quickly. This was, you know, it's not fair to say a rush job, but everything happened fast and nobody expected this. So the commercial success of a tool like this, the way that it leads directly to what we see this weekend, is it's commercial, it's popular. It's something that helps Sam go out and demonstrate the power. I mean, in a way, you know, you had some of these image generation tools like Dolly

and Mid Journey that sort of demonstrated the power of generative AI. But ChatGPT, you know, switch that paradigm. on its head and sort of, I think, brought Open AI to the four as, you know, a sort of a generational company. It made those conversations about something like AGI a lot more relevant. Obviously, everyone in the media talks about it, et cetera. So the way that that sets this up is essentially there is a great case inside the company to raise more money, to get, you know, companies like Microsoft on.

board to continue to move forward with commercial products to, you know, turn chat GPT into an enterprise tool, to open it up to developers, things like that. And by doing that, it strays more from this, you know, nonprofit style mission, right? Of we are a, you know, a research based almost think tank for the betterment of humanity. Now we're saying, actually, we're a Silicon Valley tech company and we need to grow. We need to hire people, et cetera. So, you know, that really sets up this this internal division between, you know, the people who are absolutely terrified of something like, you know, building out a godlike technology or a technology that, you know, could be used to engineer bio weapons or, you know, you name the apocalyptic

scenario versus the people that are saying, you know what, we have to raise money, we have to build this thing out. And by doing so, we're going to, you know, make it possible that we can safely develop or maybe even not safely develop a, you know, further models and work our way towards an AGI. Ross, before we finally analyze the events to the last few days, I think we have to talk about Sam Altman and who this guy is. And what Charlie just said reminds me of what I find most interesting about his relationship with his own company, his own creation, which is that Open AI is established as a kind of

think tank as a kind of think tank on steroids. And you look at Sam Altman, and this is a guy who has a nose for commercialization. He used to run Y Combinator. He is an expert at turning ideas that you might come up with in a think tank into products that are worth billions and billions of dollars. So you spent all this time with him on your international tours for an Atlantic magazine feature. Tell me about what it's most important for us to know about Sam Altman. who he is and what his ambition looks like. Yeah, well, you said it well when you look in his pedigree. You know, someone who runs Y Combinator, you know, it's interesting.

This has been, we've had lots of these great coinages, Dumers versus Boomers. Another one is like accelerators versus D-cells, right? And like, not to read too much in names, but Y Combinator was like the biggest and most prestigious startup accelerator in Silicon Valley, right, for many years, probably still is. And so this is someone whose brain thinks about how do you take an idea or some kind of nascent technology and scale the heck out of it? And so you could only imagine, you know, previously I think Sam was probably

riding along just fine with kind of like, oh, this is a research lab. He had his hands of a lot of pods. he was doing different things. But once ChatGPT hit, and you had that enormous boost in cultural cachet for him and the kind of opportunity to put OpenAI sort of on a plane with a Google or a Microsoft or an Amazon, I imagine that was irresistible for him.

And certainly, I mean, being with him in East Asia, you know, he really did seem to thrill to meeting with heads of states to the sort of waves of agility, that he received everywhere he went. I mean, he's been on one. And I think that once he get a taste of that, I mean, no disrespect to him. Like, I mean, it's kind of a tempting situation for anyone,

especially someone who has that kind of background who thinks about actualization being scaling your startup. All right. I think we've adequately set the table for this weekend. Karen, why don't you grab the reins here? It's Friday afternoon. you hear that Sam Altman has been fired without warning. Without warning to Sam Altman,

without warning to Microsoft, investors, employees of the company have no idea this is coming. You and Charlie start calling around, what do you learn about why Altman has been fired and what the sort of pre-existing dynamic within OpenAI has been that might have made us firing more plausible? We still have no idea. idea why Almond was fired. Let's be clear. No one knows why he's been fired. The board is not

talking. Sources are not talking because they have no idea. You know, like, there's only four people that really know, which is the three members of the board and like Altman himself maybe because they were the only ones in the room. And just to name them, I'm sorry to jump in there, but just to name them, right, it's the three non-employees of OpenAI, Adam DeAngelo, the CEO, Quora, the question and answer site, Tasha McCauley, an entrepreneur, and Helen Toner, who's a professor at Georgetown, who's part of the more pro-safety tribe of the AI breakdown. The one person who joined them from OpenAI is Ilya Sutskiver, the chief scientist. Sam Altman and Greg Brockman, also members of the board. It's a six-person board, but Sam, obviously, fired, and Greg

left hours after the firing. Sorry to jump in there, but just want to make sure that people have a sense of what the makeup of the board is. Keep going. So the thing that, like, Charlie and I were talking about is, I mean, you know, like, there are there, we're going to find out sort of the play-by-play of is he or isn't he going to be like going back to opening eye. What is the whatever? But really, the more important context here is actually what happened in the lead-up within the company that might help clean, help us glean some insight into what on earth happened. That might. might have come to a head suddenly with this very dramatic firing. And as Charlie was saying that there was this kind of like cataclysmic shift that happened at opening eye when Chad GBT became the success, the very unexpected success that it did. And it essentially started polarizing the extreme factions that already existed in the company that have existed in the company since the beginning and have caused each wave of the drama

that we've talked about. But in this moment in time, it's sort of really amplified because A, opening eye is extremely in the public eye now. So when the opening I anthropics split happened, like no one really knew who they were, no one really cared. And they hadn't really developed products that were so consequential to businesses, to consumers, to heads of state in terms of trying to understand the future of their country, their workforce and things like that. And so I think when ChatGBTGBT became such a success, each of the camps within Open AI instantly became way more hyped up about trying to set the agenda for their vision. So like the product side of the company just became really, really amped up about let's continue building on this momentum. And then the safety side got way more freaked out. And in part, I think there was probably.

kind of like a cyclical feedback loop where as more people outside of OpenA also started becoming either really hyped up or freaked out, that like that also reflected within like the company where these people felt like they had to speak up for the rest of humanity and kind of like dictating which direction the technology should go. And so that trickled all the way to the top. Like Altman and Brockman, they very much are of the Silicon, Valley techno-optimist mindset. They both sort of, as we already mentioned, as Ross already mentioned, Allman was leading the most famous accelerator in Silicon Valley. Brockman was at Stripe, which was a YC-combinator, a YC-funded company and had done his entire career at Stripe before opening eyes.

So they're both very like startup YC investor-minded people. And then you have Ilya, who is a research scientist had sort of been steeped in the academic world for a really long time before doing like a brief stint at Google. And then he becomes sort of really allied with the safety side, the more that he feels confident that AGI is going to arrive soon. And like the people, the sources that we talk to sort of talked about how like he started becoming almost a spiritual leader with like really focusing people on remember the mission, feel the agee, like we need to shepherd this like generational technology forward in like with the utmost kind of morality.

And so that clash really kind of came to a head. And what we're seeing is kind of the symptom of the underlying power struggle that was happening where each of these factions were really trying to go in just like completely opposite directions. Charlie, you and Karen have said that it's the safety side of the company that's been bristling at the reality that ChatchipT has started to eclipse the original purpose of OpenAI. I wonder, do you guys have a sense of how big that safety side is, especially aside from these three members of the board who voted to fire Sam Altman? Because there's other pieces of evidence that I've seen that suggests that the vast majority. of OpenAI employees want Sam Altman to stay and are furious at the board for firing him.

So there's been an open letter that's been signed by reportedly hundreds of employees demanding that Sam Altman stay and that the board essentially eject itself from the company. The Verge reported today, quote, there has been a nonstop power struggle inside Open AI since Friday with nearly all employees against the now three-person board that opposes all. Bultman, employees at the company, San Francisco headquarters, refused to attend an emergency all-hand scheduled on Sunday with the new CEO, according to a person familiar with the matter, and added, they respond to the announcement of Open AI Slack with a fuck-you emoji, end quote. So I guess my question is, there's no question, I have no doubt that this safety contingent exists

and that it is clearly powerfully represented on the board. But do you have any sense of how big it is relative to the overall size of open AI? right now. So with the caveat that, you know, we, the reporting is, is trickling out pretty slowly and that there's, you know, there's a huge subset of this company that's sitting and waiting, you know, for their fate to be decided as to, like, whether they work at Open AI or whether they're going to be employees of Microsoft, I want to say that, you know, it's not like we have some sort of God view inside there. But anecdotally, I think that this, the size of this has changed. over the weekend. I think there are people who were maybe more in the camp of being safety-minded

or split, right, or, you know, really enjoyed the fact that they worked for a company that was, you know, a capped profit entity that had a nonprofit board sitting on top of it, right? Sort of the best of both worlds there. But I think what happened over the course of the weekend and, you know, what reportedly Sam is doing, you know, behind closed doors, kind of, you know, rallying Satya, Microsoft's CTO, you know, to the cause, sort of getting a lot of the tech industry behind him, getting a lot of OpenAI's employees behind him. I think what it showed is what Sam's value was to Open AI, which is that he's a dealmaker. He's a money guy, right?

He helps get, secure that investment. He got Microsoft to commit. $10 billion and not get a board seat. I mean, that's like some baller stuff, right? From that deal-making, you know, perspective. And so I think that there was this understanding. Maybe he hadn't convinced, you know, everyone in the company that his idea of we need to raise money,

make money in order to build the safe, you know, perfect technological product at the end. He might not have convinced everyone, but I think it was a lot more convincing towards the end of the week. when you look at, you know, Microsoft using its leverage and saying, well, we haven't given, you know, you guys, most of the $10 billion investment. It hasn't been wired to you all. So, you know, maybe we pull that, right? This idea of, you know, what is OpenAI without the, you know, the war chest without any of the money? Can, can they do what they want to do? And I think it started to look more and more over the course of the weekend that, you know, the safe, the people who are so principled towards alignment and safety that they were willing to completely upend the company that maybe that was actually really short-sighted. Sure, the principle is great. And so,

I mean, I think when you see, you can see right now, at least from what is being reported, right, that 700 out of 70 signed this letter asking the board to resign. It's a pretty small faction. But anecdotally, the reporting that, some of the reporting that we did shows that it was an extremely vocal group, right? Like, I think one thing that has to be understood about some of the AI safety crowd, not just that open AI, but, you know, throughout this industry, and especially in San Francisco, is that there is like a quasi-religious fervor to it. There is a belief that this is civilizational, you know, it's not for some people, it's not just, a power struggle. For some people, it is truly like life and death. And I think that that adds

sort of like a weirdness to all this, right? It's sort of like a fun house mirror that it's really hard to speak about this always, you know, fully rationally. But I do think over the weekend, Sam may have proved, you know, proved his value to certain people who may not even have been on board on Friday morning. That's a really interesting way to think about it, that like two things can be true. Number one, that the AI safety group within OpenAI is real and it's very vocal and at least as it is represented on the board is very influential. And also number two, as it regards the employees of the firm, there's been a little bit of a hurting effect around Sam Altman in the last 72 hours as people have sort of come to realize, wait, do we really want to blow up this company? Do we really

want Sam Altman gone? No, you begin to maybe there's a little bit of peer pressure. Like, you got to sign this letter. Everyone else around me is. There's, you know, there's fuck yous that are being like sent to the new CEO of the company, and people are, like, you know, upvoting it on Slack. Like, within remote companies, I can see, or within any kind of company, I can see how a bit of a hurting effect can come into play such that the momentum around Sam is really building. Ross, we have to talk a little bit about this guy, Ilya Satskiver. He's the chief scientist at OpenAI. He's one of the founders of OpenAI.

This guy, just to summarize his last four days, number one, on Friday, he led the board decision to fire Sam Altman. on Sunday he refused to hire Sam Altman back to the company. And then today, he signed a letter demanding that Sam Altman rejoined the company and demanding that all of the board members that voted with him to fire Sam Altman be fired from their jobs. Number one, everyone on the call, do I have that summary accurate? And number two, hopefully I've drummed up enough enthusiasm around listeners to understand who the F is this guy. Ross, you spend time with him, who the F is this guy?

Yeah, and first I just want to add on to something that Charlie said, which is that, or really, first I want to just add to Charlie's narrative that in addition to Sam sort of showing his stuff over the weekend, I think that's such an excellent point that employees looking around for leadership suddenly realized like, oh, wow, yeah, if you need tens of billions of dollars, you know, to run your business and stay at the top of this field, this is probably your guy. It's also, and this hooks into your question, Derek, like, The other faction really has not covered themselves in glory throughout this whole sequence, right? Like, you can imagine being like anonymous open AI employee on Friday.

This announcement comes out, you know, Sam from your perspective has had, you know, the company's been doing well, good business performance. But you want to take it with an open mind, right? You're like, okay, well, maybe he did something truly atrocious. No details are forthcoming still 72 hours later. accept a kind of string of denials that it's anything bad. No, it's not a personal scandal. No, actually, there was no malfeasance.

You know, no, actually, he didn't really lie about anything material. It seems like, no, there's not even a dispute about safety, I think, in one of the recent statements. So I think at that point, you're like, well, why are you blowing up this cool thing that I've been involved with that my grandparents and, you know, all my high school classmates are texting me constantly, wow, you work at Open AI. That's so cool. they're so neat. I use chat GBT. You know, you're blowing up our whole future and for what?

And, like, you know, can you give us even a paragraph length statement? And not that I had any foresight on this at all, but I did have, you know, friends texting me on Friday asking me the very question you just asked about who is Ilya Sutskever. And what I told them was, I really liked Ilya. Karen, I'd be curious to get your impressions. I happen to be sort of, I'm kind of drawn to my greatest credit to the, like the mystic philosopher. And Ilya's like a really charming kind of like deep thinker, you know, he makes like,

nomic pronouncements, you know, when talking about AI in ways that I found sort of charming and intriguing. And he's also like sneakily funny. So I really enjoyed my time talking with him. However, this guy's not Napoleon. Like what I said to people on text messages, it's like, this is not the guy who's going to lead a coom. he doesn't have that kind of executive leadership

like rally around me sort of effect unless I thought the only exception to that was if he had really as was speculated and looks like didn't happen but if he had really seen something in the lab that could be explained in 30 seconds

and sounded really, really dangerous to almost any lay person then I think there might have been a rallying effect around him But as it is with a personality that's kind of pretty mild and nerdy and interior, it seemed to me that you could see already that this coup might be destined to fail. Karen, quickly to you before we talk a little bit about Microsoft and some big picture takeaways, is there anything that you have seen in your conversations with Ilya that can shed any light into what's happening

and why he might have tried at least to lead this kind of coup. And coup is a weird word for it because in a typical coup, it's the underlings who throw out the king, but in this case, actually, the board of directors is above Sam Altman and have all the power in the world to fire him. This is not the French Revolution.

It's the opposite. But anything in your dealings with in conversations with Ilya that can lend any kind of light to what's happening. happened the last few days. So I think one thing to just say before we dive into that is we don't actually know if Ilya led this. Like it is, it seems to be the case through like lots of reporting that, but that, I mean, like he was the one that delivered the news to Sam saying that he was fired. He was the one that told Greg that he was going to be demoted. But we don't know if he was, you know, the leader or the messenger. It's still like a little bit unclear. And also I would say that

it's a little bit unclear how much employees within the company were fully involved in this. Like, it seems like everyone was roundly shocked that it happened. So there are still kind of like a lot of missing puzzle pieces to precisely what Ilya's role was in all this. But I would say, like, on the assumption that he did play like a central role and was like a major leader in kind of initiating this thing, that Ilya is, he is, he's kind of like a wild scientist, like very brainiac type is not the best at people. Like he's, he's like charming in the sense that people respect him a lot.

So he was one of the co-authors of the paper that kicked off the Deep Learning Revolution when he was a PhD student. So he has a lot of cachet within the AI world as like one of the frontier people that like kind of saw the light and made this whole technology happen. But he's not like the greatest at people. Like he's a little bit awkward. He says things that are, that can sometimes be like hilarious to people and off-putting to others. He like he used to in grad school, this is from Cade Metz's book, Genius Makers. He used to in grad school, like, whenever he got like really excited about a research idea, start doing like one-handed handstand push-ups. Like, and he would say,

At a time when people did not believe that deep learning, or there was a lot of controversy around like whether deep learning would be like the paradigm for AI, whether it was the last paradigm, whether it had flaws, whatever, he would say these like very dramatic pronouncements like deep learning will solve everything, like never bet against deep learning. Like he's always had this like religious fervor against with everything. So I do actually feel like it does make a lot of sense that he would kind of with this religious fervor be like, we need to fix this. Like there's a problem and we need to fix it. This is the solution. And then suddenly as everything falls apart around him again, he's not the greatest at people. So I don't know that he would have anticipated like the human factor to like an action like this, that he would then be like, oh no, this was a mistake. Like this is bad.

Like I need to like, I need to. And the thing is, like, he will very dramatically switch his position if he feels that, you know, that what he did was an error. So it does make sense to me and it does align that he would suddenly then be like, actually, I didn't mean any of that. It's funny because, like, strong convictions weekly held is a perfectly nice idea for, like, an engineer or, like, elite scientist for a company. it's a really fucking bad idea for like the chair of the board of directors when deciding to fire the CEO of a company that's worth 86 billion dollars, you know, like strong conviction so weakly held that I might in 72 hours decide that I no longer want to fire the person that I have already fired. Like it goes again, this is why I wanted to make sure that we started with a bit of the

history. It goes to this schism in the identity of open AI. On the one hand, you have Ilya, who's this mad scientist, spiritualistic brainiac who is no business. running a team of 770 people that has to raise billions and billions of dollars for computational power and treasure, incredibly instrumental for the success of Open AI, but shouldn't be running it, leading the effort to fire Sam Altman, who is the kind of person that you would want to run an institution that has to raise money and hire brilliant engineers. Charlie, I have just some thoughts that I want to throw at you, and then maybe you can throw some thoughts right back at me. Like, I guess I want to bracket what I'm about to say with the fact

that we don't know what the future looks like and we really don't know what the future of AI looks like. And it's possible that in the fullness of time, the Boar's decision to fire Sam Altman and slow down the development of AGI will look like one of the most brilliant ideas in human history if AGI turns out to be as dangerous as some people fear. That said, I mean, is there anything else to say about this, except that it is a shambolic performance. And like, an ironically shambolic performance, especially since in the founding charter of OpenAI, there's all this talk about the importance of research transparency when inventing what they believe to be the most important technology in the history of the world. This process has been done in the least transparent way

possible. I mean, at this point, it is comical how non-transparent the board of directors has been in terms of firing their CEO. I mean, is there any way around the idea, the conclusion, that this board of directors is, I guess, to use a turn of phrase that's popular in AI speech, like catastrophically unaligned with the actual company that they're overseeing? So before even getting to that, there's another bit of irony that I've been thinking about for like the past, I don't know, 24 hours or basically ever since,

I don't know how long it's been since Sam Altman said he was going to join. Microsoft, maybe it has been 20 minutes. Who knows? Times a flat circle. But what is incredibly ironic to me is we've gone through the history just now of OpenAI and it's founding. And sort of core to this, right?

Is this idea that it is going to be protected, the whole point of this was to draw talent away from some of these bigger for-profit tech companies and also to, you know, to safely develop something in a nonprofit way. And with some of the company like Google being, you know, almost the opposing force in 2015. By making this principled safety decision to fire Sam, effectively as of right now when we're talking, what has happened is it has driven Sam. and the president of OpenAI into the arms of Microsoft, one of the largest companies,

and potentially a lot of Open AI employees who might go there as well, right? Again, the future isn't fixed right now on this. We don't know how this is all going to shake out. But if that were to shake out, I mean, essentially what's happened is, you know, the destruction of a, you know, rising tech company,

AI company, that has this non-profit, you know, research focus and driving it into the arms of Microsoft. It's kind of wild. Like maybe that was completely unintentional, but, you know,

to not game that out in any way, to sort of take this very principled stand and not see that, you know, perhaps this is going to really piss off, you know, the money that is really going to piss off Microsoft, that is really going to piss off Microsoft, that is really going to piss off,

a lot of people in Silicon Valley, including maybe Sam, who decides I'm going to go to a competitor, right, or I'm going to try to build my own company. And then, you know, from a safety perspective, if you're extremely worried about safety, do you want to launch a competitor that is no longer governed by a charter that says we're trying to better humanity, but instead may just be a for-profit company? It's pretty wild, right? And then from the standpoint of transparency, you know, I think when we were writing our piece, I think, you know, Karen really summed it up best, which is that, you know, this really just shows that this is happening behind closed doors, right? The future of this, of this technology, like, what we're watching is a power, is this power struggle between a very

small group of people, most of whom are already obscenely wealthy, or, you know, occupy these positions of immense power, whether at this tech company or somewhere else, and also have access to a, a set of skills that, you know, most of the world doesn't even know how to evaluate. And they're the ones who get to decide, right? And we all sit here and watch. We spent the last whatever, you know, 36, 48, 72 hours wondering, like, what the hell the future is of this company. And we, that's very indicative of what, you know, the larger situation here is. We're all just sitting around waiting for them to make this technology, deploy it, tell us what's going on, which jobs are going to be lost,

which jobs we're going to gain. And there's not much we can really do about it. Yeah, it's interesting. There's been a million jokes been made about how this is the second consecutive November in which the most famous young guy in startup tech, who happens to be named Sam, has been caught up in a scandal that mostly played out

on the platform, formerly known as Twitter. The other echo of history there is that, you know, I have a lot of contacts in the effect of altruism movement. And just as the implosion of FTX and Sandbankman-Fried just put a cannonball in the idea of earned to give, that is earn as much money and then give it away in order to better causes like malaria and climate change, because it turned out that when that idea, which might have made sense in principle, made contact with Earth, it led to someone being incredibly

unethical with money and then creating an incredible wealth that actually dissipates because the government takes it away. In the same way, the principle of AI safety has to ultimately exist within institutions. You need institutions in order to put forward ideas. And I know lots of people who know a lot more about this subject than I do. But I have to feel like if they are deep believers in the idea of AI safety, they have to be looking at this, just as you said, Charlie, with like a bit of horror. We've created a scenario in which we've maybe pushed Sam Altman into the hands of the second largest corporation on earth

when our initial strategy was to keep this technology out of the largest corporations on Earth. The idea of ethical AI, at least in this case, and this narrow case, has failed to make contact with reality. Karen, one question for you, and then one question for Ross to close us out. I do want to talk a little bit about Microsoft because I got some texts from friends

in the tech industry who were like, I can't believe that Satya Nadella pulled this off. This is the most successful aqua hire in history. He's going to get all of the treasure and talent from OpenAI for free. They're just going to walk down the street or I guess, I don't know, take a flight to Washington State and work for him because they were kicked out of the company. This is crazy. What do you make of this situation? Like, could this have worked out any better for Satya Nadella and Microsoft?

Yeah, we really need to talk more about how Satya is the hidden king of this whole saga. I mean, I think, like, from people that I've spoken to that have worked with Satya Nadella, like he is a strategist, he's a pragmatist, and has certainly, even when he did the first investment into Open AI, it was sort of, like, it was a very strategic decision because Microsoft already had an AI research lab for a long, long time, called MSR, that had been doing some really great research that had been very transformative to the industry, to their ability to commercialize AI technologies. And it was sort of like a kind of like, why not both? Like, why not have MSR?

Why not make this bet? And just see, like, have this basically investor mindset, like, game out, like, which one's going to be the winning hand in the future? And so it doesn't surprise me at all that he's kind of orchestrated perfectly this new upper hand out of this situation. And I think what's really interesting is what's going to happen in terms of if Altman and Brockman decide to stay, how Nadella is going to convince them to stay in the long term? because these are two startup guys that have never worked for a tech giant in their life. I don't think they're compatible at all with tech giant culture. And if Nadella is somehow able to continue orchestrating some way to protect them from the kind of tech giant nature of Microsoft and have this like side lab that allows them to perfectly replicate opening eye,

that will be like the true, masterful kind of success for Nadella. But at the moment, I'm just, I'm not sure. like he's been able to pull this off and somehow say the right words to convince Brockman and Altman to give it a spin. I mean, assuming if they don't suddenly return back to Open AI. But that's kind of, yeah,

I think Nadella has always been strategic from the beginning and this will be like a huge coup for Microsoft if they can successfully get them to stay in the long run to run this thing. Ross, I wonder how much you blame Sam Altman for this mess. I saw someone call this corporate structure a turducken, not a turkey, or not a duck in a turkey, but a for-profit company and a nonprofit shell.

If it is, Sam Altman made this turducken. It was his idea to organize the company like this. It was his idea to create the for-profit subsidiary. He's in charge, and he oversaw the Frankenstein monster that has currently destroyed his career with OpenAI. To what extent do you, you know, given your time with him and your respect for his history of corporate successes, hold him a little bit accountable for allowing this situation to happen? Yeah, I mean, you know, when you do the global victory lap, you know, you also got to do the like the global dunce cap or whatever. I feel like, you know, to abstract it slightly away from Sam,

having taken that shot at him, I will say that I think that this is an interesting moment because both from the investor side and the founder side, if you're a founder, and let's just like, you know, stipulate for a second that these are really earnest people who did think that this technology would be really transformative and thought that it shouldn't happen in a for-profit context. even taking them at their word, we now see what a spectacular failure that was. Also, if you're an investor, and by investor, I mean someone like Microsoft or anyone deep enough pockets to bankroll the kind of compute that OpenAI is needed to sink

$10 billion into a company or 13, whatever it is. Going forward, you imagine that, you know, at the very least, your general counsel is going to want to take a hard look at this creative corporate structure of this company that you're so hot to invest in. But ultimately, you know, to talk about AI risk for a second and not get into fanatical doomer scenarios necessarily, but just, again, to, you know, I think all of us on this call have some concerns about the future of this technology.

Well, we've had a period of time where the founder of the most prominent company in this space has been telling us, hey, don't worry, we got this. In fact, we set up our company in this really clever way to make sure that nothing could happen. That strategy is now failed. The tech giants have it, as you say. And so really, to me, the big takeaway is the only way to meaningfully restrain this kind of technological development, if that's what we want to do, is through the law, is through government and policy, and perhaps even internationally, because we've let the private,

sector play their games with for-profit and nonprofit structures, and they haven't worked. Yeah, the last thing I'll say is my best friend Drew, who makes occasional invisible appearances on this show, basically said exactly what you just said, Ross. She said, this is a great example of how corporate structure innovation does not work. We've tried for hundreds of years to invent a better way to structure corporations and have boards of directors that are responsible to shareholders. It's really, really hard to get it to work for a long period of time for companies that are trying to do complex things like build these kind of products. What you want is boards of directors that respond to shareholders

and try their best to do good within a legal structure that regulates these companies based on democratic will. That is the general structure that seems to work. And when you try to innovate within it sometimes, well, you get the results at the last 100 hours. Ross, Karen, Charlie, thank you all very, very much. Thanks, Jared. Thank you.