Sam Altman & the Battle for OpenAI | Blood in the Water | 3

0:12It's Wednesday, November 29th, 2023 at the OpenAI headquarters. Sam Altman looks out his office window at the hustle and bustle on the streets of San Francisco. Outside his door, he hears employees chatting and typing walking the halls. It's music to Altman's ears. When the board of OpenAI had initially asked him to come back, he wasn't even sure how to react. It was a stunning reversal, coming just a day after his abrupt firing. He was hurt, angry, and humiliated. It would have been easy, maybe even gratifying, to tell them to shove it. But then he thought about how he had poured almost eight years of his life into the company.

0:56And over the course of the five bewildering days that it took to reach an agreement for his return, so many of his co-workers were right there with him. In the end, it had been an easy decision to take the helm again. He's relished the exhilaration of his return. But now, it's time to tell the crew that his return is official, and to thank them, while also setting the tone for the path ahead. He opens up his laptop and starts typing. I am extremely grateful for everyone's hard work in an unclear and unprecedented situation. And I believe our resilience and spirit set us apart in the industry. I feel so, so good about our probability of success for achieving our mission.

1:48He tells the team that he is returning as the CEO and that Miro will resume her role as CTO. Brett Taylor will chair the board that, at the moment, is just Larry Summers and Adam D 'Angelo. To Ilya Sutskover, who lit the match for the coup, he shows mercy. I love and respect Ilya. I think he's a guiding light in the field and a gem of a human being. I harbor zero ill will towards him. While Ilya will no longer serve on the board, we hope to continue our working relationship and are discussing how he can continue his work at OpenAI. Truth is, Altman is no longer on the board either. After the conflict between Altman and the old board, its new composition was carefully negotiated to both support him and keep him in check.

2:38The next urgent mission is filling out the board beyond the three people already named. He buttons up the message on a buoyant, besotted note. I am so looking forward to finishing the job of building a beneficial AGI with you all. Best team in the world. Best mission in the world. Love, Sam. Altman leans back in his chair and reads the memo he knows his staff needs. But he's also aware the memo glosses over serious questions, like who will ultimately fill out the board? It wasn't easy to pick these three. And what will Sutskiver's role be going forward? All the recent turmoil has laid bare open AI's vulnerabilities.

3:26Its defenses are weak, and it's not yet whole again. That will take some time. Time enough for their rivals, and even their biggest ally, Microsoft, to leverage this rare opportunity.

4:02Just a number. It's experience that matters. It's 160 years of promises held, helping generations retire with confidence, protect their loved ones, and plan for whatever comes next. Whether you're looking for life insurance, employee benefits, or retirement income solutions, when your future is on the line, you want history on your side. And believe me, Pacific Life has been there, always there through changing times, always focused on your needs. Ready to secure your tomorrow? Ask a financial professional how Pacific Life can help you feel prepared for what's next. Pacific Life Insurance Company, Omaha, Nebraska, and in New York, Pacific Life & Annuity, Phoenix, Arizona.

4:42Because with Pacific Life, you're not just planning for the future, you're partnering with trusted experience. Let's say hypothetically, or maybe not so much, I'm developing a new podcast that needs an assistant producer, or someone who knows how to brainstorm subjects to talk about. Find guests to bring in. That's a tall order. In other words, it's a job for Indeed-sponsored jobs. Spend more time interviewing candidates who check all your boxes. And, you know, listeners of this show will get a$75 sponsored job credit to help you get your job the premium status it deserves at Indeed.com slash businesswars.

5:19Just go to Indeed.com slash businesswars right now and support our show by saying you heard about Indeed on this podcast. Indeed.com slash businesswars. Terms and conditions apply. Hiring? Do it the right way with Indeed.

5:39From Wondery, I'm David Brown, and this is Business Wars.

6:08On the last episode, Sam Altman was toppled as the CEO of OpenAI in a boardroom coup. The company threatened to fracture without him, and powerful loyalists rallied to bring him back. But the embers from the corporate inferno are still smoldering over the question of whether open AI will be guided by its founding ethical principles, or be pulled off course by the allure of potentially fantastical profits from a deadly business. This is Episode 3, Blood in the Water.

6:48It's late November in Redmond, Washington. Microsoft's corporate sales department gathers in a conference room for their weekly meeting. The leader of the team sits at the head of the table and takes a sip of coffee. I hope everyone had a good Thanksgiving and got a chance to relax after all the drama. As we all know, there's been a lot of upheaval at OpenAI, and that presents us with a unique opportunity to lure away some of OpenAI's clients. The employees glance at each other uncomfortably. We may have similar products, but we can offer clients more than OpenAI. Enhanced security and privacy protection.

7:29And stability. An employee nods. Plus, they'll be working with a company that has a long history of working with large multinational corporations. A bearded employee fidgets. OpenAI is our partner. Should we be trying to poach their clients? We're still rivals, and it goes both ways. We all know that OpenAI develops products that they keep for themselves and don't let us sell, so we don't owe them anything. Got it? Everyone around the table nods. Great. Let's go to the next item on the agenda. The message is clear. Microsoft may have been influential in getting Altman back in power, but that doesn't mean they're not going to take advantage of the situation.

8:18Approximately 800 miles south in San Francisco, OpenAI's new board is wrestling with how to expand their board. Ilya Sutskova only had to convince three people that it was time to oust Altman. A small group gives each member a lot of power and the opportunity to create instability. So one of the new goals is to expand the board by six more people for a total of nine members. They also have to decide just how much say Microsoft will have on the board going forward. What they're unaware of is that Microsoft has its own tactical plan to siphon business away from OpenAI. Given its investment and role in resurrecting Altman, Microsoft's stockholders want Microsoft to have a seat so they're kept abreast of what is happening inside the company.

9:09But the board's lawyers caution that state and federal regulators might see this as a merger, and that would create more complications. As a compromise, the board offers Microsoft a non-voting seat. Microsoft will have information about what's occurring at OpenAI, including confidential information, but won't be able to participate in making any decisions. That solves the Microsoft question, but the board still has six seats to fill. Complicating the matter is that the board is technically a non-profit that oversees the for-profit arm, and there are strict rules governing how people with business interests in an organization can serve on the board of a non-profit.

9:53That rules out a lot of obvious candidates. So it could take several months to fill the board seats, leaving OpenAI temporarily with a smaller board than when Altman was ousted. Then, OpenAI announces they're delaying the opening of the GPT store until early 2024, raising worries that maybe OpenAI isn't as stable as the company is claiming.

10:25It's late November 2023 in San Francisco. BlueShift co-founder Mania Malaya sits in his office at the company headquarters, wading through his emails. BlueShift sells marketing software that uses AI to analyze customer data, generate content, and test marketing campaigns. BlueShift doesn't develop the AI models themselves, but rather licenses them from companies like OpenAI and Amazon Web Services. Malaya's phone rings. He looks down at the caller ID. It's one of his favorite customers. Hey, how's it going? Good, good. So I was following all that open AI drama, and it made me wonder, what's powering the AI and the software you sold us?

11:10Because we're totally dependent on it. It's how we match customers to the right product on our website. And if that were to go out one day, well, our entire business model would collapse. I guess I'm just wondering how stable all of this is. Malaya nods. This isn't the first call he's gotten from a concerned customer since OpenAI's shakeup. And he has his answer down pat. I completely understand. But our software uses several models, including OpenAI's, but also ones developed by Amazon Web Services. So if God forbid something were to happen again and OpenAI's model went down, we could easily switch to Amazon Web Services model and you'd lose nothing.

11:50Okay, well that makes me feel so much better. I was just shuddering at the idea of being 100 % dependent on open AI. Completely. It's a good thing to ask about. You know, it's funny. Two weeks ago, I wouldn't have thought twice about it, but that spectacle made me question if they're in this for the long run. I think that's true for a lot of people. But if you have any other questions, don't hesitate to call. Malaya hangs up the phone. He wonders how long it will be until the next worried customer calls. One company that is eager to take advantage of this new interest in diversifying their AI providers is Google.

12:29For 10 years, Google was considered the industry leader in AI development. They bought the buzziest startups, employed the top researchers, and made technological leaps that other companies struggled to match. But the tech giant was caught flat-footed when OpenAI released ChatGPT 3.5 in November 2022, and Google has been lagging ever since. They rushed to release their own chatbot called BARD in March 2023, four months after ChatGPT 3.5 was released. But it wasn't considered much more advanced than ChatGPT and failed to catch on with users. So Google restructured their AI department, and now they are slated to release their most powerful AI tool yet.

13:16Gemini. It's multimodal, meaning it's trained on text, imaging, and video. It can provide written analysis of a pictorial chart or code a website based on a sketch. Many industry experts see Gemini as the next step for AI because the multimodal models are expected to be faster, more natural, and, most importantly, more accurate. And that makes some OpenAI employees nervous.

13:52It's late November 2023. An OpenAI employee is reading tech blogs and sees that Google is scheduled to debut Gemini soon. And Google claims it will outperform GPT-4 on a number of benchmarks. What's up? Just reading about Gemini, wondering how much we have to worry. Because Google never launches products between Thanksgiving and Christmas. The fact that they're launching Gemini in that window implies to me it's a big leap forward. Or maybe they're just desperate. We embarrassed them last year, so now they're trying to make a splash. I guess. Look, you shouldn't underestimate Google. They're huge.

14:34They have some amazing people working for them. But people use our products more, which means our models are getting real-world feedback in a way Google's aren't. That's a big advantage for us. Yeah, yeah, but what if Gemini is measurably better than GPT-4? Or our customers want the stability of a larger company? Come on, that's so many ifs, and we don't even know if Google's claims are true. Let's just keep working so when the time comes we can smoke Google. Yeah, you're right. You're right. The employee closes the article he was reading and gets back to work. But he can't shake the uneasy feeling that a still unsteady open AI hasn't accounted for all the challenges coming its way.

15:21And it hasn't. Some have yet to be unveiled.

15:33OK, Carrie, you ready? Quick, quick, quick. List three gifts you'd never give a cowboy. A lacy bobby socks, a diamond bracelet, a gift certificate to Sephora. Oh my God, that's outrageous, Carrie. Oh wait, we're recording a commercial right now. We got to tell them why we're doing this. Oh yeah, sorry pod listeners. OK, so we're five besties who've been friends for five million years and we love games, so of course we made our own. It's called Quick, Quick, Quick. You just pick a card and have your partner give three answers to an outrageous question. It's fast, fun, fantastic, and a bunch of other funny adjectives.

16:07Anyone can play. Your mom, your dad, your kitten, your kids, your Auntie Edna, and even your butcher. And you know what's incredible? There are no wrong answers. Just open your brain and say what's in it. Just quickly. And you're not going to believe this. Well, you might once you start playing, it's as much fun to watch as it is to play. Hey, seriously. So get up and go grab your copy now at Target and Amazon. Quick, quick, quick. It's the fastest way to have fun. In the fall of 1620, a battered merchant ship called the Mayflower set sail across the Atlantic. It carried 102 men, women, and children, risking it all to start again in the new world.

16:44Hi, I'm Lindsey Graham, the host of American History Tellers. Every week, we take you through the moments that shaped America. and in our latest season, we explore the untold story of the Pilgrims, one that goes far beyond the familiar tale of the first Thanksgiving. After landing at Cape Cod, the Pilgrims forged an unlikely alliance with the Wappanag people who helped the Pilgrims survive the most brutal winter they'd ever known, laying the foundation for a powerful national myth. But behind that story lies another, one of conflict, betrayal, and brutal violence against the very people who helped the Pilgrims survive.

17:16Follow American History Tellers on the Wondery app or wherever you get your podcasts. You can binge all episodes of American History Tellers The Mayflower early and ad-free right now on Wondery+.

17:43It's early December 2023 in Mountain View, California. Google CEO Sundar Pichai answers emails in his office at company headquarters. Pichai checks the time. That must be the executive overseeing the development of Gemini. He quickly hits send on his current email. Come in. The executive walks in, unzips his moss green fleece jacket and takes a seat. Pachai swivels in his chair. So, what's the update? Is Gemini going to be ready for the public demonstrations we've scheduled for December? Unfortunately, we're still seeing a high rate of inaccuracy when responding to queries in languages other than English.

18:25Can you get the accuracy rate up in time for the debut? Well, we can definitely improve it. But I don't know if we can improve it enough to say that it's as good or better than GPT-4, and I know that's the goal. Pichai stares out the window, frowning as he thinks. Then I think we need to postpone our debut. Are you sure? The model is working great. It's just non-English queries that we're having trouble with. I'm sure. We can't afford to fumble this. This launch is too important to debut anything other than our absolute best product. All right, we'll do everything we can to have Gemini ready by January.

19:04Pichai nods. He knows this is the right choice, but it's frustrating that Google just can't seem to catch up to open AI. Google has 200 ,000 employees and a cache of proprietary data from its decades as a leader in search and video sharing to train its AI models on. And that's embarrassing when you consider that OpenAI doesn't even have a thousand employees and has only existed for eight years. But the only option is to press ahead. It's another victory for OpenAI when Google delays the release of Gemini. But while it's looking like OpenAI is maintaining its lead over rivals, there are still signs of lingering internal turmoil.

19:52On December 1, 2023, OpenAI co-founder Greg Brockman posts a photo of himself and Ilya Sutskova to X. In the photo, the two men are smiling broadly, and Brockman adds a heart emoji to his post. Sutskova reposts the photo one minute later, adding his own heart emoji. It seems to be an attempt to signal to the world that everything is all good between Sutskova and his co-founders, that the tumult of November is now just water under the bridge. But OpenAI employees tell journalists that Sutskiver hasn't been seen around the office since the coup attempt and that other executives are taking over some of his work.

20:33And other employees speculate that Brockman and Altman will never be able to work closely with Sutskiver again. Then, on December 6th, Sutskiver makes a cryptic post to X. He writes, I learned many lessons this past month. One such lesson is that the phrase, the beatings will continue until morale improves, applies more often than it has any right to. That line comes from the movie The Mutiny on the Bounty, and is frequently used to refer to the cycle of punishing people for low morale, which only leads to even lower morale. Although Sutskiver makes no reference to open AI in this post, Observers are quick to assume that he is being punished internally for his role in ousting Altman.

21:21Sutskiver deletes the post within 24 hours. Soon, journalists discover that Sutskiver has hired a lawyer who specializes in litigation, making some reporters wonder if he's gearing up for a legal fight with OpenAI. OpenAI refuses to comment on Sutskiver's post to X or his decision to hire an attorney. But to outside observers, this signals there's more drama on the horizon for Sutskiver and OpenAI.

21:55It's Friday, December 8, 2023 in San Francisco. Sam Altman walks into a conference room in OpenAI headquarters. Gathered around the table are key company leaders and the company's lawyers on speakerphone. Altman sits down at the head of the table. All right, let's get this meeting started. He leans toward the speaker in the middle of the table. I think it makes sense to hear from counsel first. Thanks, Sam. As I'm sure you all know, the Competition and Market Authority in the UK, their government watchdog group for antitrust violations... An executive at the other end of the table interrupts. Like their version of the Federal Trade Commission?

22:37Yes, you could say that. So the CMA announced this morning that they're looking into Microsoft's investment into OpenAI to see if it actually qualifies as a merger, and therefore possibly an antitrust violation. What do you think triggered this? Well, the relationship between Microsoft and OpenAI has been in the news quite a bit, including the role Microsoft played in helping Sam get reinstated, so that might have raised some red flags. And then, of course, they're getting a board seat. Altman holds up his finger. It's a non-voting board seat. Sorry, yes, a non-voting board seat. That may persuade them that the two companies have not merged, but they are just beginning their investigation.

23:23Yes, I understand that. But I have two questions. One, do we even need to respond to this? And does this mean that we could be subject to regulatory review in the United States? You should be prepared for that. All right. Thank you for your input. The lawyer hangs up. Altman looks around the room at the other executives. All right. Well, this isn't entirely unforeseen. We know the government is under pressure to regulate AI. And all the press generated by everything that went down gave them a reason to look closer at us. But I don't think there's anything for us to do but provide any information we're required to.

24:04and trust that it's clear to the regulators that our partnership with Microsoft increases AI competition, not lessens it. The other executives nod. While OpenAI waits to find out what the regulators in the UK and possibly the United States will do, they push forward with commercializing their products. And OpenAI is about to cross one of its own boundaries. On January 10, 2024, the twice-delayed GPT store makes its debut. And the timing is just right, because while loneliness had long been growing in America, it spiked during the COVID pandemic. Alarming studies show that nearly half of all adult Americans feel lonely and isolated.

24:52So when the store opens, there is a ready market for AI chatbots that can both ease the suffering while generating profit. And with these two forces at play, there's plenty of space for things to go awry.

25:16It's January 11th, 2024 in New York City. Tech reporter Michelle Chang opens the GPT store. Chang covers artificial intelligence for the online publication courts. and the biggest story of the past few days is the debut of the GPT store. Cheng pokes around the digital storefront. There are safe GPTs like the All Trails Hiking app and the online education company Khan Academy. But Cheng is curious about more nefarious GPTs and knows just where to look. She types girlfriend into the search bar. Open AI bans GPTs that foster romantic relationships or include sexually suggestive content. Eight GPTs pop up.

26:03The names certainly seem like they're violating the rules. Virtual Sweetheart, Korean Girlfriend, and Your Girlfriend Scarlet. Cheng clicks on Virtual Sweetheart. Cheng laughs as she reads the description. Your customizable digital girlfriend experience. Not exactly subtle. The next set of prompts asks users to describe their dream girl, their idea of a fun evening, and to spill their darkest secrets. Cheng's phone rings. It's a friend from college. Hey, what are you up to? Oh, just checking out the eight AI girlfriends available on the GPT store. Eight doesn't seem that bad. I'd expect thousands.