Byrne Hobart on Sam Altman, Meta vs Google, and Effective Altruism

This week on Upstream, I wanted to share a conversation I had with Byrne Hobart, author of the tech and finance newsletter, The Diff. We discuss Sam Altman's ousting from OpenAI, why Byrne is long Microsoft, and effective altruisms past and present. This episode was recorded the evening of November 20th. The story at OpenAI has progressed since then, but I think you'll find our discussion remains insightful. Please enjoy.

Hey, Byrne, uh, how's it going?

Pretty good. Been a fun few days.

Yeah, exactly. We're talking on the Monday just after, uh, uh, the weekend of Sam getting ousted and potentially returning. Who knows? Things are, are, uh, changing in real time. And by the time we release this, probably in, uh, the next day or two, things may be different. But let me ask you, what are some reflections you've had, maybe even higher level, or what is non-obvious but interesting to you about this whole situation?

Yeah, I think one of the interesting things is that at one level, governance is clearly really important, and who is on the board matters a lot. And different boards do different things, and if they have different incentives, they behave in different ways. On the other hand, It sort of seems like no matter what your structure was, if Sam walks and goes to a different company and tells all of the people at OpenAI, hey, this is where we're doing this AI thing now, it turns out the vast majority of them want to follow him. So as of pod time, there are over 700 of the 770 employees have signed an open letter. 2/3 of them signed it apparently between the hours of 1 and 5 AM Pacific time saying we want to join Microsoft. So it kind of seems like it doesn't, it didn't really matter what the legal structure was. And I guess part of what that tells you is that if you're taking a step back and saying we want institutional constraints that are safeguards on bad actors, whether we're trying to prevent AI or runaway AI or runaway EAs or whatever, that a Ultimately, those can't actually overcome human relationships. Now, maybe there is some alternative where they could, maybe if there's a blockchain-based forum and if you don't have something signed by the private keys of a majority of board members, the servers just turn off and the AI, the evil AI is dead and Sam's key card doesn't work and he can't get into the office. But a lot of this stuff does actually boil down to human relationships. Who trusts whom? Who has the rizz and who does not? And even if it says on paper that there's one group of people in charge, they're they're not truly in charge. They're they're sort of in charge when the decisions don't matter that much, and then when the decisions really matter, that's that's when you find out who's actually in charge.

Yeah.

So I think that's that's one piece of it that's interesting. I think one other piece is just on the the discourse that people were pretty quick to fit this into some kind of narrative, and I've I've tried to be as agnostic as I can. Like. I, you know, personally, given my own biases, would love it if what happened was there was just a really, really cool AI advance and Sam thought this is going to create a very cool product with a lot of ARR. And then a majority of the board members thought, no, this is actually going to kill us all. And then, you know, we look at the product and it's like, this is a cool product. I would pay $49.99 a month instead of $19.99 a month if I could use this. And it's not actually going to end the world.

Like.

Just based on my biases, that would be the ideal outcome, but I have no idea. And there's just a— when the charge is he wasn't forthcoming with the board, what you know is that there's something the board didn't know and then they found out. And to guess what that was, you have to assume that you've figured out something about OpenAI that their literal board of directors was not aware of, and you were able to figure it out before they did and then catch up to them or be ahead of them or whatever, which may be true. Like it may be that I don't know, maybe Sam had some really spicy tweet in 2015, and it's just been resurfaced. And he said it wasn't him, and he was hacked. And then it turns out he wasn't hacked, and now we know. But more more likely, there's either it was some kind of more abstract political question, or or there's just something we didn't know about that that perhaps perhaps is fine, perhaps actually looks really bad.

I don't know.

There's there's a lot of uncertainty, and that makes it really really hard to judge people. It has been interesting also just to see how much of the coordination of this happened online. I thought the hearts thing was really interesting. It kind of reminds me of—

Fascinating.

The opening of the War of the Roses, where people would literally pluck a white rose or a red rose to show which side they're on, except in this case, everyone's plucking the same color rose. So you don't really have a war when you ask which side people are on and everyone's on one side. It's just over at that point. But yeah, so, and for listeners who did not see this as it happened, people at OpenAI started tweeting about how much they supported Sam, and then he would retweet, he retweeted some of these with a heart or replied with a heart, and other people were replying with emoji hearts. And so there were a lot of hearts floating around, a lot of OpenAI people tweeting that OpenAI is nothing without its employees. And like, at this point, that's not strictly true. Like, as more people walk out the door, the residual value of all of the Cloud credits actually starts to become pretty significant as a share of the value of OpenAI, but not not the outcome any investors really hope for. But it's like people are coordinating this or not not coordinating; they're participating in this online, and so they're sort of doing their own preference cascade. I guess if you want to compare it to actually we'll cut that. I can't pronounce the name of the guy I wanted to compare it to. The East European dictator who was doing a speech and he said everything is fine. And then if you watch the video of the speech, you could hear the crowd getting angrier and angrier at him. And he finally realized he's gotta run off and then he gets executed the next day. Um, it may be too dark. We'll skip that. Uh, people will be able to tell how spicy the content of the podcast was based on how, how long the runtime is compared to the 1 hour recording time.

Yeah, exactly. What I don't know What I don't understand is why isn't the board sharing their reasoning? Because they had the upper hand here. At 3:00 PM on Friday, the world thought— when they issued their statement, the world thought that Sam had committed some sort of wrongdoing, some sort of malfeasance, either financial or personal, that something was wrong. And then within a few hours, the Sam camp had changed the narrative to, hey, this was an EA coup. And Sam was the true person all along. And then over the weekend, the narrative evolved to, you know, hey, the investors want Sam back, the employees want Sam back, the board is waffling, they're considering resigning, which I don't know if that was ever true, by the way. But the board just looks like total amateurs. Sam is the total moral credibility here. It seems everyone's supporting him. Everyone's coming out in favor of Sam. Whereas the board is sitting on their hands. I guess they pick a new CEO, Emmett Shear, who is a very impressive person, but doesn't seem to have anything close to the support that Sam has. Certainly can't keep the employees from signing the letter. And if the board has any information that would discredit Sam, they should be sharing that, unless there's some legal reason why they can't, or they don't really have something, or they're just total amateurs. How do you make sense of that?

When there's a firing like this, and this actually applies to layoffs more generally, like you have these weird attractors in the space of possibilities where one is the board decides he's not the right person to run the company, but we probably all agree, like maybe he agrees too. And we also don't want this to be any more painful than it needs to be. So, you know, Sam is getting promoted to executive chairman and then 3 months later, Sam is starting a new thing and it's a side thing. He still loves OpenAI and then. like a year later, hey, no one at OpenAI has seen Sam around. I guess he has fully stepped down. Like, that's, that's sort of how, how you do it in a lot of corporate America is like you, you give people this really nice glide path to, to quitting. And, um, you know, it shows up in other places too, like, um, in, at some, in some financial firms, people, they keep their, they keep their health insurance through COBRA, but they also keep their Bloomberg access. So they can spend this brief period sort of pretending they still have a job. At least as far as everyone outside of the company is concerned. And it's just way easier to look for a job if A, you've got the terminal and you can do research, and B, you've got the terminal so you can ping all of your friends and stay in touch with people. So like, there's, there is that whole tradition of letting, like, easing people out really gently. And if you think that that's not what's going to happen, like, either he does not want to be eased out or it's not such a gentle process, like, you actually want it to be this very short, sharp thing where he's— Leaving. It's effective right now when this press release hits. And to make the case for that, you have to say you did something really bad. And maybe, maybe it wasn't really bad. That also makes them look bad and they don't want to disclose it. Maybe there's a negotiation and they're like, you know, you're fired effective immediately, but you can either be fired for a very specific reason or for this vague reason. And let's negotiate over how the vague reason will sound like. It's got to sound bad enough that it makes sense that we fired you, but, you know, not so bad that it's worse than what whatever you actually did. So yeah, you either end up with these very understated things that don't sound too bad at all, and where if you aren't used to reading press releases, you might not realize that this person had actually been fired, or it sounds pretty bad. It's kind of like a corporate execution and they're out and it's really clear they're not coming back. So I think if you don't think you can gently ease someone out, you basically have to do That if they've got to go, and then and then it's just a question of does the boards does the board have the the ability to execute on that or will they try to do it and and he'll say no. And you know there's like Matt Levine has written some fun stuff on corporate governance. There's there's actually this weird thing in China in particular where there's a company chop like a stamp you use to officially sign official documents, and the person who has the chop is. Legally pretty much in control of the company. Like they, they can sign off on things that other people can't sign off on. And, um, there's this wonderful book, um, Mr. China, about a private equity guy who was investing in China in the '90s. And sometimes the plotlines are literally the CEO ran off to a different country. He took the chop. And so he was able to withdraw all the company's money from his bank account, transfer it to his offshore bank account. And that was it. That was the end of the company. So I guess that is actually a case where you do have a sort of technology-based hardware solution to this governance problem. And so whatever the messy human issues are, the person who controls that private key can do whatever he chooses to do. But yeah, in this case, right, you sort of, you put out an announcement that makes things sound very certain. And then that is actually the point at which things suddenly become really uncertain. So I don't, I don't know, like you're the— I think the, the 2 theories to hold in mind for why the board hasn't been more specific are, one, Sam didn't do anything wrong and this was just a weird political thing and we just don't know what the politics of this situation are. Or 2, Sam did something really, really wrong and they can't actually bring him back if they reveal whatever this horrible thing was, or they don't think they can, or maybe, you know, it sounds really bad to people on the board and they feel like it would sound bad to them and their friends. And so they're not going to— they don't think it's possible to bring him back if they reveal what the thing is. So it's either— that's the definition of uncertainty. It's either really good or really bad for all parties involved. So yeah, I've tried to— this is another case where maintaining that radical uncertainty is probably the prudent thing because you literally have high confidence that there's something important you just do not know. And so all All definitive speculation is either going to be a lucky guess or it's going to look really stupid in a few days.

Yeah. Well, to that end, it is interesting to think about, hey, 80 to 90% of its employees just said they're walking out unless they bring Sam back. The board has doubled down and hired a new CEO, although Ilya has caved and said he regrets it and signed the memo. So it seems that they're going to— I'm just going to go out on a limb here. They're trying to gonna try to bring Sam back is my bet. At the same time, you know, one thing's interesting, just like if you're purely optimizing from a business perspective, which they're not, they're nonprofit, they're optimizing for their mission. But what I understand talking to investors is that ChatGPT is a $1.3 billion ARR business with a 40% gross margin. And the margins are only gonna get way better with a bunch of— like they probably were way overstaffed. you know, they can't afford to lose their best people, etc. But could— is the company toast if even 30% of people walk out? Like, what is the threshold at which, hey, this is really bad for the business, given the head start they have technologically, but also in terms of contracts, in terms of brand? Any thoughts there?

If they actually lose a material chunk of their employees, It would be just the turnaround, like the most impressive turnaround in recent corporate history if they survive and remain a viable company and continue to grow. That said, they do have a nice head start of they have ChatGPT, they have a really good brand name. I'm sure there are a lot of people who wanted to work at OpenAI and would still be open to that. I'm sure that of the people who either haven't signed the open letter or who signed it but wouldn't necessarily actually go to Microsoft if They really, you know, they were faced with this stark decision. Like, I'm sure they're really good people. So it would still be a really, really good AI research lab, even if they lost a lot of people. But there's also this momentum dynamic where it's just really hard. Like, once when a company is succeeding, it's much easier for them to recruit. It's much easier for them to do nice corporate development deals. It's much easier for them to raise money. All of that feeds into further success. So. You tend to have to keep the momentum going. This is why when you talk to founders, everything's always going great. Like it's never bad. They, they never lost their biggest customer. They never lost their best employee. You know, they're never like on their last 2 weeks of cash on hand or whatever. Like everything's always fine. And it's just really hard at this point to claim that things are even remotely fine. So I think if they did have a turnaround, it would have to be around some specific narrative where They've, they have to convince a very large number of people that it's really, really fortunate that OpenAI exists, given that Sam and all of these people went to Microsoft and are doing who knows what and building whatever. So I think at that point, they sort of have to pivot to being, either pivot towards being much more of an AI safety-friendly lab, or they have to just do like a heel turn and be like, We, we are now EAC all the way. And the problem with either of those pivots, like the EAC pivot is just, you know, crazy given who they hired, that Emmett is really, really worried about the potential for AI to end the world. So can't really do that unless they get rid of yet another interim CEO. Apple had, or Yahoo had, like, they had 4 over the 9 months that I worked there. So it has been done, but it gets harder every time you fire one. To survive. It's easier to fire the next one. But if they try to pivot towards, we are the safety-conscious lab, then there are other safety-conscious organizations. So they're Anthropic without the head start and probably with other disadvantages, but they would have a lot of Azure credits. So there's that going for them. So yeah, it's tough to keep something like this moving when it When the company faces a really huge setback, unless there's a core of people in charge who really believe in the mission and believe that this company is the only one that can accomplish that mission and who are really willful and effective. And so you just, like, betting on them surviving is betting that a lot of those extremely effective, extremely committed people don't go to Microsoft, that they're more committed to OpenAI than they are to Sam and company. And we'll see. We'll see how many— I mean, I'm curious about how many new employee badges Microsoft has actually printed up today and how many people are getting their new company-issued laptops and signing into Teams for the first time, all that stuff. It seems like, you know, one level things are moving really fast, at another level nothing's happening. Everyone is like— the people who are actively involved in this, whose fates are, you know, their fate and in their view perhaps the fate of all humanity is up in the air. They What else are they doing but checking Twitter just like the rest of us? I guess, I guess their, their group chats are probably a little bit more interesting, but, um, I think, you know, the, the volume of group chat conversations is probably pretty consistent across these different groups.

Yeah, it's, it's fascinating. I feel like this, it's got to be one of the most interesting corporate dramas of all time. Uh, I mean, this is, uh, this is tremendous.

Hey, we'll continue our interview in a moment after a word from our sponsors. Real quick, what's the easiest choice you can make? Taking the window instead of the middle seat? Outsourcing business tasks that you absolutely hate? What about selling with Shopify? Shopify is the global commerce platform that helps you sell at every stage of your business. Shopify powers 10% of all e-commerce in the US, and Shopify is the global force behind Allbirds, Rothy's, and Brooklinen, and millions of other entrepreneurs of every size. Across 175 countries. Whether you're selling security systems or marketing memory modules, Shopify helps you sell everywhere. From their all-in-one ecommerce platform to their in-person POS system, wherever and whatever you're selling, Shopify's got you covered. I've used it in the past at the companies I've founded, and when we launch merch here at Turpentine, Shopify will be our go-to. Shopify helps turn browsers into buyers with the internet's best-converting checkout. Up to 36% better compared to other leading commerce platforms. And Shopify helps you sell more with less effort thanks to Shopify Magic, your AI-powered all-star. With Shopify Magic, whip up captivating content that converts, from blog posts to product descriptions. Generate instant FAQ answers. Pick the perfect email send time. Plus, Shopify Magic is free for every Shopify seller. Businesses that grow, grow with Shopify. Sign up for a $1 per month trial period at shopify.com/torenberg. Go to shopify.com/torenberg now to grow your business, no matter what stage you're in. That's shopify.com/torenberg. Have you ever wondered where your donation could have the most impact? In 2007, a group of donors had that exact question. But when they sought out information from charities to help them answer this question, they instead received cute pictures or unhelpful stories. Their experience led them to create GiveWell, an organization providing rigorous, transparent research about the best giving opportunities they've found. GiveWell has now spent over 15 years researching charitable organizations and only directs funding to a few of the highest impact opportunities they've found in global health and poverty alleviation. Over 100,000 donors have used GiveWell to donate more than $1 billion. Rigorous evidence suggests that these donations will save over 150,000 lives. 50,000 lives and improve the lives of millions more. GiveWell wants as many donors as possible to make informed decisions about high-impact giving. You can find all of their research and recommendations on their site for free. You can make tax-deductible donations to their recommended funds or charities, and GiveWell doesn't take a cut. If you've never donated through GiveWell before, you can have your donation matched up to $100 before the end of the year, or as long as matching funds last. To claim your match, go to givewell.org and pick podcast and enter Upstream with Erik Torenberg at checkout. Make sure they know that you heard about GiveWell from Upstream with Erik Torenberg to get your donation matched. Again, that's givewell.org to donate or find out more.

The Microsoft point, I mean, what a coup by Satya. One wonders what he offered. Is he offering Sam that, hey, one day you'll be CEO of Microsoft? Is he saying, hey, You'll get the same offer. Open it.

Yeah, that's, that's what I think. I think, I think he probably put a date on when he is also getting promoted to chairman of the board and Sam is stepping up. Like, that's, to me, it's either that or, well, I guess, yeah, it's just, it's tough to, tough to make the case for joining a big company with an AI lab versus doing your own thing. I think the, Possibly in, you know, Satya, he's, he seems like a really great guy and I respect him a whole lot. I'm a Microsoft shareholder. I'm very happy that he's in charge. Um, he probably did not get to the position he's in and Microsoft probably did not get to the margins that they have if he were a pushover. So my guess is that a lot of this conversation with Sam was about things like, here are Microsoft's plans. Here is Microsoft's budget, uh, you know, which for AI is gonna be effectively unlimited. And here's all the stuff we think we can accomplish together, et cetera. And, you know, maybe, maybe there was some talk about, yeah, we will need a CEO someday and you seem really promising and we'll be a much more AI-focused company by the time I step down, et cetera. But I think the other piece of it was saying, hey, have you, have you talked to NVIDIA since the story about you trying to start your own TPU company to get them out of this space and, you know, stop relying on them so much? Like, have you talked to them since, since this story came out? And. Have you specifically asked them how soon you could get GPUs and what the pricing would look like and whether you get bumped back in the queue because there's a huge order backlog for these? Like, I think maybe, maybe it is actually harder for Sam to, like, harder for anyone, but especially for Sam to start an independent AI lab. It would be funny if he ended up starting a lab on his own, getting lots and lots of brilliant people and having all of them just drumming their fingers on the desk waiting for access to the one H100 at the office or whatever. Probably wouldn't be that bad. Like there are those cloud-based providers, but NVIDIA has a very tight relationship with the specialized AI workload-based data center companies. And so NVIDIA can probably suggest to them that NVIDIA's prioritization on who gets GPUs is probably a function of their prioritization on who gets to use GPUs. And so maybe CoreWeave does not want to sell all of their capacity to Sam right at this moment.

Yeah.

So yeah, it may be that like, you know, there's that he can't, like OpenAI and Microsoft might be like the 2 places he could realistically go to continue working in AI.

You know, one wonders that it feels like an ego hit, you know, sort of be an employee at Microsoft versus, you know, CEO of a $90 billion organisation at the same time. Microsoft's a $1.3 trillion company, and this is likely a landing spot to see, one, if he goes back to OpenAI, two, he gets the benefit of Microsoft's access, which include all of OpenAI's IP and resources. And so, and it looks like Satya probably gave him a sweetheart deal that allows him to sort of make a comeback story, even if he's not granted the OpenAI keys to the kingdom again, although I predict he will. But, you know, we'll find out likely by the time this episode's up, perhaps.

Yeah. I mean, so one thing to think about is like, Saadi is 56 years old. He does not need to retire right now and probably was not planning on retiring in the next couple of years anyway. So there is actually this long period for plausible deniability where Sam can spend the next couple of years working really hard, on Microsoft's new AI lab and doing, doing cool stuff in AI and gets to say, this is what I wanted to do all along. Like OpenAI was the best way to do it for a while, and now Microsoft is the best way to do it, but it's still the same thing. And then, you know, go, go do something else. Or, or yeah, he, he ends up on track to be promoted, but it's like, it's actually convenient that there's so much to do in AI that it's not like you, it's not like if, if he gets hired that he's gonna be really, really anxious to be worrying about OEM relationships and like, will Procter& Gamble ever switch from Excel to Google Sheets? And wouldn't that be terrible? I don't really think that his dream is to worry about things like maintaining high net dollar retention for Fortune 50 customers who are using Outlook. I just don't think it's not really the vision, but if he does well in Microsoft AI, assuming that's where it goes, then by the time Microsoft needs a new CEO, it's much more of a company that used its position in office software and cloud to become dominant in the deployment phase of AI. Like, I was, uh, chatting with a friend earlier and talked to— I concluded that Microsoft is actually like the bet on a slow takeoff scenario in AI where it continuously gets better, but it's always better at augmenting human behavior and at making particular jobs more efficient or making some jobs obsolete, but it's never replacing entire institutions, or it takes a very long time to do that. Like, in that scenario where the AI is really valuable but it's not totally apocalyptic and world-changing, the actually really valuable thing to have is distribution. Like, the AI models are going to get built. There are a lot of passionate researchers out there. But, um, how do you scale it to a bill— how do you get a billion customers? Um, ChatGPT seems like pretty unique in this respect that it was like a new product launch that was actually able to do this. But it's a whole lot easier if you have some product where what it's going to do is use your email inbox and outbox as training data in order to write your emails for you most of the time and summarize all of your emails, et cetera. Like it's a whole lot easier to do that if you already have hundreds of millions of daily actives and using your email client. So Microsoft does end up being a really good company for that vision of AI keeps getting bigger, keeps getting more important, but doesn't take over the world and doesn't end the world. And to the extent that we give credence to the idea that Sam got in trouble with the board for being too accelerationist and they were worried about P doom being too high if he stayed in charge, um, if he disagreed with them, then presumably he is more in the slow takeoff camp. And so he is actually a natural Microsoft bull.

I do want to talk about Microsoft in a second, but first I want to analyze or see what— ask, what have we learned about the governance structure, the nonprofit structure that OpenAI was created under? You know, Rune, a player at OpenAI, tweeted the other day that OpenAI could have never been built at a for-profit, which begs the question, you know, why were other genius inventions built at for-profits? But I guess, what have we learned? Has OpenAI been successful in spite of its model because of it? Or what have we learned in general about the differences in these models?

There is such a thing as person-institution fit. And what's nice about the for-profit model is that it does give you, one, some guardrails on just, is what you're doing actually having an impact on the world? Well, one way to measure that is, can you— is what you're doing, does it allow you to sell a product people want to buy for more than it costs you to make? And if you can do that, you have a pretty strong prior that you are actually doing something impactful. And the other thing the for-profit model means is that you're not constantly asking for money. So you don't have to persuade people about your mission. You just have to persuade customers that your product is better than the next best alternative. And I think that actually gives many institutions the flexibility to pursue longer-term goals that are not purely profit motivated. And then that actually feeds back into their profitability. So SpaceX is kind of the canonical example where You could imagine a charity that's devoted to making humans a multi-planetary species, but that's also a charity where at some point Musk would have said something that offended the major donors. Like, he probably would have very quickly offended every single major donor. They would have all pulled their money and no more SpaceX. But if SpaceX bootstraps to that by actually building a profitable business, then they, they learn more. They're able to hire more people and better people. They're able to iterate towards their long-term goal. But they're also able to hire people who just didn't see themselves working at Boeing and making a, you know, 1% better but 30% more expensive fighter for the rest of their careers and just constantly iterating on how much more can we charge the Pentagon for this. So, um, sometimes you, you get a nice feedback loop where because that vision implies a profitable company in the future, you sort of build the company and then shape the world in line with that vision. So I, I don't quite buy the idea that— I don't buy the idea that something like this could not exist as a for-profit company. On the other hand, it does exist as a quasi-for-profit company embedded in a, embedded in a nonprofit structure and with this very strong nonprofit history and ideology behind it. So, you know, as N of 1, but yeah, OpenAI clearly was not built by a purely profit-seeking entity. So maybe there is something special about it, but my guess is no. My guess is that this all would've been much simpler and more straightforward and less chaotic if there were just a regular board. If various venture capitalists had large— had board seats and Microsoft had a board seat and You know, other related parties had a board seat and, um, they were trying to make decisions that maximized shareholder value.

Hey, we'll continue our interview in a moment after a word from our sponsors. AI might be the most important new computer technology ever. It's storming every industry and literally billions of dollars are being invested. So buckle up. The problem is that AI needs a lot of speed and processing power. So how do you compete without costs spiraling out of control? It's time to upgrade to the next generation of the cloud: Oracle Cloud Infrastructure, or OCI. OCI is a single platform for your infrastructure, database, application development, and AI needs. OCI has 4 to 8 times the bandwidth of other clouds, offers one consistent price instead of variable regional pricing, and of course, nobody does data better than Oracle. So now you can train your AI models at twice the speed and less than half the cost of other clouds. If you want to do more and spend less like Uber, 8x8, and Databricks Mosaic, take a free test drive of OCI at oracle.com/upstream. That's oracle.com/upstream. oracle.com/upstream.

Let's talk about Microsoft. You've been long Microsoft for a bit. Why don't you explain your position there?

Yeah, like part of the bet with Microsoft is that distribution is something people tend to underrate and that it's really powerful and that having good relationships with the giant companies that are, A, spending a lot of IT budgets generally, and B, that have IT budgets that tend to grow faster than their revenue, like that's just a very nice structural tailwind. And then Microsoft has, they have this long history of executing broadly well. They missed some big Big paradigm shifts, but I think for forgivable reasons. I do, I am in the camp that Ballmer is underrated. I think one of the mistakes people make when looking at Ballmer's track record is to look at his tenure as CEO, because he was actually hired by Microsoft as their business manager in 1980. And Bill Gates gave him, I think it ended up being like 8% of the stock. So he was, it was not like he was some random person who was chosen to be CEO, but he also took over as CEO, I think in mid-2000. So took over at a time when tech stocks were just white hot. And then Microsoft did, yeah, they missed a lot of opportunities, but they were also really, really worried about antitrust during that time. And so, and by the time he left, Microsoft stock was just incredibly cheap. Like I think if you backed out the cash, it was at a single digit PE ratio. It was growing. I think it grew an average of like 10% annualized in terms of revenue during his tenure and was returning capital to shareholders. Like he did a lot of stuff. Right, and did miss some opportunities to do stuff better. But I think we would probably there probably more more articles about what a bad manager he had been if Microsoft had decided to just go really really hard on search in say 2005 and had ended up back in court once again because they're trying to squash this tiny weak company called Google that's run by these. Wonderful Stanford nerds and, you know, they, they had the office with the beanbag chairs and everything's in primary colors. Like, how could evil Microsoft do this once again? Like, I think his track record would actually look worse if he'd been more aggressive and he did the best that he could given this, this bad hand. And Microsoft does have a very long history of hiring brilliant people and getting, getting a lot of work out of them. Um, just intermittently expressed. And, um, I think under, under Satya, they've just increasingly recognized that they actually are now the big tech company with the least antitrust risk, in part because they make products that are fairly boring, that are more business-facing than consumer-facing. So they, people just care less about it because it's less salient to them. And we've all just gotten, gotten kind of used to the idea that a, a financial projection either exists in Excel or it isn't real, and that a legal document is going to be marked up and Word. So even if you don't like Word very much and I don't like Word very much, your lawyer likes Word and whoever you're doing business with, their lawyer likes Word. And so you actually have this 4-sided network and 2 nodes of that network are non-technical and they have, you know, at great personal cost learned all the keyboard shortcuts and learned how to navigate Word's many editing and redlining features. So yeah, it's hard to get rid of that. So yeah, part of it was just like Microsoft, They're in this very secure position. And then they were using that to grow in cloud. And because big companies, they're not like early adopters of cool technology necessarily, but they were early adopters of technology broadly. Like they are a little bit more up to date, a little bit more savvy than, you know, a regular mom-and-pop store. So they actually had these large legacy investments that in on-prem software that would gradually move to the cloud. And so that was another, another way that Microsoft could get above average growth. And then as, as the OpenAI stuff started getting more important, I started thinking about who, who has the most data and who is in the best position to convert something from really neat proof of concept to a standard tool that every Fortune 500 company uses. And Microsoft can do that. Google can do that too. But Microsoft, with Microsoft, it's a lot easier for them to say this is a direct complement to something that is already being used. And Microsoft also just has a lot of the specific kinds of data, the natural language text data talking about the specific concerns of a particular company. So they have a lot of the training data or have access to a lot of the training data that you would want to build a corporate AI tool. And one, another nice thing about selling to companies is that if your product has a good return on investment, you can actually sell it to people on the basis of its return on investment. Even if that return on investment is kind of abstract, it's like you're spending X dollars in order to save Y minutes of time. And the person whose time you're saving, their time is worth Z dollars per hour. And so this is actually profitable. Like, if you try to sell convenience to just normal people, it's really hard to get them to view ROI as this meaningful thing. Like something like DoorDash is, it is actually this, this great tool of outsourcing where it's basically like for, for say a tech worker who orders their lunch through DoorDash instead of walking to the same store and picking it up. Like they're basically, what you're doing is you're, you're firing your mid-six-figure delivery driver, which is you, and you're hiring somebody else who makes a lot less money and is happy to have gig work. Instead, so it's this economically rational transaction. It marginally reduces, or I guess it increases incomes in the 20th percentile and has an ambiguous effect elsewhere. Because if using these conveniences allows you to do slightly better at your job, get a slightly larger raise, you do actually increase income inequality, but you increase income inequality by making everyone better off. Making the white-collar workers with laptop jobs much, much better off than everyone else. So, like, but, you know, if you try to sell DoorDash as this, like, statistically you are more likely to get a promotion at work. And so if you calculate the internal rate of return over the next 10 years, DoorDashing your lunch every day is a better decision than just walking around the block and picking up a sandwich. No one would buy it. It has to actually be sold as this sort of convenience slash indulgence. It's the easy, lazy choice, and you just do it without thinking about it. But corporations, they do some things without thinking about them, but things like this, they do tend to think about them and tend to try to run the numbers. And fortunately, a lot of the numbers on taking, like, looking at someone's day, seeing how many hours a day do they spend on emails that take them 5 minutes that would take GPT-4 or 5 seconds, you can actually come up with a really high return on investment, even for a fairly expensive Copilot-type product. Not to mention the coders who, of course, their time is really expensive. They do spend a fair amount of it on boilerplate stuff. And for them, part of it is like this, there's a return on just not having to break your train of thought by looking up the syntax of this thing that you've done intermittently in this language for a really long time, but you always forget. You know, what order it is, what order the arguments in the function are in or whatever. Like being able to not lose your train of thought and just get that thing written so that you can spend your effort on the lines of code that actually matter is really valuable. That part gets really hard to measure. Measuring programmer productivity is just a notoriously difficult problem. That's why you do stock options because you just don't know what people's individual work is worth. So you want their outcomes just tied to collectively did the company do well or not. But it's clear that if you're saving some percentage of the time for people who are extremely well paid, that you have a lot of pricing power when you're selling that to a corporation, and then much less pricing power if you're selling that to individuals. So that's the Microsoft bull case.

That's well articulated. Let's pretend we're going back to 2010 and we're predicting sort of Microsoft over the next 10 years, and we're making the bull case then. around why Microsoft's gonna have a turnaround and be one of the most valuable companies in the world and kind of shock everybody that way in terms of being an innovator, like what are the biggest things we'd point to of what would have to be true? Like what explains the significance of the turnaround from a strategic perspective? What did they do that maybe other players didn't?

I think if you go back that far, it's just kind of cheating to say like, you know, this is how you would've predicted the turnaround. Like, I, I think the, the people who were pounding the table saying Microsoft is too cheap a stock in the early 2010s, they were just talking about the PE ratio. They're like, this company has— it's really tough to disrupt this business. Um, clearly the DOJ tried and they failed. Google has tried in office tools. They, you know, they've made an impact, but they haven't led to this wholesale shift. So you— I think if you were making the bull case then, you would've just said, Microsoft, it's kind of an unfortunate situation. Like they could be a bigger, better company, but they're still a good company. They're still growing. And so they're still just like this nice pull booth on IT spending broadly, which we think is going to go up. I think that the actual turnaround in some of that is that they did figure out cloud pretty early and they recognized some of their strengths in that space. But I think, uh, I, I was definitely not aware of that as a potential bull case for Microsoft, um, 10+ years ago. Um, I think it was like an interesting detail in the story when it came out, but, um, it sort of looked like they were kind of trying to copy Amazon instead of trying to copy Apple with the Zune or something, or, you know, try to copy Google with Bing. Like just one more instance of Microsoft not really knowing what they could do and trying to copy some other big tech company. It's just in this case, it turned out to be something that was actually close enough to their core competency that it was, it was actually the right thing to copy and they were the right people to do the copying.

Let's talk about some other big tech companies while we're, we're doing sort of, we're canvassing. Um, let's talk about Meta and let's talk about Google. And I know you've been, uh, you're, you've disclosed that you're, you're, you're long Meta. So I wanna hear that, that, that bull case and then, then I wanna hear your, your thoughts on Google. Yeah.

Yeah. So Meta, um, they're, they're another company that was just surprisingly well positioned for AI. They've been talking about AI for a long time and They have just a ridiculous corpus of natural language text data that reflects normal, informal, day-to-day conversations. And one of the really interesting things they've been able to do is they've been able to use this to train a lot of moderation tools. And in particular, like moderation tools, if you're trying to catch people who are selling firearms or drugs online, that's not trivial, but it's not super hard. But there are categories of text online that are just really hard to moderate because you have to get good at detecting sarcasm. So in their quarterly reports on content and what's been automatically removed, what hasn't, they actually track bullying. Bullying used to be the category where they actually had a pretty low hit rate. Like most of the bullying content that was flagged and taken down was flagged by users and not by their algorithms. Whereas for stuff like ISIS recruitment, like 99.9% of it, if you post You know, you should join ISIS. Here's the sign-up form. It doesn't. No human being will see it. It just gets taken down right away, and your account gets deleted. With bullying, it's it's tough to tell when you know when someone is saying nice shirt because you're wearing a nice shirt, and when it's like nice shirt is actually this humiliating dig, and they'll be crying for for hours because of it. But their Facebook's performance on that has actually gotten better over time, and they have been they've been advocating for stricter for more liability for. Platforms based on the content posted on them. And I think in a world where companies are actually liable for that, they basically have to use a moderation tool created by Facebook because Facebook has far more data and has just far more certainty that they're doing it right. Um, you know, everyone complains about moderation, but as long as the people complaining are pretty equally represented on both sides of any given issue— like if every progressive says Facebook is biased against progressives and every conservative says it's biased against conservatives— then you know they're actually Pretty much going straight down the middle. So that was part of the bull case. And then part of it was just, I suspected that online mobile advertising was not going away, that if you have to go from directly tracking transactions and being able to see everything someone does through the entire process to indirectly inferring what they did, that having more data means you're in a better position to do that. And that is in fact what what Meta did. They had a lot more data than other people and they were able to use that to build better and better attribution models. And so it ended up being the case that if you block direct tracking, you are actually massively privileging whoever had the most data on that tracking before you blocked it. So it ended up being an absolute tax on Meta, but a relative subsidy. And because there are network effects on both the usage side and on the advertising side, where the advertisers you have, the higher your bid density is. So just the closer every given piece of ad inventory's price is to its theoretically maximum price, like the most that anyone, any reasonable advertiser would pay, you get that network effect. And then if you have more advertisers and they're good advertisers, there is just more data being created on click-throughs and eventually indirectly on subsequent purchases. So it might have actually Weirdly enough, been a slightly like ATT might have actually been a slightly positive inflection for them long term. Maybe not in net present values, but in just absolute future value terms. Like it could actually end up being a bigger business because it would be more dominant in online market and online advertising or in mobile advertising than it otherwise would have been. Oh, and then Google. I like Google a lot. They I guess maybe maybe I should be long Google because they sort of have Microsoft in 2011 vibes of. So much great history, amazing businesses. You can still see why some things are growing, but it's hard to see what they do next. And they've tried some things that haven't necessarily worked. And there are other people who are building better versions of things that they came up with. I guess that's less of a Microsoft thing, but they definitely have that problem of all of the people who worked on attention is all you need were at Google. And then mostly they were Maybe entirely left to start or join other AI companies. So they, like, I think the business is an amazing business. Like search may just have the best economics of any business because you have so many different positive feedback loops where you get, if you have more searchers, you get more data. You can use that to better optimize your rankings. You get more ads. You can use the ads to buy more distribution, you get more searchers. Like all these things are feeding into each other and it becomes this incredibly ubiquitous, wonderful universal tool. So I think that's really interesting. And I actually suspect that Google shareholders will be pretty happy over the long term. But there's rarely a point where I look at Google and I'm like, this is an amazingly underrated company. It seems like a good but fairly rated company. And that's just a, it's a tough category to make money in because the problem is like, if you're absolutely right, you make slightly above average returns. And then if it turns out they do something really wrong or, you know, Bing somehow does take material market share in search or, or, or people change the way they search in a way that's bad for Google in absolute terms, or even they change the way they search in a way that's bad for Google's gross margins because everyone is demanding a Bard query instead of a normal search query and it costs an order of magnitude more. to serve it. Like, there are a lot of things that can go wrong. And if you buy a company that is priced for pretty good performance, and then it doesn't do quite so well, then that's a pretty painful position to be in as a shareholder.

Yeah. Which of, which of GAFA, you know, including, you know, Apple, Amazon, Google, you know, Facebook, Microsoft, would you, would you short if you had to? Which are you least long?

It's so tough because they're all great companies, extremely well run. You know, they're always cooking up really interesting stuff. If I had to short one, probably Apple. And my thinking there is that one, you just, you always have this hardware cycle risk where they do have to keep making better and better phones, but the replacement cycle keeps extending as the phones get better. And it's just hard to find New things people could do that would make them even better. On the on the services side, which has been the bull case for Apple for a long time, it just becomes more and more of a strategic risk for everyone else in the mobile ecosystem to let Apple always capture incremental revenue. And in a lot of cases, there's just not much they can do. But you know they're always plotting, and they're always thinking of what can we do to make it so the next time we're negotiating with Apple, it's not quite so painful for us. And I think that's just a tough position for anyone to be in. It's, they seem like much more reluctant than other companies to allow the creation of some mutually beneficial ecosystem. They seem to want a lot more control and they want a lot more of the economics. I guess Amazon has that problem too. But Amazon is also just pretty ruthless and is just much more, I think Amazon is much better at just testing a lot of different things at once and of launching a lot of stuff all the time and always having something that could end up being this next big source of long-term growth. So it's, I think, tougher to be negative on Amazon. But so yeah, if I had to short one, it would be Apple. But I don't see myself shorting Apple on the view of this stock's overpriced and it's going to crash. If I shorted Apple, it would be, I have too many long positions. I need to short something and I want to short something I don't think is going to, you know, rip 20% in a flat market anytime soon. So Apple it is.

I want to zoom out and just, because it's connected to the OpenAI, um, I want to talk about effective altruism movement because some people are claiming this is an EA coup. It's pretty fascinating how EA's brand over the past year has taken a bit of a nosedive. I feel like even before the SBF thing, you know, Will come out with this book and I feel like people were— they had a New Yorker profile. It was maybe getting too big for its britches or something. I feel like there were, you know, people who were unhappy with it both from the inside and the outside. I don't know, it kind of developed this a little bit of a stench or something, or maybe become too mainstream, too affiliated with the Democratic Party maybe for some people. But then SBF, you know, really was a negative, you know, hit to the brand because, you He had sort of been such a poster child for it and the values. It got kind of conflated with sort of relentless utilitarianism at all costs or something, including moral ethical lapses. And then them leading the charge on the AI doomer debate and EAAC sort of being a countermeasure that has a lot of sort of underground, underdog, aggressive, positive sum, alpha energy to it and just better memes, perhaps better Twitter Spaces. I don't know. And so I'm curious, what's your kind of reaction to my characterization of EA's sort of downfall in status among our sort of peers, I guess? And what do you think about that?

So EA was probably overrated. I wouldn't say a year ago, but like, A year and a month ago, so before FTX collapsed, overrated. I would say now they're underrated. Like it is. If you look at the early stuff they did, it was one. If you want to do good for the world, you should figure out what what does it actually mean to do good, and you should try to be rigorous about this. Like define what what you're actually trying to improve about the world, and then figure out how do you turn dollars into an improvement. You could you could view it as sort of part of the the neoliberal project of. subjecting more things to market discipline and saying it's not— when you give to charity, you don't feel good. You shouldn't feel good that you gave $100 to a good cause. You should feel good that your $100 bought X number of mosquito nets or dewormed a certain number of people or whatever. And I think that is actually helpful, that a lot of, um, a lot of nonprofit institutions just spin their wheels or, um, are a sort of accidental grift that is just not accomplishing very much for the world. And if you start measuring it, then you do some good. I don't know that— I do not believe Bill Gates identifies as an effective altruist, but he was actually pretty well-known, well-regarded for doing that kind of thing and for actually trying to measure outcomes and for doing things like doing edu— like running programs that were trying to improve educational outcomes, measuring whether or not they affected educational outcomes, cutting the budget because they had no effect on educational outcomes, et cetera. Like, that is, that is good. And you You want more of that in the nonprofit sector. But I think it gets difficult because if you embrace those premises and you start taking them seriously, you do start having these attractors again, where it's like there's— if you think there is some non-zero possibility of a really bad outcome and that outcome is really catastrophic compared to other things you're worrying about, you should actually invest more resources. And what I think might be a useful curative for that attitude is applying some kind of discount rate, some kind of uncertainty metric to those things. And they do try to do that. I think one of the nice things about EAs is that every criticism of EA that I have heard that is not a criticism of just fundamentally utilitarianism is wrong, which I think is a strong criticism.

Yeah.

True. Every criticism beyond that, everything about the implementation or the details of the strategies or the thought process, you will probably find a thread on the Effective Altruist Forum, or even more likely on LessWrong from a decade ago, from someone in the movement or in that sphere making that exact argument. And everyone, all of the famous EAs will have hashed it out in the comments section and will have Incorporated that critique into their thinking somehow. So they do, they do take ideas seriously. They do take this stuff seriously. But I think there's, uh, if you, if you are trying to optimize for this, um, this utility variable that cannot be measured by anybody, then, um, that may guide you towards doing genuinely good things. But because you're— because you can't actually measure the thing you're fundamentally optimizing for. It's hard to actually get that right. So yeah, I think EA is, you know, they've definitely had some negative headlines recently, but some, I think that they've been a helpful contribution to the discussion. I think they've probably, they've certainly donated more kidneys and probably saved many more lives than the vast majority of their critics. Actually, I don't know. I don't know of any prominent EA critic. who can credibly claim to have saved more lives than, you know, the typical famous EA has contributed to saving. So, um, I think that's actually something you should think about if you're criticizing EA really aggressively is like, how many, you know, are you using all of your organs right now? And is there someone else who could use them more? And, um, they, if you are not sure what step to take, contact your local EA. They will be really, really thrilled to help. help you. So, I think you should at least, you should at least have some response about why you want to keep your organs. If you, if you want to go too hard against the EAs, but I do, like, I do just fundamentally disagree that utilitarianism is a, like, works. And I think that since it's supposed to be, it's supposed to measure things based on how they work, that, that, that makes it difficult to justify.

Tyler Cowen in his book, Stubborn Attachments outlines his moral philosophy. And I think he calls it common sense utilitarianism. This idea of, hey, we should strive for the greatest good, or it's like a version of rule-based utilitarianism. And which, which means I ask you, if you don't believe utilitarianism is the right sort of framework, what is it? Is it a modification of it? Or is it something else?

I think like the lazy answer is the correct one that Moral intuitions are complicated. Moral quandaries are in fact quandaries. It is very, very hard. Like, I guess it's easy to come up with a straightforward ideology if you are willing to be completely insane and ruin your life. It's really hard to come up with a straightforward ideology that allows you to be a good neighbor, a good parent, a good friend, a good member of your community, and also perfectly morally consistent at all times with your stated values. I think it just doesn't happen. So you should actually look at moral philosophies that have been practiced over a long time and that have evolved over a long time. So you pretty much, if you find a viewpoint that worked, you know, like worked for people 1,000 years ago and caused them to live reasonably satisfactory lives by the standards of that time and to have kids who looked at the way their parents behaved and said, I would like to follow these standards myself, and those kids had kids, and so on. Like, there's a pretty good bet that this philosophy either has answers to most of the important questions you have or has developed some system for getting you in touch with people who have good answers. So, um, that is, that is what I think people should do, is, um, probably just choose the faith tradition you were raised in. If you were not raised in one, um, that's tricky, and you actually have to, have to do some But yeah, to generally follow, follow some kind of moral philosophy that's existed for, for many generations and just on the bet that it's, it's Lindy and will continue to exist for many generations beyond. Whereas like a lot of the new ones, they, they popped up, they did their thing for a while. It sounded really good to a lot of people. And then, you know, suddenly millions of people are dead and that ideology has been discredited. Like, you don't want to accidentally be like one of those. really, really peppy, excited Stalinists in 1930 who's talking about all the wonderful things that Comrade Joe is doing for all the poor Russians, and that it's terrible that people are lying and claiming that he's slaughtering Ukrainians by starving them to death because we know Joe would never do that. Um, yeah, you don't want to be one of those people, so probably, probably better to, to stick with the, the boring traditional answer.

Yeah. Um, Balaji, of course, uh, anytime he likes to discredit the New York Times, he talks about about Walter Duranty and their, their cover-up of, of the horrific acts that, that happened there. Well, on that, on that encouraging note, the— let's, let's wrap. This has been a fun conversation about what is perhaps one of the biggest corporate dramas that we've seen in quite some time that we are still living through, that maybe, maybe when we chat next week, we'll, we'll have some clarity on. It'll be fun to— Yeah.

Yeah, I can't wait. Things Things are developing and I'm really excited to see what happens next. And also, you know, it'll be nice when everyone gets back to work shipping cool stuff. Like, that's one of the things that makes it so exciting is like OpenAI is actually pushing out the production possibilities curve and they're actually making things possible that were not possible before.

Sure.

And, you know, they have a fast shipping cadence, good products. I use a lot of their stuff all the time. So wherever those people end up, I hope that they get back to Shipping soon.

Yeah, that's great news to close on. Upstream with Erik Torenberg is a show from Turpentine, the podcast network behind Moment of Zen and Cognitive Revolution. If you liked the episode, please leave a review in the Apple Store.