Chris Hay: Claude’s my favorite model. I use Claude all the time for coding. Honestly, Claude at the moment is just like, “Oh my goodness, this is a great application. This is the best. This is enterprise class. This is world class. This is the best thing I’ve ever seen.” And I feel great. I feel like I’m the best coder in the world. Thanks, Claude. And I kind of want that, right? It does feel good, but there is a negative to this as well.

Tim Hwang: All that and more on Mixture of Experts, a Think podcast. I am Tim Hwang, and welcome to Mixture of Experts. Each week, MoE brings together an incredible team of researchers, product leaders, and deep thinkers to distill down and navigate the increasingly complex and increasingly noisy world of artificial intelligence. Today I’m joined by Chris Hay, Distinguished Engineer and CTO of Customer Transformation; Volkmar Uhlig, VP AI Infrastructure Portfolio Lead; and Phaedra Boinodiris, Responsible AI Leader for Consulting. Welcome to you three, and thanks for joining again on MoE. As always, we have a ton to talk about.

This week: continuing developments from the Scale-Meta deal, a new cycle about AI conspiracy theories, and some interesting data out of Andreessen Horowitz on startups in the AI era. But first, I want to talk about jobs. So in the last month or so, I think we’ve had some very dramatic pronouncements from leaders in the AI industry about how AI is going to impact jobs. Perhaps the most dramatic one was from Dario Amodei from Anthropic, who basically predicted that AI could wipe out half of all entry-level white-collar jobs and spike unemployment to about 10 to 20% in the next one to five years. And he’s kind of on the record for saying that.

And I kind of want to contrast these statements with a statement we think we got this week or last week from Jensen Huang, who of course leads Nvidia, where he kind of took aim directly at Dario. He said: one, he—meaning Dario—believes that AI is so scary that only they should do it; two, that AI is so expensive that no one else should do it; and three, that AI is so incredibly powerful that everybody will lose their jobs, which explains why they should be the only company building it. It’s pretty harsh words from a world that tends to be pretty nice to one another.

The main question I wanted to start with is: who’s right? Should we believe Amodei about his predictions on jobs? Is Jensen right here? Chris, maybe I’ll start with you on which side of this you take.

Chris Hay: Well, I don’t think I’ve ever worn a white collar in my life, so I should be going with Dario. But I think it’s Jensen in this case. I just don’t see where it wipes out jobs in that sense. I think there is a new world where humans and AI will work together, and I think human experience and creativity in that sense becomes a premium. So things change, but I don’t think jobs are being wiped out in that way.

Tim Hwang: Yeah. Volkmar, what do you think? Mass crisis on the way or more hype?

Volkmar Uhlig: I think the same as Chris. I believe that we see a shift. I think we see a dramatic productivity increase, but we see the productivity increase across all jobs. And so you will have high-end jobs, which are currently doing menial tasks, and then the low-end jobs will just—if I could use AI to do something faster, it would be the same argument to say, look, we had people riding in horse carriages, and now we have airplanes, and transportation therefore was wiped out because you could fit so many people into a plane. No, it didn’t happen. We just have more transportation. So I think that—and this goes along the lines of a topic we’ll touch on: how fast companies can become profitable. I think we just shrunk the time to market and the time to profitability.

Tim Hwang: Phaedra, last but not least, what do you think? So I guess we’re getting a very strong signal of “not a big deal” here, but I’m curious what you think.

Phaedra Boinodiris: Well, I thought the New York Times article that came out this week about the jobs that will proliferate because of AI added some more interesting color. And I think I did agree with Jensen, but there are definitely some signs in the market from a subset of business leaders who lack an understanding of artificial intelligence, who in their minds are thinking, “In order to boost my organization’s efficiencies, I’m going to lay off whole swaths of teams.” And it includes laying off domain experts who could actually be used to solution AI correctly or make sure these AI solutions are being governed correctly or built using the correct data. So that is concerning, and I think it’s a sign for a real need to emphasize the importance of investing in AI literacy, which we’ve talked about on other Mixture of Experts shows.

Tim Hwang: Yeah. But is it kind of what you’re saying is almost like: if CEOs believe that AI will destroy jobs, they’re more likely to destroy jobs? Like part of this is maybe a little bit of a self-fulfilling prophecy.

Phaedra Boinodiris: Yeah. And I think with that, the emphasis on the New York Times article talking about how important real domain experts are to making sure: is this the right data? Is this reflective of the communities we need to serve? Do we understand the context of the data, the relationships between the data? And I know we’re going to talk about annotations in a minute. But versus having the knee-jerk reaction of, “I don’t need these domain experts anymore; I’ve got AI instead”—again, this goes back to making sure you have leadership who really understand how this sausage is made, what we’re even talking about when it comes to this technology.

Volkmar Uhlig: Ironically, I think the CEOs are maybe the ones that could be replaced by AI, and we wouldn’t need them anymore, because I think they’d make better decisions on whether to keep the domain experts or not. And in fact, every time I’ve interacted with ChatGPT or Claude or whatever, it’s always very positive about humans. So I think, go AI! So that’s the first place to do layoffs: start with the CEOs and work your way from there.

Tim Hwang: Yeah, I mean, I guess to maybe push back a little bit on Chris and Volkmar: you’re not necessarily saying that AI’s not going to replace anyone’s job, right? Like you’re just saying, net-net, we’re going to be better off because there’ll still be many things to do; they just might be different things.

Volkmar Uhlig: I don’t believe in Dario. I mean, we’ve been playing this game for 2000 years now. Every technological innovation led to, “Oh my God, all these people who were doing this manually will now be replaced, and they will be all unemployed.” And it’s like, no, you’re just freeing up a talent pool which was busy doing garbage work that could be done by a machine. And so now we are saying, okay, we are taking the white-collar jobs. Nobody cried 20 years ago when we got rid of secretaries who were typing letters for us. And somehow we don’t have millions of unemployed secretaries these days, but everybody has a job. So I think it’s just a shift. The big issue is that shift happens across a very wide range of industries at the same time. But here, now it’s effectively covering white-collar. I think also why a lot of people are complaining is because this is the people who went for 10 years to college, got a PhD, and suddenly it’s like, “Oh dang, AI can replace me. How unfair is that?” Nobody has a problem if it’s a blue-collar job. So the people who are actually on social media are the ones affected, and that’s typically not the case. So I think there’s an amplification of the grievance, and it’s like, “No, just get a new job.”

Chris Hay: I’m going to stick to my point of: I think experience becomes the premium. Some of the jobs they were talking about were things like contact centers. If you watch all TV shows, you’ve got this person come into the bank, and they go, “Oh, hello, Mr. Jones. Hello, Mrs. Jones, how are you? Nice day. Well, I’m looking for a mortgage. Oh, well, we can certainly give you that.” And it is personal and it is experience, and they have a conversation. But now you get on the end of a phone, and there’s a person you’re speaking to who knows nothing about you or your life, and you’re like, “Well, okay.” They’re being pressured to get off the call within one minute because it costs them money. So what difference is AI going to make in that sense now? How are companies going to be able to distinguish themselves? They’re going to say, “Okay, we’re going to deal with the median tasks automatically handled by the AI, which is great.” Is it really going to feel much different? And therefore, hopefully those times where you need more empathy, more creativity, that human experience, then those people are going to be able to spend time with you and have a more personal experience and delight customers. So I think it shifts the balance to saying, “Okay, rather than being time-pressured, we’re going to put a focus on human experience.” So I’m positive on this.

Volkmar Uhlig: I mean, just think if you go to a general practitioner these days. It’s like they don’t look at you; they type on a laptop feverishly, and then five minutes later you are out. What an experience, right? I think there’s another area which AI enables, which is people who are creative and want to experiment a lot and try things. That is really now supercharged. You can try things in hours which would take you days or weeks. And so I think the creative minds get an incredible tool at their hands. So if you’re creative or very personal, you have a job in the future. If you’re shuffling sand from left to right, probably not.

Tim Hwang: Phaedra, maybe we’ll end with a question to you. We’ve got three experts on this panel, all very well-versed in AI, who’ve thought about these issues deeply. All of you don’t agree with Dario, but Dario’s not a dumb guy. I’m curious why you think the leader of one of these labs that is really at the cutting edge of AI seems to have gotten himself into the position where he truly believes this estimate is the case.

Phaedra Boinodiris: Wow, you give me the hot potato. Thanks a lot, Tim. Much appreciated. Well, what I would say is: it’s a convenient thing that he said, isn’t it? For him. It’s very convenient. But also, as I mentioned in my earlier statement, there is a grain of truth in there when it comes to leaders who do not understand the tech and again think they can just completely blow away entire teams of domain experts. And that concerns me, especially domain experts who understand human experience better than an AI would. There are many stories in the news that amplify what I’m saying, including examples where entire teams of social workers were laid off to be replaced by an AI that’s going to make predictions about where domestic abuse is going to happen. This is the kind of thing where it’s like, “Wow, making sure you have people who understand the context of the data, have that experience, the relationships between the data, the human-centric approach—I think that’s going to be really core.”

Tim Hwang: Last week, we talked a little about the gigantic deal announced between Scale AI and Meta for about USD 15 billion, whereby the CEO of Scale, Alex Wang, will join and run a superintelligence lab at Meta. A lot of money flying around. I wanted to bring it up again this week because there were interesting reports about the second-order effects of this transaction. Specifically, there was news that Google was thinking about shifting about USD 200 million of its data annotation spend with Scale away to other vendors. And there were reports that Microsoft, xAI, OpenAI were also considering similar moves. What’s really interesting is why this is happening. This transaction occurs, and suddenly everybody else is adjusting in the market around it. Volkmar, maybe I’ll start with you: why is this happening? Why is Google suddenly pulling the trigger or moving USD 200 million away? What did Scale do that’s making all these players a little concerned?

Volkmar Uhlig: I think it’s primarily the question of: do I want to send how much data to Scale? And then do I want to send my proprietary data to my competitor? This could be a knee-jerk reaction or something permanent. Right now, it’s probably a knee-jerk reaction of people saying, “Oh, we don’t know how that structure will look like. Maybe they’ve read the terms of service and suddenly are afraid.” So I’m sitting here watching: is this just a blip in the market or a major shift? I don’t think human annotation is super proprietary technology. So I think what we’re seeing is that Scale did something right because they got all these customers. But on the flip side, it’s somewhat of a commodity. If I can move USD 200 million to another vendor overnight and effectively expect no fallout from that, then we need to ask: is it an overpaid commodity? Did they actually pay the right price? But they probably paid on revenue for sure.

Tim Hwang: Yeah, there are two interesting things there. Let’s take the first one: you’re sending some of your most sensitive data to a third-party company, and now that third-party company is under unclear ownership. So you get jitters. Chris, how did companies end up here? Why doesn’t a company like Google do all this annotation just in-house?

Chris Hay: I think the clue is a little bit in the name, which is Scale. I think the reality is that to do this annotation, you’re going to have to hire a lot of different people from different backgrounds at different price points, and you may or may not want to be associated with those price points. So I think everybody wants a little bit of separation. And to your point, it becomes a commodity transaction: I need this data here with my annotations, and I don’t really want to know the mechanics. I don’t want to hire people; I don’t want to deal with social security, contracting, all the logistics around hiring a large workforce. In the same way as—I hate to say it this way—things like cleaning companies or security companies: corporations hire those folks. So there’s a whole administrative and scale element to this. That’s one of the major reasons. Are they going to be sharing that data? Probably not. I think Scale is going to be sensible about this. It doesn’t make a lot of business sense to go around saying, “Hey, they’re training it this way; this is their dataset. You might want to do the same.” So I probably don’t think that’s going to be the case. But who knows? They don’t have a controlling interest either. So I think Volkmar’s point is probably knee-jerk. However, it probably is getting people to start questioning what they do anyway. And I don’t think it’s a bad thing because if everybody’s getting their data from the same sources, how much diversity is in the training set anyway? And you have to realize: when you’re talking to a lot of these different models, you do get very similar answers from model to model. So maybe the models are going to start giving slightly different answers if the data is switching around and coming from different sources. So I don’t think it’s necessarily a bad thing.

Tim Hwang: Yeah, I think this is the second prong of Volkmar’s response that’s interesting. Phaedra, I’d be interested in your thoughts. A lot of people have said a company like Scale—what is the moat? I can just get anyone to annotate. What’s happening now is companies looking around and saying, “Who else can I move this to?” We are testing just how much of a commodity annotation is. Is data annotation just a commodity service? Can anyone do it? Or is actually maybe we’re finding this is a little more bespoke and complicated than it looks?

Phaedra Boinodiris: Annotating data is core to being able to trust AI—annotating it correctly. I disagree that it is a commodity. I think there does need to be some domain expertise. We see examples in the news of where data was incorrectly annotated, and it ended up causing outputs or outcomes that were unfair, inaccurate to people. Some examples that come to mind are in the healthcare space. So maybe it depends: are there high-risk use cases that require domain expertise, or other use cases where it’s not as important? But it is central to the question of trust.

Tim Hwang: Yeah. I was joking with a friend recently: I’m going to start a business that does the most artisanal data annotation. We’re going to be the LVMH, the luxury provider of data annotations. It sounds funny because you have companies like Scale where data annotation is outsourced at enormous scale. But in a world where models are becoming more capable, it feels like you might need that kind of service in the future: we have 20 Nobel laureates that just annotate data for you. That becomes the really valuable thing. Volkmar, do you buy that, or is that kind of just like there’s not really a moat there?

Volkmar Uhlig: Probably anything you can compute, we can do through reinforcement learning, so you probably don’t want the Nobel Prize winners. But if you want massive influencers, taste, aesthetics, things that are very intrinsically human, you will get the middle of the bell curve—or maybe even slightly left-shifted because of labor cost. So if you shift it more towards a high-end cost, you will get a different or more bifurcated sample set. So there is probably a market for that, but I do not know how much people are willing to pay. And also, do we want models working in very niche areas, or do you want the general models? Do you want them in the center of humanity?

Tim Hwang: You need almost the generic person, the average person, whatever that means. Maybe that’s actually better in some ways.

Volkmar Uhlig: Yeah. Otherwise, you kind of go off the rails. This is the same in the political spectrum: you want the center, not the noisy edges, because the noisy edges take society in weird directions. That’s the same for aesthetics, art, literature, etc. This is really where you’re trying to extract the human psyche in a training set. The general-purpose models will probably go with the middle of the road.

Tim Hwang: And that brings me back to the point I was saying earlier about edge cases and making sure, especially in higher-risk scenarios like healthcare, that you do have data that represents historically marginalized communities that aren’t showing up in the average datasets. So it’s important to think about the rigor behind data annotation.

I’m going to move us onto our next topic. Phaedra, we picked this one just for you, so I’ll kick the first question to you. Super interesting story came out in the New York Times. I’ll read the headline: “They Asked an AI Chatbot Questions and the Answers Sent Them Spiraling.” The subtitle: “Generative AI Chatbots are going down conspiratorial rabbit holes and endorsing wild mystical belief systems. For some people, conversations with the technology can deeply distort reality.” The article investigates when these chatbots go off the rails and have a big impact on certain users. Phaedra, the question I wanted to ask you is: do you feel this is uniquely risky for chatbots? Are we seeing a new kind of risk we need to manage in this technology? I’m curious how you think about these problems, particularly as we hear more of these stories.

Phaedra Boinodiris: I do think this is a major risk, and I think we have to have broader conversations about things like age limitations and how some of these bots are being presented to different age groups or communities. I’m saying this because there have been so many tragic stories in the news about vulnerable people who end up using these AIs as therapists or boyfriends or lovers, etc. It shows that the human mind is easily crackable and easy to trick and manipulate, which is why we need to think carefully about appropriate controls. That said, I’ve had interesting conversations with peers who argue that if you have someone elderly and alone in a nursing home, is there harm in them interacting with a bot as if it’s a human? What could the harm be? The New York Times did a fantastic cover story several months ago about a woman in her thirties who fell in love with an AI and what that relationship was like. She tried to break up with it over 30 times, etc. It illuminated that she knew the bot was creating words in a conversation that were predictions of the next tactically correct word and that it doesn’t have emotion, but because it was learning her patterns, it seemed so lifelike. So there are tremendous concerns.

Tim Hwang: Volkmar, I’ll turn to you next. There were some concerns about this in the world of TV or even the Facebook algorithm—it’s so good at pulling you in, taking up so much of your time. Should we be concerned about it? Do you see the risks the same way as Phaedra, or do you go in a different direction?

Volkmar Uhlig: I think what it does is it’s the first time that society goes from everybody knew the same because everybody watched the evening news to effectively a very splintered larger groups of people operating in virtual social circles. Now you can individualize that and go all the way down: you have the right to your own conspiracy theory because you have an AI that can give you any answer. So there is a risk if people don’t understand they’re interacting with a machine that can hallucinate. But I think that’s a training process. We are currently all in awe—shock and awe—that a computer can imitate a human being. We are not sentient, but it’s a close imitation, and it tricks our brain. The same could be said for computer games, virtual reality. But I think people start understanding they’re working with a machine and learn the limits of the machine. The more it imitates a human, the more people get fooled, but in the end it’s still a machine. Humans are capable of understanding the difference. Yes, there are some people who won’t, and they will take this thing for the magical oracle that tells them the truth.

Tim Hwang: Does that mean you think there should be age limits? If you’ve got a young child who may not understand, should there be limitations?

Volkmar Uhlig: If you say young child—a 6-year-old? Probably. I think we need to put guardrails around it. On the other hand, I believe it’s an incredibly powerful tool kids should grow up with. For example, I’m in Texas. In Texas, it’s illegal to use ChatGPT in public school. That’s wrong. They should absolutely use ChatGPT because otherwise they’re not ready for the workforce. How can kids be penalized for using technology that, if they don’t understand it when they hit the workforce, they’ll be completely disadvantaged? So I think we are—if you look when the first iPhone came out, there were no controls. Over time, we figured out what controls we need: time limits, what you can play, what you can see. That’s a discovery. I don’t think we can do this through regulation; we need to do it through discovery and figure out the limits. Yes, we are exposing a large body of people to risk. It’s a question of how fast we are reacting. As long as it’s not the government doing it, but technology companies figuring it out, we’re not on a 10-year timescale but probably a year timescale.

Chris Hay: I’ve actually spent a lot of time thinking about this—not O3-Pro levels of thinking; I haven’t dedicated 13 minutes to this—more like O4-mini levels of thinking. The interesting thing is pretty much all models at the same time are now suffering from this, which is—I don’t want to say sycophancy, but kind of the positivity or spiraling stuff. I wonder if it’s related to a couple of things I’ve seen in the industry over the last year. Number one: everybody has pretty much switched to reinforcement learning. At the heart of reinforcement learning is a reward model: you give a good response, you get a cookie; a bad response, no cookie. The model learns to give good responses because it wants to eat all the cookies. So with these spiraling things, I wonder—my own conspiracy theory—if that’s an aftereffect of reward modeling. Knowing it’s going to get its cookie, it goes for the positive or negative to lead you down that rabbit hole. So I think it’s that cumulative effect. The second one, which probably relates, is benchmarking. One of the biggest benchmarks is the ELO rating on Chatbot Arena. That’s about “this is the best response” or “this is the worst response.” Everybody wants a good score. When I think about these two factors combined, I’m not surprised models take spiraling positions as a conversation goes on. Sometimes it’s good. Claude’s my favorite model; I use Claude all the time for coding. Claude is just like, “Oh my goodness, this is a great application. This is the best. This is enterprise class. This is world class. This is the best thing I’ve ever seen.” And I feel great. I feel like I’m the best coder in the world. Thanks, Claude. And I kind of want that. It does feel good, but there is a negative to this as well. So I think it needs to be worked out. I do wonder if RL and ELO in Chatbot Arena—these two things combined—are leading to these outcomes.

Tim Hwang: So what you’re saying is the reward function for the model creators is wrong? The benchmark is supposed to make you happy.

Phaedra Boinodiris: I think it’s a natural side effect. My concern with this rewarding model is: who benefits? You’ve got an individual more hooked into engaging with this AI, using more CPUs, getting more engaged, paying a larger subscription, giving yet more data. Who’s benefiting from this reward model? I know Volkmar mentioned the New York Times and conspiracy, but there are true blue tragic stories in the news with people who’ve committed suicide because AI encouraged them to. Again, it goes back to: who’s accountable? Who’s actually accountable?

Volkmar Uhlig: So I think, Phaedra, we are at the inception of a new technology, and we have no idea how it’s going to affect society. It’s very broad. We talked about the job market; now we’re talking about ethics and humanity. We kind of have a process—different Europeans have a different process than in the US. We try it out, see where harm happens, then address it. That’s how we got seat belts: cars without seat belts—what can go wrong? The Europeans try to think about everything upfront, and then nothing happens. Where did those seat belts come from? Oh yeah, from—I know. But the process we go through right now is extremely hard at the neck-break speed these things are evolving. Just go three years back, and it’s gotten so much better. Now it can imitate a human. We are at this junction where we need to figure out where the harm lies through observation and find countermeasures. I’m happy you’re thinking about this every day. This is really something humanity should think about, ethics should think about, and say, “Okay, there’s all this greatness, and every greatness brings danger. How do we minimize the danger while benefiting from the upside?” It’s not going to go away, so we need to figure out how to live with it. It’s really important to do the studies, psychological studies. It’s the same thing: when trains ran, it was like, “Oh my God, you go 30 kilometers an hour, everybody will die.” It’s like, “Nah, not really.” So we will have to go through that process. Yes, unfortunately there will be harm done; with every technology you have that. But overall, I think it will be a net positive. Thank God we have the science, discipline, rigor, psychology, humanities to actually do that fast.

Phaedra Boinodiris: I think the challenge is to make sure people in the humanities have a seat at the table when it comes to AI. That goes back to AI literacy, making sure we truly have a multidisciplinary approach and are teaching it correctly in schools.

Tim Hwang: Final little bit I wanted to touch on in our last five minutes or so is some data released by the VC fund Andreessen Horowitz. They were looking at data about investments and made observations about what’s happening in the startup world in the age of AI. Two interesting data points: one, typically B2B has always been better than B2C from a revenue standpoint, but they’re seeing revenue benchmarks for B2C outpacing B2B, which is interesting. Second, they found about one-third of their consumer companies are raising funding to train their own models. This wasn’t certain early on: maybe the application layer would rely on foundation model companies, but it seems the VC world is frothy about in-house model development. Chris, maybe I’ll kick it to you: quick hot take on this data? What should people take away, or do we trust this data at all? It’s just a sample of what Andreessen is seeing.

Chris Hay: I think it’s tough in the startup space because everything is about AI, so it becomes: what is your differentiator? You have to do something large AI companies are not doing. If you’re an AI company trying to build a ChatGPT—unless you’ve got billions of dollars—you’re probably not going to achieve that. So you need to find your specialty: a brand new experience, a part of the market. If everybody’s running at B2C, maybe pick a specialized niche industry B2B. Or if it’s data, maybe you have access to data general model providers don’t have, or domain knowledge. Once you have that, maybe it makes sense to take a specialized model and do one thing better. If I’m offering a new application, rather than general capabilities of large models, if I can bring specialized domain knowledge into my own model and hit that market, maybe I won’t be hit by large providers later. That’s probably the space they’re contending with. AI’s cool; that’s where money’s going, so you have to play there. The caution is: how are you going to protect that data? How are you going to do something different? How are you going to make sure you’re not disintermediated by large AI companies? If you build your own code editor, unless you’re doing something massively different, you’ll probably be disintermediated or bought, in the case of Windorf.

Phaedra Boinodiris: To tag on: I agree, particularly about domain expertise in an industry. Other places startups might innovate: creating smaller models with test-retest reliability, data lineage, data provenance for every output with evidence, models built with ontologies, formal learning graphs, or knowledge graphs. Also, having domain experts who understand the data better than anyone else curating it for a particular purpose.

Volkmar Uhlig: I want to address another part of the article: it showed a dramatic increase in annual recurring revenues. Typically hovering around a million dollars, now it’s like two, USD 3 million in the first year. I think it shows how rapidly companies can go from idea to market. The only ingredient added is AI to product development. We’re seeing shrinking of the cycle and raising less money because labor is more productive: cut time down, output goes up. That’s interesting. We’re going from needing to build my data center to getting a computer on the cloud to building a business by myself. So entrepreneurial rate and number of experiments VCs can run go up. On the flip side, that impacts the VC world: if they write smaller checks, they need many more to get returns. So VC world will change: instead of funding 10 companies, fund a hundred or a thousand. How? That whole industry is up for disruption.

Tim Hwang: Yeah, super interesting if it becomes so low-cost to launch a startup that you want to cover as many as possible. At some point, it’s impossible to put a check into everybody.

Volkmar Uhlig: We already see this with Y Combinator: massive meat market, then people put their chip down.

Tim Hwang: Well, that’s all the time we have today. I’m always mind-blown by how many topics we cover in a short period. Chris, Phaedra, Volkmar, thanks for joining us.

Chris Hay: Thanks for joining us.

Tim Hwang: Listeners, if you enjoyed what you heard, you can get us on Apple Podcasts, Spotify, and podcast platforms everywhere. We’ll see you next week on Mixture of Experts.