The Exchanges

Q What is the anthropic story? Can you summarize for us?

A Oh my God, what is the story? Um, the anthropic story was And I'm probably not the best person to tell this story, but a few people left OpenAI, wanted to do, um, a foundational model company somewhat differently, and they built a, and both of them hadn't figured out a business, and now Anthropic has sort of built a model that eventually became very good at coding, and it got a ton of recognition for that, and then it sort of Now, Anthropic, I would say, is a company that's, you know, trying to solve AGI, but they believe the route there is through reasoning, coding, and things like this. Um, they very distinctly do not really do much outside of text-to-text models. Um, so less audio stuff, video stuff, um, things like that, and that's really their primary focus, and in terms of a business, they just serve enterprise customers primarily. They do have a consumer app, but that's not, not really the, the prime business.

AI assessment note: “a few people left OpenAI, wanted to do, um, a foundational model company”

Q What is the anthropic story? Can you summarize for us?

A Oh my God, what is the story? Um, the anthropic story was And I'm probably not the best person to tell this story, but a few people left OpenAI, wanted to do, um, a foundational model company somewhat differently, and they built a, and both of them hadn't figured out a business, and now Anthropic has sort of built a model that eventually became very good at coding, and it got a ton of recognition for that, and then it sort of Now, Anthropic, I would say, is a company that's, you know, trying to solve AGI, but they believe the route there is through reasoning, coding, and things like this. Um, they very distinctly do not really do much outside of text-to-text models. Um, so less audio stuff, video stuff, um, things like that, and that's really their primary focus, and in terms of a business, they just serve enterprise customers primarily. They do have a consumer app, but that's not, not really the, the prime business.

AI assessment note: “a few people left OpenAI, wanted to do, um, a foundational model company”

Q I mean, the ambition is definitely there. What can India do to get competitive in this realm?

A I think one of the things that is I'll say an observation first and I'll answer the question. I think one of the observations I have in general about how, how India reacts to technology is India reacted to AI like, I want to be a part of this well after this was a thing. And that general mindset will rarely make you win at a thing. Um, and so if you compare that to China, for example, I mean, There are many differences. I mean, China has had so much local talent, so much development in math, so much research in AI for many, many years pre-LLMs that allowed them to then capitalize on that and go like, okay, look, we can probably do something here too. We have the base built out. And I think for India to do something in, in AI, like, I don't know if a lot of that base is built out. There's a huge talent leakage problem. Most of the top AI talent in India is here. Right. And we're literally a three square mile radius from where we are. Um, there's a talent problem. There's a, there's a comp problem. Why would I, AI researcher want to be paid in rupees? I don't, I want to be paid a million dollars or a hundred million dollars like they're getting paid here. Um, and then there's a general systemic problem on like, on account, but just capital is a little bit harder to get. And then resources are harder to get. And then people haven't done work on GPUs in India. If you go to, if you …

AI assessment note: “I'll say an observation first and I'll answer the question.”

Q I mean, the ambition is definitely there. What can India do to get competitive in this realm?

A I think one of the things that is I'll say an observation first and I'll answer the question. I think one of the observations I have in general about how, how India reacts to technology is India reacted to AI like, I want to be a part of this well after this was a thing. And that general mindset will rarely make you win at a thing. Um, and so if you compare that to China, for example, I mean, There are many differences. I mean, China has had so much local talent, so much development in math, so much research in AI for many, many years pre-LLMs that allowed them to then capitalize on that and go like, okay, look, we can probably do something here too. We have the base built out. And I think for India to do something in, in AI, like, I don't know if a lot of that base is built out. There's a huge talent leakage problem. Most of the top AI talent in India is here. Right. And we're literally a three square mile radius from where we are. Um, there's a talent problem. There's a, there's a comp problem. Why would I, AI researcher want to be paid in rupees? I don't, I want to be paid a million dollars or a hundred million dollars like they're getting paid here. Um, and then there's a general systemic problem on like, on account, but just capital is a little bit harder to get. And then resources are harder to get. And then people haven't done work on GPUs in India. If you go to, if you …

AI assessment note: “I'll say an observation first and I'll answer the question.”

Q So when everybody runs out of data, they create synthetic and keep improving the models?

A It's not exactly like that. I mean, in reinforcement learning, you basically want an environment where there is certain rewards you can get. Sometimes they're extremely verifiable rewards, in which case it'd be RLVR. Sometimes they're more nuanced where it's like text, like how do you verify that? But there are You can use an LLM as a judge to verify that. So there's various kinds of rewards, but it's, you know, one way to think about it is, yes, you are generating an environment where there's rewards, and you're making the model become better at that task. Yeah. In a similar way that you would do, for example, when Google did AlphaGo, and you train an AI to learn a game, you have a reward, which is winning the game, and then you figure out the steps in the middle. Um, so it's not really synthetic data, um, per se. That is also used, but it's a different Uh, step. But that's, that's what I meant by reinforcement learning. Um, let's see what else. Number three I think is not talked about enough is, um, China, which is, you know, the, the, the narratives around China are interesting, um, because most people in America, I feel, don't want to even recognize how good these models can be with Perhaps far less capital expenditure, right? Like they're, they're not spending that much for talent. They're not spending that much in even GPU resources. They don't have the latest cutting edg…

AI assessment note: “It's not exactly like that... so it's not really synthetic data, um, per se.”

Q Forget the venture lens. Okay. Just as an investor or somebody starting a job in an industry, which do you think has the most headwind in the next 10 years? Like somebody emails you from a particular industry, you'll be like, okay, I'm not going to open this even.

A I mean, I don't, I don't think I have a very strong bent against any particular industry. I'm trying to say, like, all industries are actually pretty important. Um, there are some that are maybe falling now, but it's impossible to know how, how they would do in, in 10 years, right? Like you, like, I'll give you an example. Maybe it's probably not really gonna answer your question, but if you take something like manufacturing, which is a pretty traditional industry, which may not have much Take us in, in the venture community. There are some fantastic manufacturing businesses that are going to be built, um, especially in the next 10 years that have a whole other set of interesting problems to solve. Like, how do you solve, um, a labor problem? Um, how do you manufacture in certain countries? If, can you even do that? What does tariffs mean for manufacturing in different places in the world? So I, I think there's interesting businesses in all sectors. It's really hard, and I don't know if, I'm curious actually what you guys think is, Often when you, you, you have to have an open mind as an investor to hear out the idea in a certain sector, because often it's from the places you least expect it that an interesting idea shows up.

AI assessment note: “I don't think I have a very strong bent against any particular industry.”

Q You guys want to comment on senior living? Would you send your pens?

A Um, it's a tough one. Um, I would, There is some bias against it as, as you said, but you know, maybe that decision will change over time. I think it depends on what they want to do as well. Um, I, I do think in my particular case, I'm lucky that they have a good community already at home. So that does solve a bunch of the problems, but you know, who knows how this goes. Um, but generally on your question, you know, in the U S very much, um, in the rest of the world, slightly less. Um, just because there's different dynamics, social inter, intercultural dynamics in different parts of the world. There is also taboo and things like that. So I, I'm not as sure, but in America, hi.

AI assessment note: “I think it depends on what they want to do as well.”

Q You guys want to comment on senior living? Would you send your pens?

A Um, it's a tough one. Um, I would, There is some bias against it as, as you said, but you know, maybe that decision will change over time. I think it depends on what they want to do as well. Um, I, I do think in my particular case, I'm lucky that they have a good community already at home. So that does solve a bunch of the problems, but you know, who knows how this goes. Um, but generally on your question, you know, in the U S very much, um, in the rest of the world, slightly less. Um, just because there's different dynamics, social inter, intercultural dynamics in different parts of the world. There is also taboo and things like that. So I, I'm not as sure, but in America, hi.

AI assessment note: “I think it depends on what they want to do as well.”

Q So when everybody runs out of data, they create synthetic and keep improving the models?

A It's not exactly like that. I mean, in reinforcement learning, you basically want an environment where there is certain rewards you can get. Sometimes they're extremely verifiable rewards, in which case it'd be RLVR. Sometimes they're more nuanced where it's like text, like how do you verify that? But there are You can use an LLM as a judge to verify that. So there's various kinds of rewards, but it's, you know, one way to think about it is, yes, you are generating an environment where there's rewards, and you're making the model become better at that task. Yeah. In a similar way that you would do, for example, when Google did AlphaGo, and you train an AI to learn a game, you have a reward, which is winning the game, and then you figure out the steps in the middle. Um, so it's not really synthetic data, um, per se. That is also used, but it's a different Uh, step. But that's, that's what I meant by reinforcement learning. Um, let's see what else. Number three I think is not talked about enough is, um, China, which is, you know, the, the, the narratives around China are interesting, um, because most people in America, I feel, don't want to even recognize how good these models can be with Perhaps far less capital expenditure, right? Like they're, they're not spending that much for talent. They're not spending that much in even GPU resources. They don't have the latest cutting edg…

AI assessment note: “It's not exactly like that... so it's not really synthetic data, um, per se.”

Q Forget the venture lens. Okay. Just as an investor or somebody starting a job in an industry, which do you think has the most headwind in the next 10 years? Like somebody emails you from a particular industry, you'll be like, okay, I'm not going to open this even.

A I mean, I don't, I don't think I have a very strong bent against any particular industry. I'm trying to say, like, all industries are actually pretty important. Um, there are some that are maybe falling now, but it's impossible to know how, how they would do in, in 10 years, right? Like you, like, I'll give you an example. Maybe it's probably not really gonna answer your question, but if you take something like manufacturing, which is a pretty traditional industry, which may not have much Take us in, in the venture community. There are some fantastic manufacturing businesses that are going to be built, um, especially in the next 10 years that have a whole other set of interesting problems to solve. Like, how do you solve, um, a labor problem? Um, how do you manufacture in certain countries? If, can you even do that? What does tariffs mean for manufacturing in different places in the world? So I, I think there's interesting businesses in all sectors. It's really hard, and I don't know if, I'm curious actually what you guys think is, Often when you, you, you have to have an open mind as an investor to hear out the idea in a certain sector, because often it's from the places you least expect it that an interesting idea shows up.

AI assessment note: “Maybe it's probably not really gonna answer your question, but if you take something”