From Classroom to AI Lab

Transcript generated automatically from the episode audio and lightly cleaned up (filler words removed, misheard names corrected where clear). Speaker labels were added afterward and may contain errors.

Rob May: Hi everyone. I'm Rob May, the co-founder and CEO of Neurometric.

Ryan Eppley: And I'm Ryan Eppley, co-founder CEO of RootAccess.

Anna Kirk: And I'm Anna Kirk, sales lead at Thread AI.

Rob May: The next and this is the AI in NYC podcast. This is our second show with our sponsor BePresent. BePresent is an app that you should go download because they take the same tactics that social media uses to keep you addicted and they use it to keep you unaddicted and keep you off your phone. So we weren't you weren't here last time, Anna. Are you a doom scroller?

Anna Kirk: I'm an absolute doom scroller. Yeah. Okay. In the car right over here. Yeah, I was I was trying to tell Rob that I am monitoring the situation very effectively.

Rob May: Download BePresent so you don't end up like Anna. That is going to be our

Ryan Eppley: I'm actually I'm the opposite of that. I do. Yeah, I'm a horror story by the way, but I'll like listen to podcast as I fall asleep. Sometimes it's AI in New York in New York. Yeah. 6:30.

Rob May: You're downloading it as you sleep via osmosis.

Ryan Eppley: Like I need a constant stream of stimulation. Yeah. All right. To stop me from thinking my own thoughts.

Rob May: All right. So our guest today is Albert Chun from AI Circle. Albert, welcome to the podcast.

Albert Chun: Thank you so much for having me.

Rob May: So before we talk about AI circle, I want to jump into your background because you worked at a company that I was an investor in for a long time called Invisible, which I think was like the first real scaled AI automation company. Yeah. So tell us tell us a little bit about like what they did and how you ended up there. And then we'll talk a little bit about some some lessons learned.

Albert Chun: Sure. So I think primarily their focus was twofold. One was human data in training state-of-the-art LLMs and then the other was bringing AI to enterprise. And I actually joined them. I actually encountered Invisible because I was at an intr company. I was managing a little over 200 people and I was taking care of all of these service operations work and managing books of business and all these things and like over a dozen acquisitions and I wanted a way to solve the human problem like all of this could have been automated. I got on a call with Invisible and honestly the pitch was so good my fear was I was like you're going I'm gonna have to fire like 180 people and so I didn't bring them on cuz I like it would be too good.

Yeah. And then which is yeah for better or for worse that's just how my mind was thinking and later on one of the adviserss at Invisible Dave he Dave Klein he was at Bridgewater with a bunch of the alums who ended up yeah like Ben and Lauren a bunch of people who were the executives at Invisible were a Bridgewater alum and he made a referral went through 10 interviews and got the job.

Rob May: Wow. Interviews. Mhm. And Invisible when you were there took workflows, had humans did them, but was constantly trying to figure out how to take what the humans were learning and add more AI.

Albert Chun: Exactly. That that was actually something that was really attractive to me is that they worked with a business. So they're saying, "Hey, initially, you know, they were working with mortgage brokers or or DoorDash or whatever it was, and they would take their boring workflow and automate it and saying, "Hey, like we should be incentivized to do this faster, more accurately, and less manually." And that made a bunch of sense and they would they trusted that they would figure it out over time. And I imagine they wanted to productize those learnings and you know if they built up a certain vertical they could rinse and repeat and sell that workflow or that automated workflow to a bunch of buyers.

But I was specifically brought in to take on one of their LLM accounts and to fine-tune their models and ended up working with depending how you call it around like 60 or so different campaigns to train up these LLMs into like in one of those niche expertise. So like multilingual coding, competition math, natural language, whatever it was. Yeah we would yeah deploy several hundred people

Rob May: and yeah this was building pre-training data sets or this was

Albert Chun: post training post training training yeah pre-training man pre-training has so much potential it's so simple and so all the things but that that was never part of our work I always thought it it was great like you diversify you de-risk your business you get into a different budget you get to you to trade lowquality tokens for high quality ones and just do a bunch of remarkable things, especially because common crawl is such a show and you know like everyone's working with the same base and you're investing all this money downstream to make it better and better. Like why don't you just invest a bit more upstream? But we could never quite make it work out. Yeah.

Rob May: Okay. Yeah. Cool. What questions do you guys have about Invisible before we jump into AI circle?

Ryan Eppley: I mean I think the hardest thing that I'm when I was doing some research before you came on the show is just you know how were people explaining AI to your first early customers because I think like AI wasn't a thing. It was more so like hey we have these like really complex ML algorithms and it does some magic and they're like oh like automation you're like yeah but like a bit more under underpinning that like how did you make business sense of that?

Albert Chun: This is such a great question. This is actually how and why I started AI Circle because like one I came in as an outsider, right? I was in I started a bunch of schools in a former life and worked in intrate tech and all this stuff and I stumbled upon like AI training, human data. And when I first came in, there were these huge intimidating acronyms like RLHF and SFT.

And I just asked everyone. I was like, "What does this mean?" And no one could. Everyone gave me a different answer. And I was like, "This this is so dumb that in one in the same company, you can't tell me what we're doing." And I'd get different answers. And I and candidly, when I'd go online, I wouldn't get a satisfying answer either. Yeah. And so I thought, okay, in any business, what you need is common language. You need common language internally. Any common language between business and tech and then tech and client and whatever. And so I thought okay this is a signal that like how much how many problems need to be solved here and how much need there is for that common language and you know by extension I'd argue community and so what I ended up doing was meeting with as many people as possible to understand how people were talking about AI seeing it what made sense what was what was nonsense what was actually moving the business forward

And so I ended up allocating about 20% of my time. I did I mean I did my job well. We 5x the business and got to 50 million revenue on my account alone. Ended up advising for almost every account. And so and brought in a bunch of leads. Anyway, we did right by you, Rob.

Rob May: Good. But I well I don't know if I maybe I've told the story on here. I like I sold early when when when Francis came out and took out some debt and bought and offered a decent valuation. We were like, "Yeah, it was a 10x for us." Yeah. So we took it and three years, two years, three years later, they were raising at 10x of that and doing a secondary. And I was like, "Oh man, I should have It's all good." But yeah. Yeah.

Anna Kirk: What was the pitch that Invisible gave you that made you feel so confident that you wanted to jump ship and join them?

Albert Chun: It was maybe less about their pitch and more about my own research. So I'm a person where I like to go to a place where the ceiling is very high and I like to see how high I can go which is like education I want to go to edtech or education I want to go into financial services or you know came in managing three people I want to manage like half the company. And so like invisible was the same or AI was the same way. I had this like this was in 2024 Q3 so like less than a year after ChatGPT 3.5 came out and again there wasn't like a lot of like like confidence in AI and if there was it was like everyone was confident that it was going to destroy us all right so I was like okay do I want to poses this challenge of hey a really high ceiling of all the things you need to know and the consequences of not knowing or handling this well could be catastrophic The question was, hey, do you want to sit on the sideline and complain about it, say, oh, all these nerdy tech people like ruining our lives, or do you want to be a person in a place of influence to make sure it doesn't go all the way south and I've said, let's like be in the thick of it. Yeah.

Rob May: So talk talk a little bit about that that that point that you just made which was like you came out of an education background. Which is not where most AI people come from. They come from like you know math, physics, computer science or whatever. And so how do you think about how's that influenced your understanding about how humans learn and has that translated into how you think about training AI models and whether this is a pedagogical kind of thing or whether because you most of the people look at it as a mathematical thing to get us there. That's all they focus on like what's your point of view now?

Albert Chun: So much it's like not even funny.

Rob May: So share a little bit about the ed ed background.

Albert Chun: So I studied education at Cal and then I started a school in the Bay Area got to number one and then started middle started that middle school ended up becoming opening up an elementary school in high school. Ended up teaching at a chord school a whole bunch of places. Start a robotics program and the son of anyway a bunch of like worldrenowned VCs and tech CEOs were part of this and I just didn't know who these people were as a seventh grade teacher and I was like oh wow these are like remarkable people but yeah and so when I was doing that leading schools starting schools end up starting a couple schools in New York City in the South Bronx started a dental office a medical health clinic, a mental health clinic, and all these things.

And at both of those things, a key part was actually I helped write some of the state standards in California, and then helped test the state exam in New York City. And those things were really important because I had to think of a rubric and a guideline and what's the goal of the state of California. And then on the this other side, I need to test the test to make sure it's actually working. And then I have to move and leverage all these people and shephering cats to like hurting cats to do all of these different things. In the same way it was like okay the through line to managing a workforce of my team ended up getting up to over 700 people and coordinating all these different campaigns with each with their own rubric schema guideline and how do you make it so that it increases signal so the model outputs get better and the loss function decreases and all these things like that required a lot of just like first principles thinking. Yeah. It's align on what good looks like. Let's train all these people and onboard them. And yeah, in other words, like it was insanely informative to the way that I lead a human data team. Yeah.

Ryan Eppley: So now actually I feel like the throughine between education and model training and actual implementation of those models is even stronger with like what what do you think that education will look like going forward as a former educator yourself? Like how would you have built a school today?

Albert Chun: Man, I'll be honest with you, I've avoided this question because it is so difficult. Like I don't have a good answer. I will say just, you know, I have kids of my own. My take on education is just say I'm like highly considering homeschooling. Yeah. Because I think education US just generally is I mean what are we 28th in the world or some unremarkable thing and so like just in and of itself it's not great and then you throw AI which throws everyone in a loop and teachers who Anyway, yeah I'll just say yeah we just have we have we just have so many problems in education. Will AI solve it? Will it exacerbate things? Like I don't really know. My sense is it won't get better. I think the part that I'm actually more worried about is less like okay here what are some good things. AI in general raises the floor.

Yes. Right. Which is incredible. You see the biggest lesson in education they're not happening in the US. They're happening in different parts of the country. I mean but different parts of the world like Africa, South Asia and in these different places. Wonderful. Love it. Keep it going. The part that I worry about is the social impact, right? Like, okay, so let's say you have kids who are always on character AI or whatever companion friend app and these things that seek to reduce isolation when in reality it exacerbates it. And I'm keenly worried about that level of disconnect and abstraction from like real world things, you know,

Rob May: and so you're making the case for be present, our sponsor.

Albert Chun: So 100%.

Rob May: Well, I'll tell you. So, so I have teenagers and it's funny. We just had a conversation about they're like, "Oh, Dad." Yeah. Cuz I'm like, "Oh, they they have these things that test to see if you use the ad to generate this." They're like, "Oh, yeah, Dad." But here's how you write your prompts. You tell it you like you tell ChatGPT, hey, this is going to be tested to see if it's so make it sound more human and it it doesn't and that's how you get around the testing thing. So like and for me as a parent I'm like that kind of hack is what I want you to learn as a kid, right? So like that's actually the most important skill I think.

Albert Chun: Yeah. Yeah. I think you know for me like and that's true, right? So if like if the test is no longer can you write something well like what's the test it's like can you like one one school that I look at is I think it's called Ad Astra Elon Musk's micro school where they focus a lot on like collaboration problem solving and things like that and so I actually took their assessment test it is very difficult in that it's it tries to see how you solve problems and how creatively approach it and how quickly can you recognize patterns in order to pass this this game so to speak without instructions.

This was the particular assessment that they had that seems very interesting in terms of how I think people succeed in the world. Do you work well with others? Can you figure things out? Can you problem solve? Can you work collaboratively and work remotely and do all these things? That that sounds very Are you entrepreneur? All all these things I think are like I would want to index on with my own children. Like like ChatGPT can't can help you show you a blueprint for how to start a business, but it's not going to teach you to it can't instill in you the courage to go up to a door, knock on it, and say, "Hey, I'll mow your lawn for 20 bucks." Yeah. Yeah. You know, so those are the kind of things I think school could really like if I were to swing on the other side of the pendulum, it would be to absolutely embrace those kind of like character course like character building skills. Yeah.

Ryan Eppley: So now let's let's talk about, you know, what you're up to now in building probably one of the world's largest communities for, you know, not just AI experts. Yeah. You know, but now it's like almost every business has an AI plan, right? You know, how have you managed such an array of technical talent with business talent and make sense of the communities that you run and the online kind of engagement that you're that you're putting together? Like, how have you really found success in putting together such a wide array of technical talent with business talent?

Albert Chun: Yeah, I'd say one thing I actually like to underscore is like we're not the largest and we don't want to be the largest. I have no I have no desire to be to say, "Hey, we're a million community members." Like, I would rather be known to say, "Hey, we're a group of people who know what we're talking about, who are actually doing like actually practicing the work instead of talking about it or just writing about it or, you know, engaging in hearsay." So, we're we're only 500 members and I know there are many communities you've done a lot more and that's wonderful and they serve their niche and all the things.

Yeah, I'd say to go back to what I offered about when I was invisible carving out the 20% time like in that year I probably had a little over 180 coffees with people, right? Like met in person every X day. I would go into the city, put up a wall of meetings and just meet with as many people as possible to understand how deals were won and lost. Like what's your take on RL or post training or pre-training? Does this even make sense to you? Like no one was making the connection. So the I used to actually be a fractional exec at a couple of communities that they're good businesses and they make a bunch of money and have a bunch of members and all these things and they were mainly domain specific.

So it's like hey I'm we sell here or we our CFOs here we're whatever there and I that's like not really that interesting because it's like it's a chase it's a race to like who can get the biggest name in this function. And so like cool, you're all talking about the same thing and that's plus or minus solved, you know, but then if you look at AI, like it's not solved and it's ever expanding and the post-training guys would actually really benefit from talking to the mid and pre-training guys and pre-training people like don't know a thing about coding and coding people don't know anything about human data and like they would all benefit from knowing each other because we're still I think groping for common language.

We can't like and again if we're internally calling I remember remember when agents first came out. Yeah. Some people were like agents like this is multihop function calling to use you know models or whatever a couple loops here. Yeah. And then other people are calling this other wild thing and it's like okay well like yeah it took us a long time just agree on this language and even when we did agree on the language like we try to flex like hey we all agree that rag is a thing. Well, is it contextual rag? Is it this ridiculous rag? And it's like, guys, like we get it, but let's go back to the common thing because what's actually going to help everyone win is can you simplify all of this technical jargon into business value. And that's what really people want to do. And that's why we'll have like a IML researcher over here and then head of digital transformation at a like a big bank. And they benefit each other from learning like, hey, we can't buy your model with 98% accuracy. We need 100 or you're gone or we'll relegate you to like a support chat use case or something.

And so yeah, that that's where I feel like it's the synapse is fire much more when there is a diversity of perspectives in the group.

Anna Kirk: How do you balance the diversity of perspectives that you might need to have the head of transformation at a bank in the same room as an ML researcher without that kind of hightouch community becoming diluted? Like how are you thinking about kind of that balances you Yeah. Deep in your network?

Albert Chun: Yeah, that's a great question. It's actually I think the A+ problem for almost any community is dilution, right? Like and so like we have a few like guardrails in place like we'll interview everyone, we'll require referrals and these things but that's a moving target. And then I think the other part is like the thing that balance is like people change over time, you know, like this person you thought was you passed by started a company and now they're the biggest thing, you know, and then the big exec, you know, had a fall from grace and did this thing and now they're what have you, right? And so it's like I can't pretend to say I solved it or will solve it, but yeah, that's yeah, just something that's always tough. Yeah.

Ryan Eppley: Maybe you could like you know tell our viewers a little bit more of like what does it mean for you to be creating these communities like walk us through your programming and how you make you know an interaction happen in the community.

Albert Chun: Sure. So I think the most important thing that people join a community is to meet people. And so like we have we have a goal like this person should meet three remarkable relevant AI circle members in the first week like when they join they should be invited to two events from the start and the first meeting should be their onboarding cohort where they get to meet a bunch of the people but in addition to that they should also have bespoke introductions as an example like so that those are like some of the programming I guess and then even like interview like applying online interviewing like it's like we're not like a hey you sign up for our email list and you're a community member like no like yeah we've said no to I've said no to over a thousand people who've applied saying hey we'll pay you a monthly subscription to join because yeah that the per like I like the northstar needs to be high quality membership and not some number and so that's part of it

and then the other is I think our like one thing I actually really appreciated I learned this from a book I read through Invisible. They had like a robust library they always recommend. I was one of the nerds who actually read them. And in this one one of their books they talked about when a central office should relinquish control. And they were talking about this like marketing agent or like some newspaper or something like, "Hey, big New York City guy. I can't give you like creative guidance on what someone in Ohio would want to listen to." And so, like with that in mind, we have these chapters and cities. So, like we're New York City, Seattle, London, New York, SF, Stanford, MIT, etc.

And, it's really like I don't know. I was I was in SF for like 30 years. I was born there, hometown, whatever, but I haven't been there in the last decade, right? I know my age, but like I don't know what's cool there. I don't know what's cool there anymore. So, I like we need like like we have our SF chairs like someone who's building robots at Applied Intuition or Architect at NVIDIA or you know our our AWS startups like they're they know the scene. They build the curated events. Like an example is like if I do a happy hour in New York City or a tech panel, people just show up. Remarkable people, random people.

Ryan Eppley: Amazing. I love it. New York City events are incredible. It's what we love about New York.

Albert Chun: Oh, I love it. In SF, they ask you, "Oh, so who's going to be there?" Yeah. And then they don't show up because of Bard or traffic or parking or whatever the reason, and it's like their attrition rate is low. So, you have to do unique things. So, we're going to have like a tennis social in the mission. He's like, "All right, do what the cool kids do." I would have never thought of that. Karina, our our SF chair, was like, "We're going to do this. This is how you engage, you know, you know, young scrappy founders." And I'm like, "All right, great. I'll find the sponsor and we'll host the event."

And so, yeah, that that's like a sample of our programming. And then we've done like the summit together last year. Yeah.

Rob May: Which is awesome. We did yeah we my my side venture fund we did a deal we did a deal with AI circle we put a conference together 150 people

Albert Chun: 150 people first yeah it was you know what's crazy about that was probably my third event first event was a happy hour second was a thing at Adam's house which is amazing and it was this summit I was like man anyway Rob carried us on his back was like guys we're going to do this that was a fun event after year

Anna Kirk: Is there going to be another one coming up?

Albert Chun: I hope so. Yeah, we're

Rob May: Well, we should you should do it with AI in NYC now, right? Cuz we have a bigger following than than I had on my own email list. So, yeah, we can do it.

Ryan Eppley: That was actually the day the day of that summit was the day we first met in New York. Yeah. Yeah, that's right. Yeah, we're talking about that he go, but I will be here for a year, too. Yeah. When when are you thinking about doing it this year? Have you even post some ideas?

Albert Chun: Well, so this year we're focusing a lot on we didn't do any conferences last year. This year we're going to be at all of the most of the major ones like GTC, ICLR, AI race summit. And then we want to do more regular programming. So like like Ken, our Seattle chair has like a monthly AI circle happy hour. We just launched a monthly research series with Hayes Labs out of their office to talk through research. Yeah, and so we want to do like hyper local, hyper specific, regular things people can count on and then we want to do these we'll do like sidec car events at these major conferences. And then we actually started doing this bianual like international offsite. We partnered with a couple of partners in AMIA and so we'll meet with AMIA execs there to and we'll bring the AI angle and we'll hear their exec like use case. And then I think all of this is kind of teeing up for something big in 2027 and like Q1 or Q2.

Rob May: Well, this is the I think the reason our thing went well and why we should do more stuff together. So two things that you and I think share. Number one, your your focus on quality of the event, right? Like people should learn from it. Like what I hate about conferences is they're mostly panels that start with a moderator and two people, then grow to three people, then grow to four people because somebody felt left out and you have to jam. So and then and then the moderator says every let's just give like 30 seconds to introduce yourself and everybody takes five minutes and you eat up 20 minutes of the 30-minute panel on introductions. It's like it's they're so poorly. Yeah. And the second thing is you like we've turned away a thousand people and I am people hate this but like I when I do these things I turn away people and I say no and people like you can't say and it's like well sorry we're like we're going to keep the quality high like I it's worth it if if you piss some people off it's okay.

Ryan Eppley: There's a reason people are still waiting in line to go to speak easy because it has four tables and it feels cool to be in there with with the right people you know so no I think that's great. I think you know we have some cool news to cover. I know that you have the couple teed up. I would love to hear some some input. I don't know if you saw some of the headlines here, but there's there's at least a couple that we need to talk on.

Rob May: Sure. Yeah. Well, I mean, we have to talk about open claw because that has become the thing. And now now to be closed claw with buying it. Yeah. They're Well, are they are they buying it? They're they're they're hiring the guy. They're hiring the guy. Yeah. I mean, I'm sure he's got a seven figure pay package and like whatever to do this kind of thing, but yeah, I mean, I will go last and give him my pens claw, but yeah, thoughts on open thoughts on open claw and on this deal.

Ryan Eppley: Well, it's supposedly supposed to stay open source. To what extent, I don't know. I just don't think it's as revolutionary as people say. There's been stuff like autog and related things out there for a while.

Anna Kirk: Cultural thing. It's a cultural I don't think it's a technical thing as much, you know. I think it's a cultural opportunity where people could get on, you know, boarded to this experience and they were signing up and it was cool to do it, you know. Have you talked about this in your community at all? Like what was the feedback?

Albert Chun: You know, I think the biggest thing that came up was the lack of security with OpenClaw. And then I mean to your point as to why it wasn't like technically revolutionary. It was like a cultural you know like zeitgeist type thing was like one of our members he's the founder of Zo computer and he basically did that but it was secured and I was like guys like every time someone talked about I was like just go try out Ben's you know Zo computer. And Rob, great speaker. You should really meet Ben if you haven't met him already. Like he's wonderful. And like I just thought, yeah, the virality of it was like good for him, all the things. But like, yeah, I had a little bit of chip on I don't I have no I don't I don't win if Ben wins, you know what I mean? But I was just a little bit of I was like, man, like he's been building this thing and he's done this, but like he did he couldn't catch the cultural moment. But anyway, good for him. Happy and all the things, but yeah, like to your point, like I don't know if it's like like a defensible piece of technology, a thing only he could do and all the things.

Rob May: Well, this is the biggest thing that I think people the biggest mental mistake people make when thinking about tech is historically tech has been about the technology, right? And the best for a long time in the 80s and 90s like the best technology typically won then kind of like good technology with the best sales and marketing won or whatever. Tech is like the fashion industry now.

And so what wins is a little bit random. Lots of tech is good. It's do you have a tech influencer who can support your thing or can you catch a viral wave? And so I as an inst So I've invested I don't know if the audience knows this but I've invested in since 2015. In 11 years I've invested in 135 AI companies. So I've seen a lot of stuff and I have not seen the best thing win most of the time. And some many times I was not invested in the best thing to be clear. As sometimes the companies I was invested in won and they shouldn't have and it was like it was random or it was ever but I just think it's run like the fashion industry now and it's who says what at the right time and people decide to write about it and catches some virality you know it's like yeah you know does does the I don't even know who the celebrity is cuz I'm so bad on celebrities but does does the hot celebrity wear you know does Beyonce wear the thing and now everybody's going to wear the thing that's what it's about right did do the AI influence inuencers to the ador mascot. Maybe maybe we can make that happen, you know.

Anna Kirk: I would say what I've seen and even some of my friends because you know I have a bunch of friends that are not in AI at all, right? Like this is like still them like dipping their toes in for the first couple like you know months and so and they'll text me like hey what do you think? In truth, like I think it's just very exciting that people are finding enjoyment exploring the technology, even if it's for dumb use cases. Sometimes to your point though, the first thing I always say is you really should not give it permission to wherever you're storing personal files. You got to really, you know, obscate the hardware away and escape the memory bank away. Because I feel like a lot of people don't realize that and even if they do, they're like, "Oh, well, you know, there's nothing too private on there." It's like you'd be surprised what what you're sharing if you're not explicit with like how you differentiate the hardware that you're running it on.

Rob May: Great. Well, and just AI these days like everybody that's doing AI coding is like yolo coding. Like they just don't like it's just like let's roll the dice, baby. Like who cares?

Anna Kirk: I mean it's crazy. I think if people understand the security implications behind it, then I am fully supportive. But often times, especially if you're targeting the end, the personal user, they do not understand the security implications of like MCP tool calls or all of these kind of like internal Yeah. Like autonomous execution that's happening under the hood.

Ryan Eppley: Yeah. So on on one end of the spectrum it's like you know last week we've seen okay open claw you know now they're not called clawed because all the implications with that and now they're bought by OpenAI and it's like more fun and it's more like community driven in so many ways. On the totally opposite side we have the Pentagon being very upset with Anthropic. Right. Okay, so we're the pendulum has swung pretty quickly in the last week of how fun AI can be and how serious AI can be and what the repercussions are for some of these research labs, you know? So, I'm sure you saw the headline about the Pentagon having some words with Anthropic. Have you talked about that at all?

Albert Chun: You know, actually, could you So, the snippet I saw was that, you know, oh, they use this on the raid in Venezuela and blah blah blah, but I didn't hear much beyond that.

Rob May: Yeah. So, so you're running into a bunch of problems here, right, with terms of service and use cases around and we around AI, right? And we've gone through like a really interesting trend where like defense built a lot of the tech industry and a lot of Silicon Valley and then we went through sort of through the like Obama years a lot of like tech like Google and Microsoft and those places like they didn't want anything to do with the defense industry. Which I which I which I also think was a little bit naive, right? Because you're like, hey, if I mean, some of this stuff is going to happen. You'd like the United States, you'd like your home country to be on top of it.

And then we went back to like tech's all in on defense with like Palantir and everything else. And now they're like, "Yes, let's use the tech to surveil all the United States citizens and help ICE and like all things." So they're now starting to move back the other way again. And so but I think I think tech needs to well first of all I think Anthropic should be entirely allowed if you make a product you should be allowed to say like here are the use cases and if people use it even if it's the federal government should uses it inappropriately you should be able to kick them off. Like that's actually the kind of American values of like what makes this company great, right? Is like I can kick the president off of my tool because he violated the things. But it does raise an interesting thing of like AI is going to be in military use cases. It's going to decide on targets and actions and all that and like where should that come from and how should we deal with that as a government, as a society, as technologists?

Ryan Eppley: It's pretty game theoretical. It's like if we don't will the other side of course somebody will and as you're increasing access to it for the masses then yeah somebody will. Hasn't taken this up yet. I think it's going to definitely pop up. It only happened in the last couple days that the headlines kind of came out. I mean does this does this happen in AI circle? Do do people talk about the ethical implications and some of the yeah you know difficult decisions around AI implementations.

Albert Chun: So I will say we end up talking a little bit more on the technical side. Yeah. Yeah and maybe it's more a reflection of me and the way I think through things like I had me earlier I mentioned I worked at a think tank and this was like my world for a while and then I've moved very consciously away from that world and yeah but yeah we don't have as many conversations around it candidly but when I think when I hear what I what I here like if I were and now that I'm gathering the pieces yeah I would not want to relinquish so much control over to a company in my country candidly like if I were the government it's not doesn't have to be a single person or whatever like as an entity like I would not have that like there is way too much asymmetric power there and so on a whim you could say I mean in theory they could say oh yeah we'll have all of your enemies have access to this and you won't or because our politics swing this way. No technical person will open it to our country but in this country all of their LLM providers give them full reign or access like that is that's frightening candidly

Rob May: but it's also hard at least in the US it would also be hard for the government to go do their like cuz you could say spend a billion dollars a year build your own model lab but it's hard for the government even in a scenario like that to attract the best and the brightest people because nobody wants to go to work there.

Albert Chun: True.

Rob May: Partially because whoever's in power, you know, half the AI people don't agree with them and don't want to go work there. And the other thing is they typically it's hard for governments to pay the best because the salaries are public and then people go why is this person making $800,000 a year and you know people get mad.

Albert Chun: So yeah, like you know, it's interesting when I think of like maybe the like China or you know when you see China do like this is a maybe a bad analogy, but you know when you go to the Olympic Games and you see all Beijing and they had this incredible spectacle with all of their shows and all these things. I remember one commentary saying the only country that could outdo them in that performance is North Korea because that's the only other country that has that much more control over them. Right. Yeah. And so when you look at countries like that, you they get outputs because they can force the inputs. Oh, you know what? I'm gonna have 35% of our children now study AI because I can, right? And so they can force inputs, rely on the outputs, and so that's tried and true.

And America has always won because we're like, hey, you know what? We're just going to incentivize them and we're going to let people follow the incentive. If you want to make 800 grand a year, yeah, go study AI. Go to Stanford. Leave your country and come to ours and build it, right? And so, and enterprise, you won't say no because I'm going to pay you a billion dollars in ARR to use your tool. And so, like, it's worked out for us. I'm not really fearing that, you know, some LLM in US won't give them control, but like when you really like we'll find ways of doing it, but like it does present a scary scenario. I think the people who are probably more afraid are other ancillary countries. I shouldn't say ancillary, it's like there's not just two people, but you know, I like if I were other countries, I would be I would be frightened to think, oh wow, the latest and greatest technology reside in these two or three countries. Like I would triple down figure out a way to solve my own AI problem. Yeah.

Rob May: Yeah, I think the government the government could do a lot by investing in more model labs. You know, doing stuff like that particularly if you know, you could take some bets. I mean, most of the labs are still focused on LLMs, transformer based architectures. I mean, you could you could roll the dice on let's fund the five most interesting non-transformer-based things and let's give them a lot of money and push them forward and maybe we make a step change at some point. But yeah, I don't think yeah, I but again part of the problem you have is most politicians are lawyers or reality TV stars and so they don't they don't always know a lot about technology. There aren't a lot of technologists that go into politics because it would be annoying to most of us who are have engineering backgrounds. So it's tough to have them make good decisions.

Ryan Eppley: I think yeah the real winner is the other labs. I think OpenAI was probably looking at the headline like, "Okay, cool. That that looks like some big government contracts potentially coming our way."

Rob May: Well, they open AAI just did a big deal with Massachusetts, the state of Massachusetts. Which I saw because they, despite the fact that I left Massachusetts because I did so many, AI events there for the last 10 years, they invited me to this, you know, come to this big kickoff they had. So I went up there and saw that and it was the governor more Healey, Sam Altman called in. Yeah. So, it's so yeah, opening eyes getting into that business. That was kind of cool.

Ryan Eppley: Interesting. They now need to start forming political alliances.

Rob May: Well, yeah. I mean this at tech has been historically a little bit disconnected right from politics partially because it was west coast focused and DC's east coast and also because there was just this like libertarian live and let live we don't care what happens with the tech we just want to build it thing which I think was a little bit of a naive point of view and it came back to bite them you know in the social media days when people started clamping down and they realized like we don't know any of the politicians we don't have any influence here and that's when they really got into lobbying and all And now with AI, all that's going to be on steroids, right? I mean, it's just such an important technology. A lot of money, a lot of big decisions to be made in the next couple years. Yeah. Yeah.

Ryan Eppley: We're running up on time, but Albert, at the end of every episode, we ask a New York specific question. So this week's is, what is your favorite New York cafe to go work at? Who wants to go first?

Albert Chun: Awesome. Yeah. The Hungry Llama in the West Village. Not only is it like just objectively great coffee, cookie, brunch, whatever. It's Yeah, it's Yeah, I would actually go there just for the food. It's that there are these communal tables and I actually met multiple AI community founders there. Actually helped one launch and just, you know, gave them playbooks and things and met like actually our New York City chair there. She's works with frontier AI startups at AWS and yeah, it was just I met like TPMs and all these big tech people and it was just like I just because I like striking up conversations with strangers and yeah.

Anna Kirk: What are the cross streets of that one? I feel like I've haunted every cafe in the West Village except for that one.

Albert Chun: Like there's not much around it honestly.

Rob May: I don't know where it's lama. I'm I'm looking up say it's Washington and Charles Street. Yeah, you're way over by the you're way over by the river.

Albert Chun: Everyone is wonderful there. Like honestly, my goal is to have a an event at that cafe at some point. It's just it's great.

Rob May: If they listen to the podcast. Yeah. What do you got? Where where your work in the cafe?

Ryan Eppley: You know, it's funny. I don't go there as much as I should, but my favorite place to go actually I'm gonna I'm gonna change what I told you guys before. Though. I'm going to say the Starbucks Reserve in the Meatpacking District.

Rob May: Oh, that's a good one. You know that that big one down there? They got a lot of they got a lot of space.

Ryan Eppley: And when you go in, they've got also you go downstairs and they have space and so it's like it's really good. That's a good one.

Anna Kirk: Yeah, that is a that's actually a really good one. I'll be honest with you. When you said Starbucks, I was like, "Oh, you're going with a hot day." Yeah. Okay. That's Yeah.

Rob May: Now, the Starbucks here in general in this in Manhattan are hit and miss. Like some of them don't even have seating or they have like two tiny tables and you're like, you can't really schedule a meeting there unless it's a big one.

Ryan Eppley: Yeah. Yeah. I don't know. Maybe it's just because I spent a lot of time there over the last year. You know, our company just recently got our first like headquarters and our first private office and things like that. So, all of 2025, I was in a lot of cafes. Paper Coffee, like 28th and like 6th. Is that where you and I met? Where was the place? We sat down and had a meal there for a bit, right? And they also have a terrace outside that no one goes to. It's like this beautiful spot with a couple outside pieces of furniture. Nice trees and it's like even quiet enough that you can take a call outside in New York City there and no one's ever there. Maybe some people will now see it and they'll they'll be showing up, but that place has closed deals for me. I hire someone at that cafe. You know, there's been there's been some some times had in that cafe. The cafe has done all of it.

Rob May: You've had that's actually the third co-founder.

Anna Kirk: I would have to say so I love this place that just opened up called Urban Backyard. Specifically the lo the location on Sixth Ave in spring. You just you walk in and there are like plants everywhere and it's just I'm very motivated by environment and that's one of those places where you walk in and it is impossible not to be happy. So one of my favorites you got to go. They have a great ube latte.

Rob May: Okay. I like that. Yeah. The only problem with cafes for me is like I only drink like black drip coffee. So they'll have like this long menu. I'll be like can I get my black drip or And don't you get frustrated like standing in line behind all the people and they're complicated drinks. You're like I just need 10 seconds right there.

Albert Chun: That canister. I used to be a black coffee person. I don't know if it's like what it is but I've mellowed out and started getting like lattes and cappuccinos. And there's this drink I've never heard of it until recently. It's called an Ein spawner.

Rob May: We have no idea what that is. What's going on here?

Albert Chun: So, it's I think it's like an Americana with whipped cream instead of milk. And it is Anyway, that's my my guilty pleasure.

Rob May: Okay. Yeah, I'm too old to tolerate the calories and drink. I have like if I can just get black coffee, which has almost no calories, I that's what I need. So, nice. Albert, if well, thanks for coming on today. For people that want to apply to AI circle, realizing you have a two-thirds chance of being rejected, but like if if they want to do it, what's what's the process? Where should they go?

Albert Chun: It's just going to a-circle.org.

Rob May: Gotcha. Cool. Well, those of you that are watching, go check out AI Circle. Please rate us if you like the show. Rate us. If you don't don't don't do anything. Yeah. If you'd like to be on the show or if you have a guest suggestion, please reach out. You can do that through our website. We have some cool stuff coming on. We're we're going to launch New York GPT, which is going to be a chatbot for the city. Have a lot of information. We're going to start doing some more events now that we're we're we're getting this thing to scale. So yeah, check us out, watch the other episodes, and thank you guys for watching. The next