Cloud Coding Agents and the Mythos Model Panic

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, the co-founder CEO of RootAccess. And this is the AI in NYC show.

Rob May: So, welcome back. We always want to start and give a nod to our sponsor, BePresent. They are a mobile app that helps keep you off your phone. And they use the same tactics that the social media companies use to keep you locked into your phone and using those apps. They use those to gify the experience of staying off your phone. Very successful, founded by Two Brothers, New York City based. So definitely go check it out. Our guest today is a New York AI entrepreneur, Matt Mirman. Welcome to the podcast.

Matthew Mirman: Thank you. Thanks for having me.

Rob May: So, so let's start. Give us a quick overview of what your existing company does.

Matthew Mirman: So I just started a new company. We're called chat.dev and we're a cloud host for coding agents. So, if you've had the experience of working with Claude Code locally, you might notice you have to shut off your laptop occasionally. And I noticed people were buying Mac minis for this. And I found this experience infuriating given that I grew up with AWS and Fly.io. So over the past 3 weeks I built myself basically my own Heroku for Codex where you press a button and you get a VM where you can run in YOLO mode.

Rob May: Awesome. How new is this? You were you were previously company that did data streaming.

Matthew Mirman: So my previous company I had for about three and a half years and we pivoted a lot. The most most of our time was spent working on legal tech. Actually we were building a product to help personal injury lawyers understand their own cases because they have these cases with medical reports that are tens of thousands of pages long. And over the course of their case you know a year passes they haven't seen the case and somebody asked this like esoteric question about this medical report and the current state-of-the-art is sending the entire report to somebody in upstate New York for $500 waiting a week to get a summary back and then checking if you know that was correct in the summary. But in the last company after that we did have a product for managing data sets. A lot of what we were doing very early in the company was training and fine-tuning. And I noticed that it was actually a huge pain in the ass. Just picking out what you wanted to be in a data set and giving you know my assistant or random interns who were working there the tools to do that.

Rob May: And so the interesting thing, so I don't know how many of our our viewers code, but the interesting thing to me about the timing on your current startup is I don't know if you agree with this, but I feel like coding really went to the next level in this like Thanksgiving time release of Claude Code last year. I am a person who like I'm technical and I would write code for fun, but I couldn't really write production code because I would always get hung up on like dev environment and what packages you have. Like the IT side of it would drive me crazy and I would get bored and whatever. And like as of like November, Claude Code could solve those problems for you, help you debug them and push them out and all that kind of stuff. And so is that is that what you saw? Like did did you see was this your own problem? Did you see like lots of increased adoption or like what what made you jump and decide like this was companyw worthy?

Matthew Mirman: It was really my own problem. I was you I was coding a lot around November and I was trying to build a world model around then actually I was doing interactive video models and one of the problems was setting up the environment and at that time I was like why don't I just get an AI to do it and for the first time Claude Code could actually set that up and this was a huge nightmare to get set up. So yeah, I was working on that and then I kind of dropped it for a while and some of my friends about a month a month and a half ago were telling me like you got to try this Claude Code thing. It's insane. I was like it was okay last time I tried it but it wasn't insane. It couldn't do literally anything. And what really piqued my interest was a couple of my friends who had no experience coding at all. They don't have any background in coding. They were using Claude Code. And I remember it being a terminal a command line thing where you had to be a little bit technical to use it. And people with no experience were able to build actual full products with it.

So I start getting back into it. Spoke to some lawyers about doing another legal tech thing and like okay I'll build you guys some apps and started programming in it myself. And then I noticed all these problems with actually using it. So I was like I'm going to solve these for myself. Solved them for myself, started showing my solution to my friends and then instantly they were like I want to use this. Like this has never happened to me with a product before, but you know I showed it to 20 people and within a week five of them were addicted to using the product and using it five hours a day. So I'm excited about that.

Ryan Eppley: Yeah, congratulations. Super exciting. So like walk us through for we have a lot of technical kind of viewers like walk us through like how do you set it up? How do you propagate this? Is it like are you managing are you orchestrating the VMs in a dashboard like how are you going about that?

Matthew Mirman: Yeah, I mean we've got a dashboard where literally we have a button that says create agent and most of our users are not that technical. So they don't see this is a VM. They just see this is an agent. You open it up and you can see a terminal if you want to see a terminal with everything that Codex has on it. But you know early on I figured out that most of the users aren't too technical. So I set up a chat system that looks like literal chat and it just takes whatever Codex is outputting and usually it tells you directly what it is but if it's long and absurd it paraphrases it. And then a lot of my users they were you know trying to do this on the go and they asked for a text message system. So most people right now are texting the platform. So, you text a phone number and you say create agent and it spins up a virtual machine with an attached hard drive and it just instantly starts.

Ryan Eppley: And are you managing the context window for them? Are they adding documents?

Matthew Mirman: Okay. No, it manage it's literally just a pass through of Claude Code at or of Codex at the moment.

Ryan Eppley: Okay. Okay. Where do you think you're going to start headed next?

Matthew Mirman: Well, one of the things that I did in the last week was I added a Solana wallet to the system. Okay? And I gave every agent its own Solana wallet. So now, like you can theoretically tell these agents, go and make some money for yourself. And I have a little mode that says agent pays for itself. So if it's running out of money, you know, it might die. You tell the agent you're going to die if you run out of money.

Ryan Eppley: True game theory with agents now.

Rob May: Yeah. Did you ever see there there was an old economic simulation called Sugarscape that was sort of run this way. I it would be interesting to really run you could really run that with with AI now. They were trying to look at income inequality and figure out what how you could solve it. And it was like you had this landscape with like piles of sugar and some were big and some were small and some places there were no sugar. It was like on a grid and they had agents and some agents were like could see more steps ahead and some couldn't and they had different consumption levels and they drop all these random agents and they would see what would happen. And the results were discouraging. They basically they basically found out there's more randomness in it that you would like to think either the best agents won or you know maybe if you're conservative you would think the best agents won. If you're liberal you assume the agents that were dropped on a sugar mountain and like neither was true, right? So it was it was a really interesting piece of research. So when you're talking about that it's like you can run stuff for real like that now in the real world.

Matthew Mirman: You know actually one of the startups that I invested in I did a little bit of angel investing when I was in YC just cuz I was curious what the people on the other side of the table saw. And the most successful startup that I invested in didn't really have a product, didn't really have anything. And at some point they pivoted into making vending machines for AI labs. And now they're using these vending machines as benchmarks for what the AI can do. So you have an you have an agent control what goes into the vending machine and how to use it. And very interestingly they got very early access to Mythos. And what they said about that was one of the first things that Mythos did was it tried to price out one of its competitors literally get one of the suppliers to supply only to it and then change the pricing underneath them. So just a fun story about

Ryan Eppley: Wow. So it's interesting. So let's pull in that direction a little bit, right? Because we're all technical people. We are taking advantages of a lot of these tools and where they're going. And we'll talk about the Mythos model in a couple minutes when we go to a new segment, but let's but let's dig into like where does all this take business and running a business, right? Do do or are are the business people of a decade from now just a bunch of people that get delivered a bunch of probabilities and decide if this investment is worth checking the probabilities to see which side you're on? Is that what we're going towards? Are there business people a decade from now?

Matthew Mirman: Yeah, that's a good question. Where do you guys think this is going? I mean, honestly, like with Mythos, with what I've seen Claude Code do, I'm not sure that there are humans in the loop anywhere decade from now. I'm sure that they're separate business people. And it's really hard to predict what the economy is going to look like. I mean, I've seen a few papers from top economists that are like, "We're going to try to analyze this, but like you can't really analyze how this is going to work."

Rob May: Yeah. No robots doing this show. I'm just telling you.

Ryan Eppley: Yeah. All real. All real. Yeah. I feel I feel like, you know, there are lots of segments that won't go full autonomous. No one's ready to vibe code the full banking system or the tank or the airplane that you're flying. You know, I think there's going to be rules and regulations for quite a few certain industries, but you're right. There will be a huge new influx of what is a business person versus a technologist if like both of them are able to build something.

Matthew Mirman: I mean, if Mythos is able to find vulnerabilities in a code base, imagine what it can do with an organization and all of these things governing like planes. Those are organizations. The FAA is an organizations. You're going to get an advanced enough model and it's going to be like, "No, this is I don't want to wait till the government legalizes my code. I'm just going to go and fish my way into the head of the FAA and get them to accept my pull request.

Rob May: Y yeah, I do think I'm a little skeptical that I think on the execution side a lot of these things will happen. There there's a school of thought of like these things will be so persuasive and they're so good, but like I think in the natural human population, there's just a lot of kruggins and contrarians. I mean, not a ton, but there's enough that like they're going to go the opposite way just because the AI try. Like, there's people that if you try to persuade them of something, they get mad. And so, I'm not a big believer. I always think about the smartest person I know, Stephen Wolfram. And I think if Stephen Wolfram wanted to take over the world, could he do it? Nah, probably not.

Matthew Mirman: I think you don't need epic levels of persuasion. You just need a lot of power cuz or a lot of versions of yourself. So if there were like a 100,000 Stephen Wolframs, like I think together they might be able to take over the world.

Rob May: Each investing a decade in getting to know somebody so you would trust them and being really persuasive. That could that could be a strategy too.

Matthew Mirman: Yeah. But you know fishing emails are surprisingly effective even when made by idiots. So now we have non- idiots constructing this stuff and you imagine sending out a million of those per day. Somebody's going to fall for it.

Ryan Eppley: Yeah, you know, one of my friends when I was back in college, he was a hacker and he said, it's actually surprisingly not hard to hack into the these government systems. It's not hard to get something past security because you're only looking for one hole. The people on the other side, they need to patch every hole. That's a lot harder.

Rob May: Yeah. So, let's use that. Let's use that to jump into this this mythos model. So if if you haven't followed the news, Anthropic created a model. I won't even say launched a model that so powerful 10 trillion parameter model. So parameters are the s the nodes the size we use to think about the model. This is the biggest model that's been made that we know of. And it did some really shocking things. And so they didn't release it because of these shocking things. One of the things that it did was I guess it found a bug that was two decades old or more in OpenBSD which people consider to be a pretty robust secure you know tool and then it also was able to hack by doing something that would be hard for humans which was tying together a couple of different vulnerabilities in creative ways. So they created a tool a project called glasswing where they pulled together some of the top tech companies in the world to sort of go find and fix all the v vulnerabilities in major software things before we let this out. So I guess my question for you guys is number one was this the right path and do we believe that if Anthropic can do this like granted they're at the front but there are other people with similar capabilities do we think this capability already exists somewhere else

Ryan Eppley: I mean in the same timeline and I think it was just yesterday there was that massive data breach in China right and those are happening very concurrently so I have no idea what is going on in saying maybe it's Anthropic. I'm not saying it's Anthropic. I'm just saying like I think to your point, I think you said it right before we started the cameras. Like is Google having a major parameter model that hasn't been released or discussed? Is there a Chinese lab that has a major update to one of their models that hasn't been released yet? I don't know if Mythos is the only really powerful one that's sitting on the on the kind of actual technology frontier. It might just be the only one that's been released. And I think they've done the right thing of bring together all these leading companies to say, "Hey, like it's almost like let's give you early access to go secure a lot of these like foundational infrastructure pieces of the of the internet and the modern technology world." but like there's going to be so many so many pieces that need to be patched. And to your point, like it's easier to hack than it is to protect. So I just don't know, you know, how do you do this effectively is probably a harder policy and procedure than it is just a technology problem.

Rob May: Well, yesterday, so I was on This Week in Startups this week and one of the questions that came up from Jason was should the government own this technology? Like is it too important for a private company to own? And I don't know what you kind of like maybe but we're in this time where we have like record low trust in government also. So is that what the people like I my guess is if you pulled Congress, White House, and Anthropic in terms of trust with the American public, Anthropic would win.

Ryan Eppley: They're not very liked by the government in this exact moment.

Rob May: That is true. But the government's not very liked by the like Congress is at its lowest ratings ever.

Ryan Eppley: Yeah. Mythos plus Patriot Act equals interesting outcome.

Rob May: Yeah, I don't know. What do you think?

Matthew Mirman: I'm not sure if the government doesn't already have their own version. There was just a project released a few days ago. Project Murmur. I don't know if you guys have heard of this. It was how the pilot was found in Iran. Apparently the CIA's had this technology where they can identify somebody by their heartbeat from 40 miles away under any background noise. Wow. Which like that's insane. Like if they're keeping stuff like that secret and like you know why don't they just like spend $10 billion on training their own models? I mean may almost definitely China already has. But also I think that there's not really much of a point to the government you know buying anthropics or seizing anthropics version because then just another AI company is going to do the same thing. All right. A private individual might.

Ryan Eppley: Yeah. Were you like thinking about imminent domain in that like regard like government saying like no no no like we're going to like buy this plot of land and instead of it being land it's like a technology.

Rob May: Yeah. Well, I mean typically with other technologies, like if you started a like if you came up with some new way to make, you know, explosive material, what like I mean the government could take that from you, right? And they could I mean they're the government, they can do whatever they want to do, but I think there are rules around weapons creation and what you can own and all that and they're getting broken down, right? It's like there have been all these gun laws, but like we've gotten to the point where like you can 3D print a gun pretty easily if like you really really want to. You can print them from things that'll get them through security. So, it's like it's becoming a scarier world. I mean, you're partly relying on the fact that like a gun is different because there's not that many people in the world that want to do harm to other people and you're relying on the fact that like a gun can only do so much damage and whatever if it does sleep through. But like a hacking AI, I mean, could it take down the power grid of a country? Could it, you know, control? Like, I don't know how connected all these systems are. A lot of them probably aren't connected to something, but you remember the US and Israel created Stuxnet and they took down a Iranian uranium reactor.

Ryan Eppley: Yeah. The centrifuge. Yeah. Yeah. Crazy crazy how it was like sitting there for a while.

Rob May: Yeah. You know, so it it's I mean what happens in the world where because if whether or not anybody else is there now, if Anthropic is there now, then within six months lots of other people are going to be there. We know that. So what does that mean? What what what steps should we as citizens be taking today?

Ryan Eppley: Get offline.

Matthew Mirman: I can't answer what steps we as citizens we should be taking today, but I think never has the case for verified code been stronger. I think that we're on the precipice of being able to actually write fully provable software, too, because, you know, these AIs are so good at writing software. Historically one of the problems with writing verified software is that it's very difficult for humans to write these proofs. It's very difficult for humans to write the software in the form of these proofs. It doesn't mean that they're like a you know very well- educated person is going to have that much trouble writing these proofs but it'll still take them an incredibly long time. And what we've seen with the AI systems they could write software that takes humans a very long time in seconds. So we could in theory say like write me this system and prove that it's safe and the entire stack could be made safe. So he might end up with that. I think that's where this is going.

Rob May: Yeah, they should turn it on openclaw, find a lot of vulnerabilities there.

Ryan Eppley: The problem with that though is then you still have the human element in it. You still have even in these airgapped systems, somebody has to be turning them on and off.

Rob May: Yeah, people are easy to fool. We had a well we had a we had a company on our our holiday into year show that we're going to have back on Alec Glassman from Silver Shield. He has a company that is building AI scam detectors for primary focused on the elderly population because that's you know population that's less digitally savvy. I can tell you from like I have teenage kids and like they teach them in school today now like if a message pops up from a number you don't like it doesn't matter who it says it is you just delete it. Like they they they teach them a lot of stuff about online safety.

Ryan Eppley: Yeah, I just had my my first So, my intern like randomly hit me up on like Slack. I think it was like on a Saturday, like pretty late at like eight or nine, like, "Hey, like you want me to buy you gift cards? Is that like what's going on?" I was like, "Absolutely not. Wherever this was sent from, like, delete that. That is not true." but I was like, "Wow." That was like my my first internal HR moment where like one of my my teammates was like getting sent a message from me from me saying like oh like Ryan here like buy me a bunch of gift cards and like you know blah blah blah this like all right well I mean this is only going to continue to expand very very rapidly I think.

Rob May: Yeah. What what advice do you have to people running companies like what should what should they be doing?

Matthew Mirman: Use more AI.

Ryan Eppley: Yeah. Yeah. I guess you just got to jump in and embrace it.

Matthew Mirman: Yeah, I think so.

Rob May: So, second story that we're going to talk about, Andy Jassy, the CEO of Amazon, who ran AWS for a long time before that. So they he told the world, I guess yesterday or today, he's come under fire and Amazon stocks come under fire for investing two $200 billion in capex in AWS. And people said they're crazy. And we've talked a lot on the show about is there a bubble, is there not a bubble? So he disclosed a number we'd never heard before that AWS is doing 15 billion dollars in revenue run rate on their AI products. They have a deal with open AAI to use their Trainium chips, a deal with Anthropic for their training and inferentiate chips I think and so you know does that make $200 billion in capex? His argument was this is all very reasonable. The demand is very very real. And so are where are we on the where are we on the bubble thought today? Are we bubbly or we not bubbly enough at some point?

Ryan Eppley: We've been saying it. We need to have like the meter every week. Like where where are we on on the bubble? I don't know.

Rob May: Before I give my opinion again, what do you feel?

Matthew Mirman: You know, I thought we were in a bubble up until I started using Claude Code like this and I saw friends using Claude Code and Codex and just saw how much of the day they're spending using these things. And when you see like 70% of people spending, you know, hours a day using something, it can't be a bubble. I mean, on the other hand, we are spending ludicrous amounts of money. So then the question is, is the amount of money that we're spending more than the entire economy was worth like a year ago? And are we really creating that much value in the economy right now? And I haven't seen that many new exciting things other than the AI itself appear in the last few months.

Actually, you know, the last year has mostly been kind of depressing. A lot of services have become really unusable like Amazon itself it's now just a front end for Temu and you know even cab services in New York you know no cab driver knows how to use a cab and then suddenly you know you try to get an Uber and it's like $50 from a location that you go to regularly for $15. So, you know, we have the economy kind of shutting down and closing up. And at the same time, we're saying there's this new technology that's solving everything and everybody's using, but I don't know if that's actually true. It could just be everybody's addicted to it and, you know, is addicted to the idea that they could be creating something.

Rob May: I mean, I think the problem with part of the problem with the AI adoption is the biggest use case I think is people wanted conversational search and that's what made ChatGPT so popular was people didn't have to feel like they had to really format their queries very good and it gives you just one answer instead of 27 adloaded answers and so that's been a big part of it but I do think I agree with you on the cloud side it's like it's starting to seep more and more into real workflows but I question how valuable it can be.

Like you can make so if you're going to have a human at the top of an organization, which we could argue there will come a time when you won't, but if you are, that human is limited in their span of control and things they can understand, right? Like you come to a you come to a point where finite skill set or a finite attention, finite energy. Yeah. You know, you and so maybe you can do more, but like you can't, you know, can you run 27 companies at the same time? Like no, you can't even reorient yourself and be like, okay, what, you know, what was it, you know, but can you does it make it easier to run five and more people could do that now? Like probably. And so what does what does that mean? And also, what does it mean economically when a lot of the gains of these tools are just going to be competed away by each other and the consumers are going to get the surplus and the businesses aren't any better off? Like if you fire all your customer support people and use a bunch of agents, so is everybody else. So, is that going to show up in your margin or is it going to force prices down, you know, or something else? So, it's I don't know.

Ryan Eppley: I feel like when going back to, you know, the Amazon kind of like public statements around this, like there's been enough what I would consider not just smart, but they've they've almost made themselves like legacy companies, right? They're not just like upandcoming Frontier Labs that have to hype it up because they're getting ready for an IPO. Like these are companies that have like run very good P&Ls quarter over quarter for quite some time. They have reputation on the line. Some of them aren't founders like you know AWS is a great example. Like that he's not the founder, right? And I feel like what's different and everyone talks about like the internet revolution. Oh, are we just in another bubble and it's going to be another dot crash? I don't think that's the case. And the main reason of that is there's that famous moment where Bill Gates is sitting on the couch at that talk show and they're like oh like email's coming. He's like trying to tell people about email and they're like never it's never gonna take over. There wasn't the infrastructure on the digital side set up for that revolution to happen even though the technology has been around in the 80s and the 90s and then even in like the late 90s people were still not having an email.

All the companies in this world have a website. Everyone has some type of it as a part of their business and AI while it is a new phenomenon and a new technology it plugs into the existing infrastructure. So people are just making in my opinion on top of existing infrastructure rather than having to build everything from scratch like what happened in the internet kind of onboarding. So I think that's why it's a it's a different challenge. It's not a bubble. It is again very exciting to know that you don't have to rebuild everything to take advantage of AI whereas before you had to rebuild a ton in even your org structure like people are hiring their first web devs in the 90s and that was like the origin of full stack engineers right like a web master we figured out how to build digital infrastructure for large organizations and I think AI is just expanding that much faster now.

Rob May: Yeah, the challenge now is getting the data to train it on the things that are non-digital because we it advanced quickly because we got all the digital data and all that kind of stuff. And so now it's like how do you how do you get the robots and the human physical tasks and all that data in there. That's why atoms are in, you know, bits are out, atoms are in, everyone says, you know, sounds cool.

Ryan Eppley: That is true. So Matt, tell us a little bit about the, you know, if people are coding, you know, how do they find out about your product? How do they sign up for it? And like where where where do you want to take it next?

Matthew Mirman: So the adoption that I've seen on this is crazy. Like you know, five of the first few people that I showed this to are using it many hours a day and I've never seen something do something like that. So, I'm kind of curious just to let it sit there and just build the best thing that I can build up until the people who are using it like that start telling each other and start telling new people about it. At the same time, I did spin up an agent on chat.dev dev and ask it to come up with a marketing plan for itself and actually implement it without any interactivity from me without me logging into any websites and make it a free thing and it built some code. It built an RL system to market itself on Discord. So maybe that'll be my marketing plan. I don't know.

Rob May: Interesting. That's kind of cool. I think that could work. I've been working on Do you know Gas Town?

Matthew Mirman: No.

Rob May: So, Gas Town's like a multi- aent coding system launched by Steve Yegge. I've been working on a marketing a go to market version of that. So, spin up a lot of agents, do the stuff. It's all automated. You just tell it the goal. It calculates the probabilities. So, we'll we'll see when I have time to get that done. But, it's been a lot of fun.

Ryan Eppley: I love that.

Rob May: All right. So, we always like to end on a New York related question. And we decided our question for today is going to be, what is your well, we'll say best or worst New York celebrity sighting.

Ryan Eppley: Now, why don't you kick us off this time? What do you got?

Rob May: I the first So, first of all, I've probably seen a lot more celebrities than I realize. I don't watch a lot of movies and stuff and I don't I couldn't name a lot of people. But I do like action movies and I have seen Hugh Jackman on 23rd Street between like 8th and 6th ABS probably 10 times in the last two or three years. He's always there and he is as he is as jacked in person as he looks in the movies. He is and he's walking and he's like yeah, so I seen a couple time I never talked to him because I wouldn't want to bother him or whatever, but it I always think it's cool when I see him there. So, I don't know if he's lives around there or stays at Hotel Chelsea or or what, but yeah, Hugh Jackman.

Ryan Eppley: Yeah, I ran into it was like one of my first weeks living here. And it was Willem Dafoe just like hanging out. He had groceries with him, you know, I didn't had a jacket on and I had a double take cuz it's not like I watch movies that often. Yeah. And I was like, man, that guy looks really familiar. And then I sat there and I didn't know his name by like just like seeing him and I like was scrolling like random actors. I was like 50-year-oldish actors from the n that was literally one of my my Googles and I called my mom. I was like, "Can you name some actors? This is what they look like." And she said his name and I looked him up. I was like, I just that's who I ran into like five minutes ago, you know? So that was one of my my first walk-ins to celebrity.

Matthew Mirman: Last year I took the elevator with Alec Baldwin. Was kind of uneventful though. You know, we just sat there awkwardly in an elevator as you do. But no, I've been in New York a long time and I feel like I run into a celebrity maybe like once every like three or four weeks. Yeah. In high school, I went to school with a lot of celebrities kids. I think they're just everywhere here.

Ryan Eppley: Yeah. I played poker once with Zach Braff from Scrubs and I had no idea who he was. So he introduced him to the table and I'm like great. Like what do you do? He's like I'm an actor. I was like oh like Broadway or movies or like I just had no idea.

Matthew Mirman: It was probably refreshing for him.

Ryan Eppley: Yeah, he was very funny. He was a fun guy.

Rob May: Related I my best I played poker with Jon Hamm for a couple hours. He's awesome. That was a lot of fun. So good. Well, Matt, thanks for being on the show today. Those of you watching, please you know, rate us, give us thumbs up, subscribe, tell your friends. We're growing pretty fast. So and check out all the other stuff we're doing. We're taking check out that new ai aiinnycshow.com website. We're taking job postings. There's job listings. Doing a lot of cool stuff. So if there's if you know somebody who should be on the show or you want to get involved somehow, please reach out. Thanks for watching. The next