BUILD MODELS

“If a brain doesn't have good memory, well, it's only as good as the memory. And so what I'm using is Obsidian. And Obsidian is great for creating markdown files of a lot of the skills that you have, a lot of documentation. You work with these agents all the time. They forget what you're talking about and you need something like this that's going to remember. It's extremely high leverage because it's able to get the information you need faster.”

“Some of the mergers are done on this basis and as a consultant I can tell you well um are you sure? Is this really the same process? Are you really working the same um way? Let's have a deeper look. Let's go at least one level deeper. Let's go from the cloud level to the kite level. maybe even to the sea level and look at more details in the process.”

“DeepSeq R1 does some amount of this imitation learning after the reinforcement learning... if you just do reinforcement learning, by the way, sometimes this model starts code switching in the middle of solving math problems. It's just suddenly speaking in Chinese and English and back and forth, or some other foreign languages that may not make sense to human readers. So reinforcement learning only cares about whether you got the final solution right or not. It doesn't care about how you got there. So strange behaviors can be emergent and then it can be even reinforced.”

“When you, you are the most successful when you can offer two things, reasoning and specialization. So reasoning capabilities with our agentic platforms, a agentic frameworks in the platform, we are bringing that to the fore. Specialization is something that is very, very crucial. So this can be achieved primarily using, specialized models and fine tuning. Student teachers student distillation gives you that control you need to have on, providing personalized experiences as well as providing, having some control over your latency metrics.”

“I think the core is that, you know, you really need to treat agentic AI as a system. It's truly a system. You have to start with governed data. You have to kind of put in that risk controls baked into multiple layers of your, of your application or your system. You have to look at, latency as something that needs to be optimized end to end. And, understanding that your biggest gains do come from postproduction telemetry is also critical.”

“All these capabilities, all these SOPs are all documented for you. And you want to put these all into skills.md files. So they should all live within a skill.md. The better you are at one, doing your job, but two also documenting what you do. I highly recommend making a bunch of Loom videos or just dictating using whisper flow, just saying what you want to say and making them into skills.”

“DeepSeek is kind of losing its crown as the preeminent open model maker in China and the likes of, Z dot AI with their GLM models, MiniMax's models, Kimmy Moonshot, especially in the last few months, have shown more brightly. The new deep seek models are still very strong, but that's kind of a it could look back as a big narrative point where twenty twenty five deep seat came and then all and it kind of provided this platform for way more Chinese companies that are releasing these fantastic models.”

“The idea is that during pre-training, the model is forced to be completely passive in the way that it learns to predict which token comes next. But what if we encourage the model to think for itself? Before predicting the next token, what if we encourage the model to think for itself by generating something like a chain of a thought? And then predict the next token... the key idea of our approach is to make the reward information gain of predicting next token with thought compared to without thought.”

“Copywriting course used to be, hey, come to our thing, we'll teach you how to write good copy. Now you can ask ChatGPT or Claude to just write the email for you, write it like Matt McGarry or write it like Neville, and it'll figure it out. The need for that service has gone down and down and down, and we've been niched more into strategy rather than just execution.”

“There’s going to be more of a premium to style. Making content for SEO—like a 10,000-word post with all the little keywords—is no longer rewarded because that is a dime a dozen now. Having style and substance and weirdness and entertainment is going to be more valued because it's something AI still struggles to replicate with actual soul.”

“I think social media is blogging. You have a piece of content, a picture, a text update, and you post it on a network where people read it and comment. I went from making long-form blog posts to just social media content because I get more out of it now. I've stopped posting traditional blog posts because a blog is almost like a dumb social network where you just post and pray people see it.”

“I disagree with some of their presumptions and dynamics on how it would play out, but but I think they did a good they did good work in the scenario defining milestones. The camp that I've fallen to is that, like, AI is, like, so called jagged, which will be excellent at some things and really bad at some things. So I think that when they're close to this automated software engineer, what it will be good at is that traditional ML systems and front end, the model is excellent at. But the distributed ML, the models are actually really quite bad at.”

“All these capabilities, all these SOPs are all documented for you. And you want to put these all into skills.md files. So they should all live within a skill.md. The better you are at one, doing your job, but two also documenting what you do. I highly recommend making a bunch of Loom videos or just dictating using whisper flow, just saying what you want to say and making them into skills.”

“There must be a better way of it, fundamentally better way of doing this. And can we find it? In some ways, the nature found a solution, which is the human brain. The nature found the solution, and human brain requires so little energy. Our brain apparently use less energy than one light bulb.”

“This is truly compounding technology. A lot of the work that we do just compounds it. We don't throw it away. It gets better. The operating system work gets better. The dev tooling gets better. The models get better. And so we're really gonna get a I I think you see it in Waymo as an example. If you can put a little constraint on commercials that has a small ability for you to more likely see the other end of that walk.”

“The most often thing that people who are not building within the, you know, kind of cut just call it competitive business world or competitive technology world. It's like, we'll just slow everything down. Right? And you're like, okay. The problem is is competition doesn't slow down. That's actually not not the way that that kind of competition works. Like, if you just said, hey, everyone. Wait for me to get my economy in order, and then you can start shipping your export things. I mean, you know, take a look what what's gonna be happening within Europe with the spread of BYD cars from Hungary, for example. It's like, oh, no. No. Wait for a decade for us to get our auto industry in in in place. That's not gonna happen in terms of in terms of how this operates.”

“Anthropic lost in court and was owed $1,500,000,000 to authors. Anthropic, I think, bought thousands of books and scanned them and was cleared legally for that because they bought the books, and that is kind of going through the system. And then the other side, they also torrented some books. And I think this torrenting was the path where the court said that they were then culpable to pay this billions of dollars to authors, which is just, like, such a mind boggling lawsuit that kinda just came and went.”

“So we wouldn't want to go back to pre Industrial Revolution, but maybe we can figure out ahead of time by learning from it, what those dislocations were and maybe mitigate those earlier or more effectively this time. And we're probably going to have to, because the difference this time is that it's probably going to be 10 times bigger than Industrial Revolution, and it'll probably happen 10 times faster. So more like a decade, then unfold over a decade, then a century.”

“If you said the the minimum that Mythos is is an ability to have an infinite number of, quality cybersecurity engineers who are penetration testers, and it's not infinite obviously because there's a compute cost for running them, but you could take a thousand of them and run them in a direction, you end up you know, if it's just that, you end up with a rechanging of the cybersecurity landscape because we have billions of lines of code that essentially haven't been touched and partially because there hasn't been either an economic model for the number of cyber criminals or the number of people employed by rogue states to go after any other than a certain set of systems. But it broadens the range by the ability to just kinda spin up many new penetration testing engineers, you know, with a kind of an AI mechanism. And that's the minimum that I think you could look at, the the discussion around Methos and its cybersecurity issue is.”

“Collaborative modeling is this idea of we model together uh we the tech people the developers with them the domain experts the users so we bring together nontechnical people and technical people and build the domain model together that's why it's called the collaborative modeling and there are different methods eventtor storming domain storytelling and so on they are all lightweight sticky notes stick figures so there's a low entry barrier.”

“I was training the Gwent three base model with RLVR on math 500. The base model had an accuracy of about 15%. Just 50 steps, like in a few minutes with RLVR, the model went from 15% to 50% accuracy. And the model you can't tell me it's learning anything about fundamentally about math in 50 steps. So the knowledge is already there in the pre training. You're just unlocking it.”

“If a brain doesn't have good memory, well, it's only as good as the memory. And so what I'm using is Obsidian. And Obsidian is great for creating markdown files of a lot of the skills that you have, a lot of documentation. You work with these agents all the time. They forget what you're talking about and you need something like this that's going to remember. It's extremely high leverage because it's able to get the information you need faster.”

“There’s going to be more of a premium to style. Making content for SEO—like a 10,000-word post with all the little keywords—is no longer rewarded because that is a dime a dozen now. Having style and substance and weirdness and entertainment is going to be more valued because it's something AI still struggles to replicate with actual soul.”

“Building an element from scratch is a lot of fun. It's also a lot of to learn. And like you said, it's probably the best way to learn how something really works because you can look at figures, but figures can have mistakes. You can look at con concepts, explanations, but you might misunderstand them. But if you see the there is code and the code works, you know it's correct.”

“To do reinforcement learning on an end to end model, you now need to actually simulate all the sensor data. This becomes we we call our our work in this neural simulation, but it's think of it like a hybrid of Gaussian splatting and diffusion methods where you really care about performance. Performance is everything. If you can't do enough simulation fast enough and cheap enough, you actually can't get results that are worthwhile in the end.”

“In the physical AI world, we're not really constrained right now by, like, the intelligence of the models. It's actually what Peter's talking about is actually deploying them in the hardware to give you. And so then there's just a reality is of safety critical systems. So those end up being the your limiting factors rather than, let's say, a limiting factor for a foundation model company which is going to be just capital.”

“So, I think, customers that are new on the multi agentic journey are one of our more naive users. So we have to offer them prebuilt blueprints that they take and run for their use cases. There are savvy customers who know what they want to do, and you offer them those developer kits. So we have all of these different ranges that we offer to the to our customers depending on where they come from, what their use case is, and how can they get to the fastest path to production in the least, constrained manner as possible.”

“People tell me I need to make a Claude skill and sell it, but I'm hesitant because each successive model version needs less and less instruction. Right now you can build skills, but I think the better every single LLM gets, you just won't need that prompt engineering. It is too easy for someone to get the free version of what I would try to sell, so I'd rather use it as a lead magnet.”

“All of the evidence from neuroscience and psychology is that most of what comes into people's eyes is never processed. You're doing fairly fine-grained processing of exactly what you're focusing on, but as soon as it's away from that, you've sort of only processing top-down this very abstracted semantic description of the world around you. Human beings are working with semantic abstractions.”

“The moat of just having a piece of software is not really big anymore. So it's really hard to sell those small tools unless you're in a very niche industry or there's some extra regulations around it like healthcare or law. In the past, I thought if I could just make software and sell it, I would be the king of the world, but the defensibility of it is just eroding very fast.”

“One example would be just seed rounds for startups. That basically haven't even got going yet. And they're raising at tens of billions of dollars, valuations just out of the gate. It's sort of interesting to see, can that be sustainable? You know, my guess is probably not, at least not in general.”

“This Single Brain, you have all this data from within your company. It could be your CRM, all your chats, all your analytics, your Google Docs, all these things. They are feeding into this Single Brain over here. And then your entire team can query whatever data that they want throughout the Single Brain. They can ask about how sales is performing or what are the deals in our pipeline right now that are stuck?”

“Month one, when you're trying to roll this out, it's going to be tough. There's going to be a lot of hallucinations. Things are going to break. You're going to have to reset the models. Then month three, the flywheel starts to spin and then you're starting to see major lifts in terms of what you're doing. It's going to find 10x opportunities on the sales side and content opportunities because it's learning to work with you better.”

“Copywriting course used to be, hey, come to our thing, we'll teach you how to write good copy. Now you can ask ChatGPT or Claude to just write the email for you, write it like Matt McGarry or write it like Neville, and it'll figure it out. The need for that service has gone down and down and down, and we've been niched more into strategy rather than just execution.”

“Physical machines today are more akin to the state of the phone market before Android and iOS existed. Part of the reason that Larry at Google decided to get into Android was they wanted to run Google products on a bunch of phones. At the time they had 50 different operating systems. It was virtually impossible for Google to make their app run on all 50 devices equally well. The state of the physical industry right now is a little bit like that.”

“Of course when we look at option number A keeping two systems in parallel. Well we will probably have no synergy effects. So we can ask the question does it really make sense for the one company to buy the other if we then still keep these two systems there? That's hypothesis number one.”

“I think social media is blogging. You have a piece of content, a picture, a text update, and you post it on a network where people read it and comment. I went from making long-form blog posts to just social media content because I get more out of it now. I've stopped posting traditional blog posts because a blog is almost like a dumb social network where you just post and pray people see it.”

“At the moment, it's a little bit like the systems are just, it's like talking to a person and they just, when they're in a bad day, they're just literally telling you the first thing that comes to their mind. Most of the time, that will be okay, but then sometimes when it's a very difficult thing, you'd want to stop pause for a moment and maybe go over what you were about to say and adjust what you were about to say. But perhaps that's happening less and less in the world these days, but that's still the better way of having a discourse.”

“My personal consulting is all now people who went full AI for their newsletters and emails, and they are already coming back to human-edited content. The AI writes stuff okay, but to completely take over an entire newsletter with AI doesn't work as well as you would think. It currently doesn't have the same vibe, and for some things, it's just better to have a human touch.”

“So what in the past, what used to be thought of as non functional requirements such as latency today is product feature. It is it is baked into the experience of a developer. So these are some things that, you know, we are seeing a paradigm shift in terms of what we need to bring to the fore to the developer experience to keep in mind when you're when you're implementing your systems.”

“If you're simply, you know, trying to predict the next video frame, that's not so difficult. But what you actually want to do is understand the consequences, likely consequences of actions minutes into the future. And to do that, you actually need much more of an abstracted semantic model of the world.”

“I think it's fair to say that, you know, vision understanding sort of stalled out; you got to object recognition and then progress just wasn't being made. There's really an interesting research question as to why that is, and at heart, the ideas behind Moonlake are an attempt to answer that, believing that there can be a really rich connection between a more symbolic layer of abstracted understanding of visual domains, which aren't in the mainstream vision models, which are still trying to operate on the surface level of pixels.”

“The moat of just having a piece of software is not really big anymore. So it's really hard to sell those small tools unless you're in a very niche industry or there's some extra regulations around it like healthcare or law. In the past, I thought if I could just make software and sell it, I would be the king of the world, but the defensibility of it is just eroding very fast.”

“So sometimes people call it jagged intelligences. So they're really good at certain things, maybe even like PhD level, but then other things, they're like not even high school level. So it's very uneven still the performances of these systems. They're very, very impressive in certain dimensions, but they're still pretty basic in others. And we've got to close those gaps.”

“Swipefile has turned more into that where it's all AI analyzed—human picked, AI analyzed. We built a thing where we can now grab Instagram stuff, YouTube stuff, and save all those swipes, and then it gets saved and automatically analyzed. We booked more revenue with SwipeFile this year than all of our copywriting courses combined.”

“Alfred here ran the quarterly Vendor and Ops Audit. It found three immediate savings: redundant CRM licenses, 12 unused seats at 43,000 dollars per year; overlapping SEO tools, three tools doing the same job at 18 grand a year; and then underperforming ad campaigns. Total savings that you can get is 500 grand a year. Basically, we had Alfred do a CFO analysis and it saved, I kid you not, 500 grand.”

“And similar things of course happen to companies that merge when one company buys another or for other reasons they they merge. And how can we avoid heartbreak? How can we avoid the pitfalls? That is what I am going to tell you in the next 45 minutes. Or at least I'm going to tell you some ideas about that.”

“The next few years are definitely gonna be an increased value on physical goods and events and then even more pressure on slop. So there'll be so they'll keep the slop is only starting. The next few years will be more and more diverse versions of slop. Hoping that we society, drowns in slop enough to snap out of it and be like, we can't. Like, none. Like, it just doesn't matter. We all can't deal with it. And then, like, the physical has such a higher premium on it.”

“The most clever sponsorship model I see is where they pay you a flat monthly fee for a set ramp-up period, like six months. Then, unless you hit certain performance goals by that seventh month, you only make the percentage of that goal you actually hit. It gives me some upside of getting paid no matter what for a while, and gives the sponsor a natural out if the relationship isn't performing.”

“I was around, and, basically, Elon, I think, is doing everything possible to throw anything he could possibly invent at OpenAI. And, you know, among them was, you know, kind of claiming that they misled him. Like, he wanted to identify himself as a cofounder of creating it, but then it's, oh, you misled me into into philanthropic donation with a later plan to turn this into commercial. And having been around it, I know it's factually incorrect. We'll see what happens. It seems that it's like it's the kind of the definition of legal harassment, and kind of trying anything possible to kinda say, no. I didn't make a huge mistake when I basically told OpenAI that it should become a company that I own I, Elon, should own 80% north of, and if you're not gonna do it, I'm leaving. And, oh, look. The the organization did well.”

“We moved from a classic ML world to a world where we have LLMs, generating responses. And now we want to move on to a world where actions need to be taken, specific goal oriented actions need to be taken. And when the problem that we are working on is a complex one, with multifaceted, aspects associated with it, That's where multiagentic comes into place. So, basically, we have a large complex goal which we have to break down into specific steps, and each step is basically narrowed to a specific agent.”

“Month one, when you're trying to roll this out, it's going to be tough. There's going to be a lot of hallucinations. Things are going to break. You're going to have to reset the models. Then month three, the flywheel starts to spin and then you're starting to see major lifts in terms of what you're doing. It's going to find 10x opportunities on the sales side and content opportunities because it's learning to work with you better.”

“The next stage is agent-based systems, which I think we're going to start seeing. We're seeing now, but they're pretty primitive. Like in the next couple of years, I think we'll start seeing some really impressive, reliable ones. And I think those will be incredibly useful and capable, if you think about them as an assistant or something like that, but also they'll be more autonomous. So I think the risks go up as well with those types of systems. So I'm quite worried about what those sorts of systems will be able to do, maybe in two, three years time.”

“The Single Brain concept, not only that, you have a fleet of agents that it's also powering. So this fleet of agents, they are the ones that are doing the thing. So we have our agent called Alfred, and then we have Arrow here for sales, we have Oracle for SEO, Flash for content side work. I mean, look, this piece over here, by the way, this piece got 354,000 views. This was created by Flash.”

“There's a lot of delving now that wasn't happening before. Yeah, probably, yeah. You know, actually, whenever I see the word delve in anybody's writing, I'm like, hmm, what did you do?”

“My personal consulting is all now people who went full AI for their newsletters and emails, and they are already coming back to human-edited content. The AI writes stuff okay, but to completely take over an entire newsletter with AI doesn't work as well as you would think. It currently doesn't have the same vibe, and for some things, it's just better to have a human touch.”

“I can give you one example of our recent work called the prismatic synthesis. It's a synthetic data generation algorithm which is prismatic because it acts like a little bit like a prism that can scatter the light to make it more diversified... we're doing this using Dipsic R1 32 billion parameter model as the teacher model... our goal is to compete against the alternative, which is to use much stronger teacher that's 20 times larger... we find that that one million data points is actually better than the one million data points that you generate from the stronger teacher model, the best teacher model.”

“The reality is that although the visuals do look fantastic, those visuals actually aren't accompanied by an understanding of the 3D world, understanding how objects can move, what the consequences of different actions are, and that's what's really needed for spatial intelligence. So, I mean, a term we sometimes use is that you need action conditioned world models, that you only actually have a world model if you can predict, given some action is taken, what is going to change in the world because of it.”

“Alfred here ran the quarterly Vendor and Ops Audit. It found three immediate savings: redundant CRM licenses, 12 unused seats at 43,000 dollars per year; overlapping SEO tools, three tools doing the same job at 18 grand a year; and then underperforming ad campaigns. Total savings that you can get is 500 grand a year. Basically, we had Alfred do a CFO analysis and it saved, I kid you not, 500 grand.”

“It's held for 13 orders of magnitude of computer something. Like, why would it ever end? So I think fundamentally it is pretty unlikely to stop. It's just like, eventually, we're not even gonna be able to test the bigger scales because of all the problems that come with more compute.”

“My name is Henna. Um, I'm known as the guy with the many kids in the community. So, I have three sons, three daughters, and they have three different mothers. So, I know how to operate in complex situations. And that's important for what I do at work where I'm a coder, coach, and consultant at a company called WPs.”

“In more traditional development, right, you you oftentimes would have, more black and white answers to questions. But what's changed now is with these models, everything is statistics. Right? Like, you no longer have a black and white answer, but it's like, well, how many orders of magnitude or how many nines of reliability can can I get in the system, and how can I how can I prove that to be true?”

“Anthropic is known for betting very hard on code, which is called code thing, is working out for them right now. So I think that even if the ideas flow pretty freely, so much of this is bottlenecked by human effort and kind of culture of organizations where anthropic seems to at least be presenting as the least chaotic, it's it's a bit of an advantage.”

“The what's different about us is we're deploying intelligence onto a lot of things that don't have screens. Most of the value we provide is putting intelligence that is in safety critical environments. So that the those two words are really important because learned systems can make mistakes if you're asking for something like tell me about these podcast hosts, but you can't do that when we run driverless trucks in Japan right now. We can't have errors.”

“And I'm working on in my spare time, my two minutes of spare time, you know, physics theories about things like information being the most fundamental unit, should we say, of the universe, not energy, not matter, but information. So it may be that these are all interchangeable in the end, but we just sense it. We feel it in a different way. But, you know, as far as we know, this is still all these amazing sensors that we have, they're still computable by a Turing machine.”

“I think part of what you've seen in the large tech companies is they would prefer to be buying companies versus doing these kind of, you know, kind of strange deals, but feels that the regulatory environment is too hostile to for them to do that, especially earlier. And so are looking for other ways to make it work. And so you say, well, what's a deal that you could make work that isn't an acquisition? It's like, well, you know, get a a, like, a, like, kind of something that's not quite a BD deal, but not kind of corp dev deal. Corp dev like, it's kinda, like, halfway in between, and the measurement is, does the business continue, you know, with with vigor and effectiveness afterwards?”

“The Single Brain concept, not only that, you have a fleet of agents that it's also powering. So this fleet of agents, they are the ones that are doing the thing. So we have our agent called Alfred, and then we have Arrow here for sales, we have Oracle for SEO, Flash for content side work. I mean, look, this piece over here, by the way, this piece got 354,000 views. This was created by Flash.”

“People tell me I need to make a Claude skill and sell it, but I'm hesitant because each successive model version needs less and less instruction. Right now you can build skills, but I think the better every single LLM gets, you just won't need that prompt engineering. It is too easy for someone to get the free version of what I would try to sell, so I'd rather use it as a lead magnet.”

“Llama was the, I would say, pioneering open weight model, and then Lama one, two, three, a lot of love. But I think then I think what happened just hypothesizing or speculating. I think the, leaders at Meta, like, the upper, executives, they I think they got really excited about Llama because they saw how popular it was in the community. And then I think the problem was trying to, let's say, monetize the open not monetize the open source, but, like, kind of use the open source to make a bigger splash.”

“The interviews that we give now, I think, are way harder than they've ever been, but but we also allow, right, selective use of AI tools to solve the problems. And I think in that, you you start to see more of a bimodal distribution of engineers. Right? You you start to see, like, wow. There's there's this subset of of people that they they really get it. They're all in, and they've clearly invested the hours needed to learn these tools.”

“So for example, in the inflection cases, we're not gonna be able to establish our agent with building frontier models in the way that we hope. We need to pivot. So we're gonna pivot to p two b. How do we get the capital? How do we make that happen? Well, we do a deal where that that enables that to happen. But that that was a that's a that's a pivot deal, not a like, it wasn't Microsoft came along and kinda said, hey. A little bit of cash. You know? You completely upend your business. I was like, no. No. We've we've actually already decided that we need to change, and we're trying to figure out how to fund that change and make that change go through.”

“Chat Concierge for us was, our beachhead initiative around deploying multi agentic deploying a multi agentic solution. So, Chet concierge is essentially a a auto deal dealership, project or application that was deployed out to our auto dealers to basically bridge that experience between dealers and their customers and make it very seamless. And we need to understand we are moving to a world where the car buying experience doesn't start at the dealership. It starts before. It starts when they go to their website and try to figure out, okay, what's the inventory?”

“Yeah. We've just announced partnership with a deep one. We already were collaborating with them, but it's a much deeper one now with Commonwealth Fusion who I think are probably the best startup working on at least traditional tokamak reactors. So they're probably closest to having something viable, and we want to help accelerate that, helping them contain the plasma in the magnets and maybe even some material design there as well.”

“I don't think nowadays, 2026, that there will be any company who is, let's say, having access to a technology that no other company has access to. And that is mainly because researchers are frequently changing jobs, changing labs, they, rotate. So I don't think there will be a clear winner in terms of technology access. However, I do think there will be, the differentiating factor will be budget and hardware constraints.”

“The mission really is democratizing general AI, so that it's not just companies who can purchase a lot of GPUs, are able to create LLMs and adapt to LLMs and serve LLMs, but also people like myself and colleagues who are academics, so for example, cannot buy as many GPUs, and is there something really meaningful and fun that we could do, even with a smaller counterpart? And at the end of the day, I believe that fundamentally it should be feasible. It's only that the world has invested so much more into exploring what happens when you scale things up so much.”

“When I was actually working with Nvidia on the Synthetic Data Foundation Model Training Project, we were actually generating a lot of these synthetic data and showing that these synthetic data are actually as useful as real-world data when it comes to multimodal pre-training. But then, there's a lot of dollars being paid out to external vendors or other folks to manually curate these types of data.”

“Well, look, it's probably my longest standing passion is world models and simulations in addition to AI. Of course, it's all coming together in our most recent work like genie. I think language models are able to understand a lot about the world. I think actually more than we expected, more than I expected, because language is actually probably richer than we thought. But there's still a lot about the spatial dynamics of the world, spatial awareness and the physical context we're in, and how that works mechanically that it's hard to describe in words and isn't generally described in corpuses of words.”

“Overtone pluralism means when you ask a question that's politically thorny, for example, that could have different answers, the best way might be for LLM to just present all of them. All of the reasonable opinions as, hey, the answer is that people have different opinions. Here's one view, there's another view, and be able to include all of them as opposed to picking the majority opinion. Because that marginalizes out the rest.”

“Mode collapse is a real concern with LLM generation. So, what we find in our paper is that even when you ask open-ended questions, like, you know, tell me a joke about time, or tell me something wise about time. Even when you ask, by the way, hey, give me a random number between and 10. It's not random... The bigger problem is after post-training, like sequential fine-tuning and RL, the probability, output probability of the model becomes even more skewed, like zoning in to the stereotypical answers that people tend to like.”

“Every job that uses language or information will have an AI component that will range, in certain cases, from automation, in certain cases, much more AI work, and in certain cases, kind of AI amplification or augmentation of human labor. And I think that will be true of all of those things. And so almost every industry, because if you think about, like, even, like, steel manufacturer, you still have meetings, financial analysis, you know, kind of, capital, other kinds of thing, you know, legal, other kinds of things, all that play in it, you know, strategic planning, and so so that means that AI will touch everything. Now and I, by the way, inherently also agree with you that, one of the general problems when you get new technology, kind of put this in super agency, is you go, well, I can see if I take jobs as a fixed number that do not grow and do not change, and I can see that a bunch of them will be, you know, now will the horses then care the buggies will go away and we'll have cars and, like, all the grooms people will go away and all the horse carrying jobs will go away and all the horse cleanup jobs will go away.”

“But then we thought, well, wouldn't it be fun if we plugged Genie into Simmer and sort of drop a Simmer agent into another AI that was creating the world on the fly? So now the two AIs are kind of interacting in the minds of each other. So the Simmer agent is trying to navigate this world, and Genie is, as far as Genie is concerned, that's just a player and an avatar doesn't care that it's another AI. So it's just generating the world around whatever Simmer is trying to do.”

“You need to make sure that you have the right hardware. So we have a Mac Mini here, we have two DGX Sparks from NVIDIA, and we run it on local inference. This is what we run it on for my agent fleet, for example. And the thing is these local models, by the way, they're getting a lot better, like Google's Gemma. The cost savings are going to get better and better over time.”

“The most clever sponsorship model I see is where they pay you a flat monthly fee for a set ramp-up period, like six months. Then, unless you hit certain performance goals by that seventh month, you only make the percentage of that goal you actually hit. It gives me some upside of getting paid no matter what for a while, and gives the sponsor a natural out if the relationship isn't performing.”

“All the more important that observability comes to the forefront in stochastic systems like a multigenic application. All the more important for us to be able to replay agentic actions and try to understand how it function. Agent behavior needs observability along many different dimensions in terms of what are the tools involved, what was the reasoning mechanism that led to that tool invocation. And, overall, what was this context that passed across systems?”

“This Single Brain, you have all this data from within your company. It could be your CRM, all your chats, all your analytics, your Google Docs, all these things. They are feeding into this Single Brain over here. And then your entire team can query whatever data that they want throughout the Single Brain. They can ask about how sales is performing or what are the deals in our pipeline right now that are stuck?”

“I'm thinking about an example that we, you know, we've all heard about. I forget the the very specific scenario, but, a business had a chatbot. I think it happened in Canada, had a chatbot on their website, and the customer asked for a discount, and the the chatbot basically gave them a discount. You know, this would clearly be disastrous in a, in a car dealer type of scenario. Like, how do you, make, you know, generative AI and agents safe, you know, for, you know, these dealers that have a lot at stake?”

“You need to make sure that you have the right hardware. So we have a Mac Mini here, we have two DGX Sparks from NVIDIA, and we run it on local inference. This is what we run it on for my agent fleet, for example. And the thing is these local models, by the way, they're getting a lot better, like Google's Gemma. The cost savings are going to get better and better over time.”

“Swipefile has turned more into that where it's all AI analyzed—human picked, AI analyzed. We built a thing where we can now grab Instagram stuff, YouTube stuff, and save all those swipes, and then it gets saved and automatically analyzed. We booked more revenue with SwipeFile this year than all of our copywriting courses combined.”