The AI Model Built for What LLMs Can't Do

Most AI companies are racing to build bigger LLMs. Eve Bodnia thinks that's the wrong approach. Eve is the founder and CEO of Logical Intelligence, which is developing an alternative to the transformer-based models dominating the industry. Her argument: LLMs’ architecture makes them fundamentally unsuited for some mission-critical tasks. A system that generates output one token at a time, with no ability to inspect its own reasoning mid-process or guarantee its results, shouldn't be trusted to design chips, analyze financial data, or even fly a plane. Her alternative is the energy-based model (EBM), a form of AI rooted in the physics principle of energy minimization, not language prediction. Rather than guessing the next probable word, an EBM maps every possible outcome across a mathematical landscape, where likely states settle into valleys and improbable ones sit on peaks. Dan Shipper talked with Bodnia for AI & I about why she believes LLM progress is plateauing, what it means for AI to actually understand data rather than just pattern-match across it, and how her team is building toward formally verified code generated in plain English—no C++ required. If you found this episode interesting, please like, subscribe, comment, and share! Head to http://granola.ai/every and get 3 months free with the code EVERY To hear more from Dan Shipper: Subscribe to Every: https://every.to/subscribe Follow him on X: https://twitter.com/danshipper Timestamps: 00:00:51 - Introduction 00:02:09 - Why correctness and verifiability matter in AI 00:09:33 - What an energy-based model is 00:14:21 - How EBMs construct energy landscapes to understand data 00:19:00 - Why modeling intelligence through language alone is a flawed approach 00:26:54 - What it means for a model to "understand" data 00:37:21 - How EBMs solve the vibe coding problem and enable formally verified code 00:43:21 - Why LLM progress is plateauing 00:49:54 - Mission-critical industries haven't adopted LLMs, and how EBMs could fill that gap

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Dan Shipper?

  Can you define EBM for us?

• 0:01Speaker 01

  EBM are naturally non-autoregressive. There are no sequences of tokens. And that's what makes it fundamentally different. Like, imagine you're trying to navigate the map and you have a left brain. To navigate, you're sort of allowed to choose one direction at a time. And sometimes you take the wrong turns just because you hallucinate. Like, there might be a hole in the road and you're just gonna fall. And you might see this hole, but you cannot turn back because you're an autoregressive LLM. EBM gonna have the bird view all the time. So if you see there's a hole, you're gonna choose a different route.

• 0:48Dan Shipper?

  Eve, welcome to the show.

• 0:51Speaker 01

  Hi, thanks for having me.

• 0:53Dan Shipper?

  Great to have you on. For people who don't know, you are the founder and CEO of Logical Intelligence. Tell us what Logical Intelligence does.

• 1:01Speaker 01

  So Logical Intelligence does a few things. First of all, we see ourselves as a foundational AI company. So we work in both with EBM and LLMs. So everything is built in-house. We prototyped on LLM initially and we're building EBM at the same time. And that sort of gets plugged in in the long term.

• 1:26Speaker 01

  We focused on correctness of software and hardware as a product.

• 1:33Speaker 01

  Because I believe there is a lot of issues with AI being placed in mission-critical systems today. Like, you know, can we do the code gen? Can we do the chip design? And the answer is yes. Yes, people use LLMs today. But very few actually questioning of like, how these results are actually like, correct? Does it make sense what it produce? And it seems like there's a big gap on market today, having deterministic AI, verifiable AI. So we're trying to fill that gap.

• 2:10Dan Shipper?

  I the the place my my brain goes first is why does correctness or whether something makes sense? Why does that matter if it works?

• 2:22Speaker 01

  Um, actually, let me ask you a question back. So speaking of correctness. Um, I don't know. Well, imagine there is AI driving a car and you are in that car and that car is an L L M. And someone tells you like, um, you know, twenty percent of the time it's gonna hallucinate and you might end up like in in like wrong place. Uh, how would you feel about it?

• 2:46Dan Shipper?

  Well, I I think in my in my case, I'd be like, wow, that's kind of interesting. I'm I'm curious where it takes me. Um.

• 2:53Speaker 01

  Okay.

• 2:55Speaker 01

  Yeah, let me give you another example.

• 2:56Dan Shipper?

  Yeah, sure.

• 2:57Speaker 01

  Plane, how about the plane? You take a plane from SF to New York. And someone says, you know, like twenty percent of the time it might just like the next word not gonna match and it's gonna go down. So how would you feel about it?

• 3:08Dan Shipper?

  Yeah, my my my feeling about that is planes are currently run very well by deterministic systems. So I don't know why I would need an AI for that.

• 3:17Speaker 01

  Um, I feel like we just cannot avoid AI anywhere. Like next ten years, people are gonna try to place AI everywhere. Automate systems with AI and, you know. Technically, you might not need. We we survived somewhat without AI up to this moment, but now it's just like a next step of evolution that people just want AI everywhere. Like for the banking, you don't need AI initially, but we learn it's really helpful to automate like certain processes in decision making. And it's gonna save us a lot of time and allow a space to be creative instead of like debugging and fixing things. Um, so I just feel like it's unavoidable future.

• 4:03Dan Shipper?

  I think maybe what I'm getting at is, um, what am I getting at?

• 4:11Dan Shipper?

  It seems like if you want a guarantee of certainty using,

• 4:18Dan Shipper?

  the only way to sort of guarantee certainty is to use something that you can express in code or logic.

• 4:26Speaker 01

  Hmm, that's a part of it. So the certainty comes from internal verifiers and external verifiers, at least for us. So for example, if you take LLM, obviously, it's a language-based model and architecture doesn't allow you to do internal verifiers. So you, you like, it's like a black box for you. You don't have access to what's inside until it's all processed. But you have access to the output and many people and companies sort of take LLM, train it for certain tasks, and if it requires logic, they attach texture and verifiers to it,

• 5:07Speaker 01

  such as languages like Lean4, which is a proof of, I mean, machine verifiable language, proof language, which allows you to check this output using mathematical frameworks.

• 5:21Speaker 01

  However, you know, it doesn't solve the problem of things being

• 5:27Speaker 01

  just so expensive, because what's expensive is your architecture, which is still playing a guessing game out here. And even if you attach external verifier, even you fine-tune this LLM specifically for the task you're trying to create, you're still not solving the problems of tokens being expensive. It takes compute for you to play a guessing game. So this problem is solved by the EBM's, but we're talking about LLMs for now. So here we have the situation when there is, right, internal absence of verifier, but there's external one. So now about the EBM's. EBM's don't have tokens. It's token-free model. There's no guessing game of this kind. So essentially,

• 6:13Speaker 01

  you could oversee all the possible scenarios.

• 6:16Dan Shipper?

  Can you define EBM for us?

• 6:18Speaker 01

  Yeah, I'll define in a second. So for now, just think of it as something which doesn't play a guessing game. And something which has architecture, which is essentially allow you to self align itself as a processing the information. And it's no longer a black box for you. So as it's performing, you can open it anytime during the training and you could see what's happening in there. So you cannot do this with L L M's. Just nature of architecture is different. So you have for verification tasks, you have this notion of self alignment because of the EBM architecture and the absence of token makes it cheap.

• 6:57Speaker 01

  But also you have external verifier on top of it. So you have verification sort of on both sides, inside and outside. Um, hopefully that makes sense.

• 7:08Dan Shipper?

  I think so. Let me, let me play it back to you and you tell me if, if, if I'm getting you. So basically, I think what you're saying is we're living in this, in this world, which is really cool with L L M's, which is we can generate lots of output with them. And the output is um, uh, really useful for, for a lot of different things. But in order to tell if the output is right. The, the best we can do is sort of guess and check. We generate the output and then, for example, if it's code, then we go and check the check the code with integration tests or manual tests or whatever, just to see if it works. And that totally works, but it is expensive and time consuming. Um, and uh, one of the problems is it's very hard for us to know, okay, how did the L L M get to this answer?

• 7:52Dan Shipper?

  We can't like go look inside of it.

• 7:54Speaker 01

  Exactly.

• 7:55Dan Shipper?

  And I think what you're saying is there are other types of models that are a little bit more inspectable.

• 8:04Dan Shipper?

  And that give us a sense before we even try the um, before we even try the output to understand, does this work? Does the output work? We can get a sense from the model by looking at its internals. So sort of like, how good is this solution? How good does this model think this solution is? And it's sort of like being able to ask someone like, are you sure about this? Like, how, how, how good is this before you like, go check their work? And a language model can answer that question. But a language model's answers are working at a, at a different level when it answers that question than, than this models. Then, then these E M B models are working. And the, the, the answers from E M B models are more likely to be correct.

• 8:48Speaker 01

  Uh huh. Yeah, so you always have an opportunity to see what's inside with the E B Ms. And you control the training.

• 8:57Dan Shipper?

  E B M, sorry.

• 8:58Speaker 01

  Yeah.

• 9:00Speaker 01

  So the, the E B Ms, you control the training. It's no longer a black box for you. You control sort of how the training goes. Well, you do some extent with the L L Ms, but you need to wait until the training is done. Before like, you actually go and like, see what's inside. In here, you could do like a real time.

• 9:20Speaker 01

  Yeah, and also you can attach the same external verifiers which works for L L Ms. So you have sort of double, double verification things.

• 9:29Speaker 01

  Yeah, so you asked me, what is a E B M? I just wanna give like a historical note. Because I feel like there's so many terms today and just people throwing those terms without defining it. So E B M just simply means energy based model. What is energy? Energy based? It comes from physics. It's a very popular term when they're trying to minimize the energy. And if you're doing theoretical physics, like your full time job is just to write Lagrangians, which sort of correspond to uh, terms associated with the energy in your system. Like, hey, this is my kinetic energy, this is my potential energy.

• 10:11Speaker 01

  And then you're trying to derive equations of motion of it. And the way you derive the equations of motions is you're doing the minimization. So that's pretty much how whole theoretical physics works. Hey, just start with the energy terms, then you minimize this energy and you derive equations of motions. And equations of motions are gonna give you a conservation loss. So you're gonna know exactly what are your laws about your system. And this principle is fundamental principle, like everything wants to minimize energy around us. Um, yeah, so like even as we talking to each other, we sitting on the chairs, we not like jumping and running around because it's a natural state when we minimize the energy.

• 10:54Speaker 01

  So we're just using this minimization energy principle as AI is processing information in high level terms. So the term energy based minimization doesn't really mean anything specifically to AI. It's just the whole like idea of like, hey, let's take some energy and try to minimize it and discover what's the laws about it. Um, so our model is called official name of that model. Uh, even though we call it Kona, just because we like big fans of coffee culture and Kona is one of our favorite kind. So we decided to start with that. Uh, the formal name of the model is called energy based reasoning model with weight and variables.

• 11:38Speaker 01

  And I'm gonna like describe exactly what this words means. So we already understand what the energy minimization is.

• 11:44Dan Shipper?

  Can I actually pause you because I want to make sure that we do understand what the energy, energy based minimization is.

• 11:49Speaker 01

  Okay.

• 11:49Dan Shipper?

  Yeah.

• 11:50Speaker 01

  It it's just for now, think of it is just something which minimizes the energy. It it means this AI architecture has a framework which allows you to construct the energy function of your system and minimize it.

• 12:02Dan Shipper?

  I get it. I I just think that, um, so I just want to make sure for for people listening, they understand what it means to minimize energy, what what energy is and what it means to minimize it. So I'm curious, um, give me a, give me a, uh, uh, tell me if this concrete example is is about what you're like, sort of, uh, like, close to what you're talking about.

• 12:24Dan Shipper?

  So if I'm going to, let's say, uh, I'm gonna go lie on the couch behind me. And, um, and I'm trying to, I'm trying to predict or understand how is my body going to be lying on that couch? Um, given the laws of gravity.

• 12:46Dan Shipper?

  The couch is uneven, my body is uneven. And so I'm trying to like sort of understand the fit of like how my body is going to end up settling onto that couch.

• 12:55Dan Shipper?

  I'm going to end up settling onto the couch in a way that minimizes energy. So there's going to be like a good fit between my body and the couch. Versus like me being sort of like jerky like this and and having lots of different spaces. Is that the sort of the sort of energy minimization that you're talking about?

• 13:10Speaker 01

  Yeah, yeah, you're just, it's all about your body finding the most comfortable configuration for you. Which gonna correspond to the like the lowest potential of your body. I would even tell the like even more high level example of this, like, you know, you Dan, you just like imagine you're tired. You like don thousands of podcasts and you just came home and someone is asking like, okay. Dan is a variable here. Let's try to figure out what's his equations of motion in the house. And where he's gonna most likely to end up. So you're probably gonna end up in the couch.

• 13:55Speaker 01

  With like a nice show and probably some drink.

• 13:59Dan Shipper?

  Yeah.

• 14:00Speaker 01

  Yeah, so that's gonna be a law like, okay, when Dan is tired, he's gonna go and sit on the couch and just relax. But to get there, we're gonna look at all your possible states, like you washing the dishes, you know, walking around the house. So those are gonna be different states. But your most probable scenario is gonna be on the couch. So essentially, all of this picture can be mapped into something we call energy landscape. When there the high, it's gonna have like, it's gonna look like a map. So you're gonna have highest points, you're gonna have lowest points. The highest points, we can associate less probable scenarios. So probably if you're tired, you're not gonna dance around, although I don't know.

• 14:42Speaker 01

  But, you know, typically people assume that if you're tired, you're probably gonna want to relax. So that's gonna be the lowest point. And as we're trying to figure out where you are during the training, we're gonna observe you multiple times during different days. And you know, how much of the workload you have, it's gonna be a variable. Your internal state is gonna be a variable, how your body feels. And eventually, we're gonna train this landscape to be, you know, based on what we see in the real world, right? The lowest point is gonna be you on a couch.

• 15:18Dan Shipper?

  We've all been there. You're sitting in an important meeting and you're trying to pay attention. You're trying to stay present, but you have this lingering, underlying anxiety that you're gonna forget everything, that you're gonna miss the important detail, forget the decision, forget the action item. Let something important slip through the cracks. That's why I love Granola. It's an AI-powered notepad that works in the background while you're in your meetings. It takes notes on everything that gets said, transcribes action items, and helps get rid of that feeling. You don't have to worry about whether you're gonna miss something because Granola has you covered. And now it's you stay present in meetings. I've been using Granola for a long time, almost since they came out. And it's amazing for this. It doesn't join the meeting like some of those other clunky meeting note takers. The UI is really fast and well-considered.

• 16:02Dan Shipper?

  And it feels like it's sort of just transcribing all the important moments in my work life. And that gives me the confidence to get great work done. And what's even cooler is you can chat with your notes afterwards. You can run detailed research reports on how your week was, how you act as a leader, how you performed in particular difficult conversations, and how you can be better. It's really a power tool for anyone who cares about their meetings and also cares about how they show up in those meetings. It also has these things called recipes, which are pre-made prompts for common tasks like negotiating, coaching, or summarizing. I even have a recipe that I made that's in Granola that you should check out. Once you try it on one meeting, it's really, really hard to go back. The notes are always better than what you could do manually. And it helps me be much more present instead of frantically typing all the time.

• 16:45Dan Shipper?

  Head to granola.ai/evry for three months free with the code EVRY. That's granola.ai/evry for three months free. And now, back to the episode. Okay, that makes total sense. Now I want to relate this to LLMs for a second, because you can imagine that there's an LLM that's trained to predict where I end up after a long day of podcasts.

• 17:09Dan Shipper?

  And you can imagine it probably would also end up predicting that I would end up on a couch. What are the differences in the ways that it makes those predictions? That make energy MACE models better for this scenario?

• 17:23Speaker 01

  Okay, that's a good thought exercise. So, okay, now you are LLM. So, okay, let's talk about back to EBM, because what we described is very literal about EBMs. EBMs are all about constructing energy landscapes and how we navigate those energy landscapes. And energy landscapes is sort of the maps of your states based on the data we observe. So in your case, we're just gonna look at you in all possible scenarios. All of these possible scenarios gonna mapped into energy landscape, highest point, less probable scenario, lowest point is more probable.

• 17:59Dan Shipper?

  Very probable I end up on the couch.

• 18:01Speaker 01

  Yeah, yeah, yeah.

• 18:08Speaker 01

  There might be some other additional low points, like sometimes you might go to gym.

• 18:15Speaker 01

  You feel tired and you might go to gym. So it's gonna be, you know, lowest points compared to everything else, but some of them are gonna be lower.

• 18:23Dan Shipper?

  Yeah.

• 18:24Speaker 01

  Yeah, so that's the situation. So this is how energy-based model, actually, we think. It just takes a data and map it directly to this energy landscape, and then we use certain algorithms to navigate this. But there are different kinds of energy-based models today, so I'm gonna talk about it a little bit later. But the whole idea is just, hey, let's map it into the structure and navigate the structure. And as you see, as we like map into this, there are no tokens. We don't predict any tokens and so on. So that's already a crucial difference. How would LLM think? LLM, yes, it's gonna rely on the training data, and it's gonna be a lot of training data, like a lot and a lot of observations of how you behave.

• 19:09Speaker 01

  And to figure out where you would end up, it's gonna be attached to probabilities of your next token.

• 19:20Speaker 01

  If that makes sense, and those tokens gonna come from words. And what usually bothers me about L L M's, it's intelligence, which is language dependent. Like, our brains, we are intelligent, I'm relatively intelligent. So, like, I speak different languages and none of my thoughts processes really depend on any language. Like, I could just think in an abstract way. And then I speak different languages and decode the information in the channels. And with the L L M's is like, if you're searching for the next token in certain words, like

• 19:56Speaker 01

  intelligent process, I I would say the information processes in French are gonna be different from what's in English, just because like words naturally gonna be end up next to each other. So see what I'm saying? It's like, and then we have so many languages in the entire world, and you have so many L L M's trained on different languages, so you're gonna end up reasoning, you're gonna end up having reasoning processes different for each of the language, which feels really wrong. So in this case, observing you walking around the house has nothing to do with language then. It's a pure visual spatial reasoning tasks, just looking at your body navigating the space-time

• 20:40Speaker 01

  and geometry of your house. So we need to map that information in the language space, find the right words and embeddings. And then we start associating those tokens with the probabilities based on what the data we see from you. So we're trying to map something absolutely has nothing to do with language into language space and think about it in that space, which feels really wrong.

• 21:09Speaker 01

  And I don't know, I I just realizing that. For many people, it's counterintuitive, just because a LLMs is the first form of AI we sort of know. And it's the most popular form of AI today. Like, for for many people, it's by default, like, oh, yeah, we're just gonna use language to navigate the world, to drive a car. But and I'm like, every time I'm speaking, I'm like, well, let's wake up, let's like actually see when you drive a car. When you walk around your house, how much language you actually use, are you trying to predict next word as you navigate yourself around the house? Probably not. You just use your visual data, your state of the body, and you know, you just move your body, right? Without speaking.

• 21:54Dan Shipper?

  Uh, there's a lot here. I I'm really into this conversation. So I wanna start with. A, it seems absolutely right to me that um, there are many different ways in which we.

• 22:09Dan Shipper?

  Process information or many different ways in which intelligence can occur and and only a few of them are verbal. Uh, but there's.

• 22:19Dan Shipper?

  There's certain things that come up for me when I when I think of this. One is.

• 22:26Dan Shipper?

  Language models happen to work with languages as their primary way of working. But really, they just work with sequences of tokens that have um, weak correlations, many thousands of weak correlations between each token. That helps us to know which comes next. So even though it might be unintuitive to model my behavior inside of my apartment. With like specifically with language, although I I'm gonna, I think there is some interesting things there that it might be related to language. We we could model it as just like a sequence of movements, right? That were um, uh, one, one movement is weakly correlated to the next one that we sort of.

• 23:07Dan Shipper?

  Have a trajectory of movements that tell us where where I'm going. Why is that not a good way to model things?

• 23:15Speaker 01

  It's a good way to model things and you don't need a LLMs for it. You need a form of AI which is not attached to language. But it can be compatible with language if you want it to, and that's what our model is about.

• 23:30Dan Shipper?

  Right, I guess what I'm saying is, um, uh, for- for- forgetting about the, the language part of it, just like modeling my movements as this string of, uh, correlated events, like an event stream, where each token is like one next thing I do.

• 23:46Speaker 01

  Yeah, you, you can do it, and people do it today, right? People even do image recognition using language models. Uh, you could be really creative, but it's like, that's what makes it expensive and super slow, because you're trying to play a guessing game what my next token could be, and this is what makes it extremely expensive. So, like, you could do it, but you don't have to do it. You just can use different architecture, which is more suitable for non-language related tasks, such as spatial reasoning, or applied engineering is another example of spatial reasoning. Like, when you build a bridge, you don't go to literature department, you go to engineering school and learn formal methods, right? So here, we are trying to use literature department everywhere, and I'm like, hey, we don't have to.

• 24:30Speaker 01

  There are BBMs, there are also other forms of AI, which you can experiment with, and you don't have to do everything through language.

• 24:39Speaker 01

  It's a matter of like, it's, it's, right? It's like energy-based minimization principle when it comes to your resources. If you have infinite money, and you don't care about the time scale, sure, you can do everything. You can attach it to language, you can attach it to, I don't know, your cat movement around the house, and connect it to the cat movements, and you know where cat goes, and your next token goes, and we decide where you're gonna go. You could be really creative, but if you wanna minimize your resources, and you don't have opportunity to wait, like, for example, if your AI controls the circuits, you probably cannot wait even, even a second. It's all milliseconds, microseconds.

• 25:21Speaker 01

  So, it just, this form of AI is not suitable for those tasks.

• 25:26Dan Shipper?

  So, basically, if I'm understanding this right, if I'm spending tons and tons of tokens, and I'm looking for a more efficient, more direct way to, um, predict some, some of these solutions to these, these problems, an energy-based model is gonna get me there faster than a, you know, modeling it with tokens. Is there, is it also able to do with less training data?

• 25:50Speaker 01

  Um, yes, actually, the beauty of the of the E B Ms is it's really good at working with sparse data. Because like, you know, there's evolutions of like traditional E B Ms, which were applied for the L L Ms. Then there was diffusion models. And diffusion models came from the fact that sometimes you don't have enough data to train a models or your data is just, data set is incomplete. So there are ways to reconstruct those energy landscapes by injecting certain noise and changing the navigation strategies. Um, so that's what diffusion's models were about. And the E B R M with latent variables is just like, hey, on top of the diffusion stuff, we also understand the data.

• 26:33Speaker 01

  We're not just taking any data, but we also understand why the data looks the way they are. So that understanding goes to the latent variables. Just like latent space in your brain sort of understands the world around you and keeping you on top of your tasks and allows you to like predict and plan. So it's the same idea here.

• 26:54Dan Shipper?

  So now we got to the blatant variable part of our part of it. So I I would love when you when you use the word understanding, I I think that must mean something very specific to you. Can you help me understand that and how it relates to latent variables and what those are?

• 27:07Speaker 01

  Yeah, so that's also back to your question was that like how L L Ms are different with from the those kind of E B Ms we're creating. Um, L L Ms don't understand the data. It's just you feed a lot of data into it and it's sort of like, hey, I I got it like, okay, I know what's the most probable scenario here and here we are. However, here, E B M, you can feed a lot of data. It's not just gonna look at like, hey, I I see the biggest pattern here. It's gonna try to understand the pattern. And that understanding, that knowledge, gonna go to latent variables. So what is understanding about data? Like, it's just basic knowledge about the world.

• 27:49Speaker 01

  Basic rules about the world, like i- if there is a couch behind then. It's probably because he likes to sit on it. Or be- because he likes it on the background, so. There are little rules you can guess about. You being as a data point and your couch being data point. And then there's, you can try to create those kind of rules, like for everything, right? For you navigating your apartment, there are little rules, like. You know, there's a kitchen for cooking, there's, I don't know, bathroom, there's sofa, there's your bed. So that, that understanding. Allows you to have your own mental world model for your brain.

• 28:33Speaker 01

  Um, which helps you to understand your environment. And if something changes in your environment, you understand the rules, like. If somebody brings you a different couch, different shape, you're still gonna know what to do with it. So that's an example of how you can infer what to do with something new, based on what you already know.

• 28:55Speaker 01

  So with, with people, it's kind of comes natural because of the evolution and so on, but with AI. We need to teach it, so we need to like mimic that evolution. And what latent variables allow you to have here is like, hey, let's look at the data, but let's also try to understand the data. Let's look at, you know, if you deal with numerical analysis, we're gonna look at all possible correlation functions. And the model is gonna be creative, it's gonna try to figure out what's the total. State of the energy and minimize and figure out the, the, the laws about your data. Um, but uh, there are so many creative ways how you can infer those rules.

• 29:32Dan Shipper?

  What is, uh, so is, is a latent variable equivalent to a rule in this scenario? Like, uh, if there's a couch in my apartment, I sit in it?

• 29:41Speaker 01

  It's not equivalent to a rule, but it's equivalent to. Something which holds the knowledge about the rules of your data. It's like a knowledge storage.

• 29:54Dan Shipper?

  So it it it has many rules in it?

• 29:57Speaker 01

  Yeah, you could have many rules.

• 29:59Dan Shipper?

  So one latent variable has many different rules.

• 30:02Speaker 01

  Yeah, it's just like a knowledge data set essentially about your data.

• 30:08Dan Shipper?

  Is it an explicit data set? As in, like, does it have key value pairs of rules? Or is it a?

• 30:15Speaker 01

  It's it's in a form of energy landscape. It's just another energy landscape you're gonna navigate. So essentially, we take the data. We look at the data. We construct some sort of structure for AI to deal with the data. So it could start learning the rules about the data. And once it understands the rules, it stores its knowledge in the latent variables in the form of energy landscape. And then we navigate that energy landscape later.

• 30:44Dan Shipper?

  Interesting. And like, could it, for example,

• 30:49Dan Shipper?

  explicitly write out for me theoretically explicitly write out for me here all the rules that I know? Or uh, is it?

• 31:00Dan Shipper?

  It it stores all it stores all of them in in this energy landscape, but?

• 31:04Speaker 01

  Yeah, we we can access that. We can access that. And that's what makes it what that's what like IBM potentially makes it um, powerful for data analysis, because data analysis is all about searching for patterns and rules about your data. So and it's it's something where language is not going to be helpful to you. If you try to attach the rules about your data and those data is like numbers and some relationships and functions to like American English and words in American English. And then you try to search for the next word that kind of like you're losing a lot of information. So in this case, you have an opportunity just directly work with the data and understand the data.

• 31:43Dan Shipper?

  I think one of the one of the things I'm trying to understand is when I hear rules about the world and how things relate to each other, I think of symbolic AI. And I'm wonder and and and obviously the those approaches ended up being pretty brittle and requiring too much compute and stuff like that.

• 32:08Dan Shipper?

  And I'm wondering how an energy landscape that acts as that is a that stores a bunch of rules about the world doesn't fall into the same problems.

• 32:20Speaker 01

  Um, well, because I guess we avoid tokenization in this case, we just map it directly into different data structure. So C E B Ms are naturally non auto aggressive. Like there are no sequences of tokens and that's what makes it fundamentally different. So essentially, I don't know if it helps, there could be another analogy, like you're trying to navigate uh the maze and you are L L M person. So you have a L M brain. Um, well, maybe maze is not a good example. Like imagine you're trying to navigate, I don't know, the map of San Francisco. So and you have a brain. So you're like, okay, I'm in Mission Bay.

• 33:02Speaker 01

  Lets me turn to embark there. So you cannot, you cannot choose. So essentially, you you're just forced to choose one direction at the time. So you like choose to walk and back there and you're just gonna keep walking and walking. And you can, if you wanna turn, you just need to choose one direction at a time. And imagine you like trying to get to the Bay Bridge from like, I don't know, King Street. Like, you know, it's typically twenty minutes walk, but depending how you walk. So to navigate there, you're sort of allowed to choose one direction of the time. You don't see any other options. You like have tunnel vision.

• 33:47Speaker 01

  And you just kind of keep walking, walking, one decision at a time. And sometimes you take the wrong turns just because you hallucinate. You know, some words just naturally next to each other and it doesn't allow you to turn right when you want to turn you left. And then you just keep wondering and wondering until you try to reach the Bay Bridge. And the roads you're taking might never take you there. Like there might be a hole in the road and you're just gonna fall, but you and you might see this hole, but you cannot turn back. Because you're auto-regressive L L M. You have to go into that hole and that's like sometimes you run out. So this is the reason why sometimes we prompt it and it doesn't give you an answer.

• 34:31Speaker 01

  It's just because it's searching and searching and searching. It's spending more and more compute and it doesn't have a bird vision. It's just doesn't have ability to turn as it performs the tasks. It doesn't know what's right and what's wrong anymore. It just like randomly chooses one direction at a time and keep walking until you try to read it. So you might never reach that destination.

• 34:53Speaker 01

  And that's why you need a lot of training. So how is it different from the E B M? E B M gonna have the bird view all the time. And you're allowed to take different routes. So if you see there's a hole, you're gonna choose a different route.

• 35:08Speaker 02

  It may not look like it, but Dan Shipper is currently hard at work testing the latest Codex and Opus models.

• 35:17Speaker 02

  Working looks pretty different in this new world.

• 35:22Speaker 02

  We call this Hammock Mode.

• 35:29Speaker 02

  Whoop, Hammock Mode's over. Looks like Dan has to jump in.

• 35:35Speaker 02

  Hammock Mode, an idea by Everi.

• 35:40Speaker 02

  Everi, the only subscription you need to stay at the edge of A I.

• 35:46Dan Shipper?

  That's really interesting. It basically, I've been doing a lot of coding with language models recently to sort of test the limits of vibe coding.

• 35:56Dan Shipper?

  And one of the things that I find with, or have found with big production apps is in particular, if you have vibe coded something,

• 36:09Dan Shipper?

  you, over the course of vibe coding it, you may have slightly changed exactly what is this project even supposed to be about?

• 36:20Dan Shipper?

  And what is the, what are the problems that I'm trying to solve with it? And if you then go look at the code base.

• 36:31Dan Shipper?

  It feels like all of the code is locally correct, but it forms this sort of like patchwork of like hot fixes and solutions, where if you zoomed out, you'd be like, actually, there's a much simp- we should just throw all this out, and there's a much simpler way to think about how to do all this stuff. But it has a trouble- it has trouble when it's presented with a lot of context, then zooming down into, "Okay, I need to create a unified solution here that is not a patchwork of different things, but like carries one concept throughout the entire system." And it gets sort of- it ends up being distracted a lot by whatever it's looking at at the current moment.

• 37:14Dan Shipper?

  Is that the sort of problem that you think this type of system can help with?

• 37:20Speaker 01

  Uh huh. There's actually a lot of problems in what you're describing. So, um, yeah, so solving the problems with vibe coding is one of our use case. We're dreaming about generating formally verified code and automate the coding entirely. So moving you from vibe coding in one specific language to coding in natural language. So you can code in natural English, for example, and no more C++ or Python needs to be involved in there. So that's an idea. And with the coding is like, at the state it is today.

• 37:57Speaker 01

  Yes, we prompt LLMs and it gives us something back, but it's still on you as an engineer to figure out what's right and what's wrong. So there's gonna be a set of rules LLM can try to help you, and even if it has external verifier, which just gonna check whether your old logic in your GitHub space is sort of compatible, compliant to what you're trying to create, and if the new logic is compatible with your old logic. So these things, external verifiers can check. They could just say, "Hey, we know the old logic. We know the new logic. We're gonna see how it's merged together. We're gonna write mathematical proof,

• 38:41Speaker 01

  making sure that, you know, this logic is compatible with what you already have, and provide you a certificate." It's all like, you don't have to review any of it. It's machine verifiable. It's all happening on compiling level. So all it's gonna say is gonna send you a message in natural language like, hey, look. This part of your code is not compatible by logic. This is potentially how you fix it. And this is the things we cannot fix for you. So. We moving you from vibe coding to vibe code specifications. Those rules and. Information about your code is called code specification. So, once you understand.

• 39:24Speaker 01

  Like, this, this is the first problem, right? We're trying to solve. It's just logic can be incompatible with what you already have. The second problem is, is this code actually doing what you want it to be doing? And this is what AI cannot solve for you. Because AI cannot look in your brain and know what you want. Example would be like, imagine you're coding, web coding autopilot. So you have specifications from the hardware perspective. You have specifications from your logic perspective. Like, hey, make it. And there's also instructions, right? How the car is supposed to behave. So there's behavior. Parameters for your code.

• 40:05Speaker 01

  So, code being able to be compiled is one problem. The second problem is, is this code doing what you want it to be doing? So, for example, how fast it is on the hardware and so on. And if the answer is yes. Another set of questions is like, okay. Is it gonna hit a pedestrian by chance? Is it actually gonna navigate the map of, I don't know, San Francisco? And the answer is, I don't know, right? So in here, you need to write a bunch of tests. And, and test your entire system you created, like, oh, is it overall behaves the way it was meant to be?

• 40:46Speaker 01

  And, so this is, this is another form of specifications, right? And essentially, the behavior part, sometimes can, we can guess it, like if we have a lot of data. We could have another L L M or E B and proposing you like, okay, people who try to do the autopilot of the sky, and this is what they're looking at.

• 41:07Speaker 01

  But it you might be doing something absolutely new and we just don't have data about it. So it's gonna be on you to tell the behavior. And this is where the big, the big thing starts for me personally. If you have L L M as a form of A I driving something important, where people trust their lives, like a car or plane or, you know, similar. Um, L L M can misbehave based because you cannot constrain it. It just hallucinates. And E B M can be constrained. You can come up with a set of constraints and E B M just forced to follow it. So it's on you as a human to make sure you know what you want for A I to be doing.

• 41:53Speaker 01

  And then from our end, from the technical point, we make sure A I always obeys the rules by given by human. So, and it can go really far, right? We're talking about the cars and planes, but look back to the language. Sometimes model can say something super sensitive to mental person, like struggling with depression, and it can go really wrong. So even the language can be dangerous. So here we like, what what I also like feel like we're solving is this problem of A I, just sometimes we don't know how it's gonna behave at different environments.

• 42:36Speaker 01

  Uh, but we do know how E B M will behave. Like at least architecture is designed to be constrained. And there are ways formally to force those constraints

• 42:48Speaker 01

  to be compliant.

• 42:50Dan Shipper?

  So it seems like you have a really promising architecture and a model you built, uh, or several models you built. And it's very different from the predominant paradigm right now, where companies are pouring like hundreds and hundreds of billions of dollars into building data centers and training new L L M's and all that kind of stuff. What do you think about the current state of the industry and?

• 43:13Dan Shipper?

  Um, investment in LLM versus other model.

• 43:16Speaker 01

  Opinions about that.

• 43:17Dan Shipper?

  Yeah.

• 43:18Speaker 01

  Oh. Um, it's an ecosystem, right? Silicon Valley especially, it's an ecosystem and there are lots of micro versions of those ecosystems around the world. So LLMs historically the first form of AI which gave us a hot effect. Like two thousand twenty one, two thousand twenty three, when those just, they just started appearing, people like, oh my god, this is the new future. It's amazing. Um, so this is why, like, people start believing that, okay, if it's really good at talking to me, eventually it's gonna be good at doing data analysis, my taxes and other stuff. So

• 44:02Speaker 01

  all the investment communities start pouring money into LLMs. And there were a lot of money to be put in that back then. And right now, people see that, okay, we grow the compute. We trying to change the architecture a little bit and it's sort of reaching out plateau. And there's so much money already put in there. Like, what do you do with this? Um, it's like billions of dollars, literally, you can just like, forget it and like, okay, you know, let's dismiss it, let's pour money into something new, nobody thinks this way. And we don't have probably enough money in this entire economy to like, just make decisions like that, billions of dollars there, billions of dollars there

• 44:46Speaker 01

  for the AI specifically. Um, so this is why it's so hard just for investment communities, just like, take that step, understanding like, okay, this is not working, maybe I invest into something radically new. And I'm not saying people don't do it, like, people do it. Just percentage wise, it's a lot smaller. What people feel comfortable is to take something LLM based, which is changed a little bit, so it has a little bit of elements of novelty, but it's also LLM based. So they can still use the portfolio companies and so on. Um, so they pour money into that. And I understand because like,

• 45:33Speaker 01

  If I were an investor, I would just always look at what variables would give me risks and how can I reuse what I already have. So it's naturally for you to keep investing into L L M's like architecture just because you already invested a lot in the past. You're already committed to this and maybe start investing a little bit into something new. Um, and there's a lot of big tech companies who are a part of this ecosystem, right? So there's a lot of circular deals happening, like those companies who create the L L M's. They create ecosystem for companies who create in data centers. And those who create in data centers, they have dependencies with the hardware industry.

• 46:18Speaker 01

  So it becomes like a one giant thing, which is impossible to break. And when we came with alternative architecture, we're like, okay, let's not just try to put it as something out there radically different, which you have to abandon L L M for. We, we are very much compatible with L L M's, like you could put L L M on top of us. A B M's compatible with transformers, transformers can, you know, work with any L L M's. We can be that layer where still all your L L M investments valued. You wanna make them cheaper, everyone wants. You can outsource the task to us related to spatial reasoning. Like if somebody comes to big tech L L M and say, hey, can you try to do my taxes?

• 47:03Speaker 01

  L L M not gonna solve this, but if it's attached to E B M, we could take care of that. And you can take care of anything language related. So we could actually try experiments to reduce the cost for your L L M portfolio companies and be a part of the ecosystem which is already out there while we creating a new ecosystem on the side for alternative forms of A I.

• 47:26Dan Shipper?

  That's, I think that's really smart. It's a great strategy. I'm really curious about something you said a little earlier that, that progress is plateauing in LLMs. That's, that's news to me. Like, I feel like every month or two, I'm testing a new model, or I'm like, holy shit, this is actually way better. And it does feel like if you look at the top model companies, if you're, you know, talking, talking to OpenAI or Anthropic or Google, they're, they feel like there's a lot of more room in the LLM paradigm. What are, what do you think the, what do you think I'm missing, or the big model companies are missing?

• 48:00Speaker 01

  Um, personally, you know, when I'm saying plateauing, it doesn't mean it's like reaching out flat. It's like you incrementally better and better. But is there gonna be another phase transition, like another breakthrough? I don't anticipate that, just because we already reached so much complexity of those networks using billions parameters, so much compute, so much of frameworks, like creatively parallining this reasoning processes, and it still doesn't transition you. Um, so the reason why I figure out it's not gonna work in the long term for some tasks, like applied engineering, is when I just start speaking to different companies in

• 48:45Speaker 01

  that space. Like, we speak into, um, like, digital assets companies, like banks, trading firms, where a lot of data analysis is needed. Also, drug discoveries, essentially, just people who look in at a bunch of data, not just patients talking, like, language set of data, but also, like, the blood markers, the genes, and so on. Um, so a lot of this is data analysis, which is still done by people today. Um, there are also, like,

• 49:20Speaker 01

  Decision making pipelines, like sometimes you just need to distribute the energy on your energy grid. And you need to know how much energy to pump in your system. So what it means is you need to analyze the data in the short term, in the long term. Construct a prediction, how much data. I mean, how much power you actually need to put into your system next. In the next millisecond or second or an hour. And all of this is still done by people or a combination of people and some programs which are controlled by people. So L L M's are relatively new and A I is like, you know, not relatively new. It's been like few years for us. And all of this mission critical industry is still not automated by A I.

• 50:05Speaker 01

  And even like, I'm just asking like, oh, how much of your data analysis L L M is doing today? And the answer is zero. And I'm like, why? What's the issue? And the issue is the big tech L L M's. They're mainly like B two C, so it works for you, for your coding and for your personal needs sometimes. But for businesses, they don't wanna share their data with them. They don't wanna share data in that big brain for all. They wanna have. Privacy and they wanna have their own custom A I, like custom version of A I specifically designed for their tasks. And this is what like L L M's cannot do for you in the form we have it today.

• 50:50Speaker 01

  So there's no B to B model. There are B to B model for like co-generational tools, right? Do you have enterprise package for the co-gen for, I don't know, businesses, but. It's still done by people. Even coding is still done by people. So it's like, it's interesting to see that there's still a huge gap, especially in applied engineering, data analysis. Anything which requires a layer of verification, like L L M's are not there.

• 51:19Dan Shipper?

  I totally agree with you that there is definitely, there are definitely still a lot of gaps in LLMs. I'm curious, given this and given what you're seeing in the customers you work with, the companies you work with, do you think the big model companies are sensitive to this? Are they working on energy-based models? Are you working with them? Like, if they're not going to get to the next paradigm, do you suspect that they'll start to adopt stuff like this?

• 51:43Speaker 01

  I do know that some big tech LLM models, I mean, the companies have EBM models in-house. So, which is a positive signal for us, right? So, you know, the leaders who were there before we came, they started with LLMs and now if they started building the EBMs, after we start building the EBM, it's a positive signal, right?

• 52:07Dan Shipper?

  Fascinating. Eve, this is an incredible conversation. I feel like I learned a lot. Thank you so much for coming on the show.

• 52:13Speaker 01

  Appreciate you. Thank you, Dan.

• 52:15Dan Shipper?

  Of course. If people are interested in following you or following your company and maybe using some of your products, where can they find you?

• 52:23Speaker 01

  I'm mostly on X. Yeah, we have Logical Intelligence account and my personal account on X. I'm still learning to be more active on social media.

• 52:38Speaker 01

  We also have LinkedIn page, so we're trying to update it.

• 52:41Dan Shipper?

  Cool. Awesome. Well, thanks for joining.

• 52:44Speaker 01

  Thank you so much, Dan. Bye.

• 52:54Speaker 03

  Oh my gosh, folks, you absolutely positively have to smash that like button and subscribe to AI and I. Why? Because this show is the epitome of awesomeness. It's like finding a treasure chest in your backyard, but instead of gold, it's filled with pure, unadulterated knowledge bombs about ChatGPT. Every episode is a roller coaster of emotions, insights, and laughter that will leave you on the edge of your seat craving for more. It's not just a show, it's a journey into the future with Dan Shipper as the captain of the spaceship. So do yourself a favor, hit like, smash subscribe, and strap in for the ride of your life. And now, without any further ado, let me just say, Dan, I'm absolutely hopelessly in love with you.