Why Energy-Based Models Could Be the Next Big Shift in AI

0:00Eve Bodnia:My intelligence is not attached to any language in my brain. I think in an abstract way, not every AI has to be LLM-based. And there was a realization to me, it's like not everything is related to language in this world. Like robotics is not attached to language. EBM is one part of the story. EBM attached to the LLM is another part of the story. This is the whole point of energy-based model. You never have to guess the next word. You don't really play in a guessing game anymore. You see your energy landscape, you know where the right answer is.

0:37Welcome, humans, to the latest episode of the Neuron Podcast. I'm Corey Knowles, editor of the Neuron, and we're joined, as always, by Grant Harvey. How are you, Grant? Doing good. Looks like some big news dropped recently here from a company called Logical Intelligence, eh? Yeah. Yeah, that's right. Logical Intelligence. This is some pretty huge news. So Logical Intelligence just announced that Yann LeCun, we're talking the Turing Award winner, former chief AI scientist at Meta, and one of the godfathers of deep learning, just joined as the founding chair of their technical research board. And this company is building something completely different from ChatGPT and Claude.

1:14The founder, Yves Bodnia, has a wild background as well. She's a physicist with a PhD in quantum information and algebraic topology. She's published 22 papers on dark matter and quantum mechanics, and she's saying that her new Kona model represents the first credible science of AGI. So today we're going to break down what energy-based models actually are, why they almost can't hallucinate, where they belong versus language models, and whether this is actually a path to AGI or just a really strong constraint solver. So today we're going to bring her on. Eve Bodnia, welcome to The Neuron. It's great to have you.

1:51Eve Bodnia:Thank you. Cheers. Cheers. Happy to be here. Excellent. Well, let's start with you. That's such an awesome background, thinking of dark matter and quantum physics. How did that background lead you to start Logical Intelligence? Actually, this background is a little bit more complex than it sounds. I think since like I was a kid I was just naturally curious and I was trying to understand how this universe works and I was trying to like pick a field which has like no limits to myself like I felt that I'm not going to be a good doctor or like an engineer because there's a level of things you can master and kind of like I was not good enough to go further and on the theoretical science I felt like I could just go up and up and I was relatively okay with mathematics and theoretical physics and I'm like well maybe I'm just gonna start with physics and just throw myself into it and try to understand how things work what's like the fundamental laws of nature and become a professor and I just made the decision maybe when I was like around 11 years old and And since then, my whole life, I was like trying to optimize for finding the best people, the smartest people around me so I can learn from them and like the best resources available for this.

3:22Eve Bodnia:So I moved a lot with my family and ended up being in the Bay Area. I went to UC Berkeley for my undergrad and I met Professor Daniel McKinsey. Hi, Dan. he was he was like so deep into dark matter but also he was focused on just general understanding how you know symmetries and how it's like how the symmetries work and how it's applied to describe the laws of nature so it was not just dark matter it was mainly like the particle physics which is like one of the most fundamental areas and I was like attracted to mathematical foundations of it and eventually once you expose the different areas you start seeing the patterns and I was like well I kind of like understand a little bit how particle physics works and the same mathematical methodology can be applied to like how brain works well there's some frameworks like not or not every frameworks but some frameworks can be applied how the brain works and once you start questioning how the brain works you're naturally questioning what is intelligence and how it works and i met michael friedman who was back then at google quantum ai and we had collaboration at uc santa barbara during my phd era and he's like i why he just shared like if what are you doing with like brain chip development space we also doing the same in AI space and we start naturally talking about it and I'm like well maybe there is some fundamental laws describing intelligence just from the physics perspective rather than you know traditional computer science techniques and I just went deeper and got an idea of sort of what is energy-based models are but in my own mathematical language and then I spoke with our chief of AI and he's like, well, those ideas already exist in a different form and Yann LeCun is pioneering it.

5:28Eve Bodnia:So I'm like, okay, I want to learn more. So it was like a natural progression of things. It was never me like imagining myself being a tech whatever person in Silicon Valley. I was never imagined myself doing this. I was like, oh, I'm just going to be professor I'm going to be teaching I'm just going to go deeper publish my papers so when the solution came for the architecture my first instinct was like oh I just going to publish a paper and get a tenure somewhere and then I met a friend and he's like oh if you're in academia it's kind of going to be hard for you to move the same speed as AI companies so maybe you should consider start a company and I was eight months pregnant by then and I already had my own home I was like very nested you know and i'm like no no no there's no way no way and then it just like sinks in in your brain and you just i'm like okay let's let's just do it and here we are that is so awesome yeah what a great story that's cool yeah it's cool to see how it kind of evolves naturally and it's like you're following your passion but then you're also following the science and then that leads you to this to where you are now it's cool yeah and you know academia is amazing place like you're just doing science unconditionally and industry i feel like you have to force to make it conditionally because your business has to be profitable and you have to like take into account other variables but on another hand you have a lot of resources from your investors who are also your trusted partners and you like aligned on the vision and that's what makes it powerful and i'm very happy that i made this choice i'm still doing what i was doing it's just like the scale of things is larger yeah for sure could you could you give us the the overhead simple view of what is an energy-based model right and specifically contrasting it to language models as well yeah yeah yeah i feel like it's better for me to like understanding my view on the llms and then i explain that my views on IBMs and you're kind of going to see the difference.

7:36Perfect.

7:37Eve Bodnia:So the LLMs, they're great. They're great AI models for language related tasks. And when I just started playing with the first LLM, it just came out from OpenAI and I opened it. I was like, wow, that looks amazing. If you ask like personal questions, it responds. And, you know, then I try to go deeper. It's like, oh, can you help me with math or like with my research? And it's still pretty good at like providing you some you know resources but then you kind of go in the links and you see like things a little bit off and i'm like well i'm just wondering how and as i was understanding a little bit more i just realized like the way it's done it's just taking all the data and it maps in a language space and then it tries to like predict the next word like in the form of tokens and it's like it naturally hallucinates because like sometimes wars just naturally close to each other in in some languages and also it makes your uh intelligence language dependent like i speak multiple languages like my daughter speaks spanish i speak russian i also speak ukrainian i speak english wow when i that's awesome when i think in general when i think i i don't like my intelligence is not attached to any language in my brain, right?

8:58Eve Bodnia:I think in an abstract way, but yet I have a chance to decode it in any language I speak. But also sometimes I don't have to like speak at all. I can just, you know, move things around and I don't have to speak. So, and there was a realization to me is like, not everything is related to language in this world. Like robotics is not attached to language. If you're trying to control the circuits, like lower the lowest logic level via filmware or hardware you don't need to have any language in there right so yeah language is a big part for us people to communicate with each other and create programs which communicate to us but there's a lot of a lot of just information around us not tied to any language and i'm like well it's it's a great realization and if LLM historically were the first AI models which are tied to the language, people naturally think, oh, AI means LLM.

9:58And I'm like, I just need to understand how I could

10:02Eve Bodnia:just teach people that not every AI has to be LLM-based. And there are models out there which think in an abstract way, just like your brain. And you can have a chance to decode it in different form of action like it can be movement it can be software speaking like your ai model speaks to another software or it can be your ai speaks in different languages so you're supposed to have a choice and each of these choices is not really relying on like an extension you're supposed to attach to yourself like for example sudoku people asking like why sudoku for the ebms like sudoku is an example when you can solve it with your brain you don't have to write a program to do this and you don't have to search for any patterns in any language to solve it the whole purpose is just to show people like hey there's a world around the you is not just language there's a lot more there's like spatial thinking involved and that was like the whole purpose because i realize like many people just don't see the difference and the model we designed is exactly for that it's not it's not thinking it's any language but it has its own vector abstract representation like machine things in machine language let's say it's not a technical term but it thinks it's an own language and then you can have a choice like do you want it to be drawn in a form of image or in a form of video, in a form of language, or you might just want to continue thinking in the same way and talk to another software.

11:40Wow, that's awesome. I've been thinking about this forever in the sense that, you know, not everyone, at a very simple level, not everyone types as their primary form of communication, right? Like an artist draws, a musician will play music, like there's other ways for people to communicate than just language. So it's brilliant that there is another way of doing this and i wasn't sure if you had to tokenize everything in order to get like uh you know like for example like a voice model that we know of as today is a voice model technically tokenized like text still like is it still technically a language

12:21Eve Bodnia:is still language right it's just language in like the audio forms form so it probably makes sense but you don't have to do this like the energy-based model we created it just takes data and it maps it in its own abstract representation right we call it energy landscape and then you kind of see all the scenarios at the same time and the whole science becomes how do you navigate this landscape in the fastest possible way so in this case we don't have any tokens at all like there's no token It's token-free model, but the language can be attached to it. So we have a version of the EBM, which is suitable for robotics, which doesn't have any LLM layer.

13:06Eve Bodnia:In this case, LLM is just like a user interface. Because language for us, it's like I speak to you and my language has no intelligence. It's like my brain has some intelligence, but my language is empty unless it's attached to my brain. So your smart language or just any language, it's a manifestation of your intelligence. And you can mimic it, but you don't have to sometimes. So in this case, the EBM, it has an ability to speak to people through the LLM if you want the language to be out there and LLM just like a user interface. But we also have a version which does not. So this is why I started talking about ecosystem.

13:47Eve Bodnia:You know, people talk about AGI. AGI is a fancy word. And at the beginning of this video, whatever you just said, like, I see the signs of AGI. It's not me. It's the journalists see the signs of the AGI. For me, naturally, I ask, like, what is AGI? Like, can we define what is AGI? And if you start talking... I think that's an important step to pause and define it at some point. Exactly. So, you know, every person you're going to speak to, they're going to have their own definition of AGI. And to me, AGI is some form of general intelligence which can plan, which can adopt. It can have some sort of prediction.

14:31Eve Bodnia:It doesn't have to be precise prediction, but it's a part of the planning, right? So we as humans, as evolving, you have memory in your brain. Your brain has certain parts and certain hierarchy. It communicates to each other. There's short-term memory. There's long-term memory. all of this is doing is helping you optimize for planning and prediction so you can survive and that's what makes like us sort of humans intelligent and it's like it's the same idea here right so there's going to be some evolution of ai if we wanted to interact with the real world you need to have ability to plan and predict and adopt because the world is a very like it can be pretty tough environment.

15:17Eve Bodnia:Like you can have different weather for self-driving cars. This weather can be changing or it can be like manufacturing situations when there's something unpredictable came up and you need to be able to respond to this quickly. And that's what makes it safe, right? How quickly you can respond to this changing environment, how quickly you can adopt your intelligence to it. So to me, if you have this ability, this is maybe your general form of intelligence which is ready to evolve maybe that's my definition of agi i like that okay yeah that's fair so it's basically it's basically just how quickly you can adapt and use the information that you're taking in from all of your senses that would that be accurate or maybe but you also need to be able to preserve the task right yeah right if you just adopt for like no reason there's no point evolution has a very well defined task like hey we live in beans we want to survive right different ais there's going to be some forms of it that's going to try to minimize the resources it's using time for computing the things and also like if you give a task if you are an ai driven self-driving car you want it you want to reach your final destination and you don't want it like randomly dropping you off in the forest just because like oh next word is gonna be here sorry we're gonna go different way yeah i've seen some i've seen some uh self-driving car announcements involving language models and it sort of freaked me out because i was like you really well does it need to go back and forth to the cloud to do this and all this stuff it's a little bit scary yeah yeah it's like this is why we're focusing on building the models for specifically this kind of situations because, you know, the world is going to have AI everywhere.

17:09Eve Bodnia:Like five years from now, AI is going to be everywhere. And we just want to make sure that we as people like safe in this AI driven world and whatever AI we put in our system, it helps us and it actually does what it's meant to be doing. So back to your point on hallucinations, like your brain hallucinates naturally so is mine like all of us we're not precise machines this is why if you want to build a house or like a bridge you go to engineering school right you like you're learning the formal methods to help to to to narrow your thinking and have some measures to check it so this ci is the same way like naturally it's it's it's not uh it's not precise but there are ways to make it self-align and there are ways to make it formally precise.

18:04Wow. Yeah, let's talk about that. So how does that work in terms of the, I guess, like the exact mechanism at a high level for folks who aren't super technical? How does it constrain itself to do that?

18:19Eve Bodnia:So because it's token free model and your input data already mapped in the energy landscape, you can, as an engineer or as a human, you can set up where the constraints are during your training. So you're still going to train it a little bit, right? You're going to show it some complete data set with the answers, or you can show it some sparse data with the answers. And there, what it's going to tell you internally, it's going to map the shape of this energy landscape. So the highest point on this energy landscape is going to be less probable scenarios, and the lowest is going to be highly probable scenarios.

19:01Eve Bodnia:And this is very much matching sort of this theoretical physics modeling, when we always want to minimize the energy. So typically, if you're good at theoretical physics, you're going to write the Lagrangian, which is going to reflect your kinetic and potential energy in your system. And you're going to minimize this Lagrangian and derive equations of motion. And then you're going to make predictions how the model is going to behave. So in this case, it's the same situation. We're going to find the minimal points of this energy landscape, and we're going to oversee this landscape. So you always have this bird view eye and you know you're gonna know exactly where the right answers are and as you train in your model you have ability to self-align it as well because sometimes your energy landscape is going to be a little bit off and just depending on what modeling you're using we're using the model which has correction terms so it can bring you back to the original landscape so and of course you it's it's a cold perturbation theory and technical people understand what i mean but typically if you like have a leading term and then you perturb it a little bit like you still can bring it back to where it was originally and it's a subject or delta like how how strong your perturbations are and you could define the perturbation as a subject of your environment and so on and this is what llms are unfortunately you'll never be able to do so just because the model is different, the architecture is different.

20:36Eve Bodnia:So this is the self-alignment part. The hallucination-free part comes for the tasks where you want the answers to be precise, mathematically precise. Obviously, you cannot formally verify poetry or similar tasks. But if you want to verify your data analysis or you're generating the code and you want to make sure it's correct, here you can attach it to external verifier like lean for and some people using other languages we personally use lean for and then you can formalize the output and have your answer kind of checked on the level of compiler oh that's awesome i know the sudoku uh uh the sudoku test on your website right uh yours is wicked fast compared to all the other ones and so i wonder like is is that so that all of that's happening in like split seconds basically and then it's also verifying it in that same time yeah so this is this is the whole point of energy-based model you never have to guess the next word or you never you don't really play in a guessing game anymore you see your energy landscape you know where the right answer is and you just know where to go right away and this is what's saving your time.

21:53Eve Bodnia:So our models are very small. Like we, we scale in this model from like 20 million parameter to 200 million parameter. And there is a range in between and we run it on like the cheapest H100 GPU. So I was like, yeah, I was like a lot inspired by your brain, right? Like as you're speaking to me right now, if somebody says, Hey, can I create a digital twin of yours? And I'm trying to map like your internal body state, uh, the information you process as you're listening to me, your visuals, it's going to take like enormous amount of GPUs and energy and orchestration, but your brain just naturally can do it like less than 20 Watts.

22:33Eve Bodnia:So if you make the architecture right, well, I'm not comparing people like full disclosure with the machines because we evolved like for many years. So there's some, some very hardcore evolution behind, but here the idea is there if you make it right it's not supposed to take you that many gpus that's yeah i noticed something i noticed that really stood out to me in playing with kona is that to watch it reminds me of like diffusion is is there a similarity or relationship there in that style like watching it it it's that same feel as it's going over and it looks like it's iterating over the same thing like but i understand there's there is a big difference correct um yeah so it can be quite deep discussion um because there is so many like different versions of diffusion models yeah and to me to have a technical discussion like i always want to define things first so we don't we don't go on ambiguous right but there's some similarities with diffusion models obviously and the ideas of energy-based models in general they're not that new they've been out there for like 20 years and some early like science even 40 years ago.

23:51Eve Bodnia:The problem was nobody tried to build energy-based reasoning model. People try to apply like energy-based techniques and modeling to like existing LLMs or image recognition, but to actually design the reasoning part itself, this is where things are really new and we just got lucky that we made it. i yeah that's awesome i love that i appreciate the honesty too i i really feel like this is an exciting direction when i think of you know and it's hard to get out of the lom trap in my mind when thinking of it i've got to say too because i keep thinking of like nodes in agents like the speed is insane uh and as a guy who does a lot of sudoku it's also really impressive it well sudoku is just one of the things it's just we we thought of like what's the simplest uh way to illustrate that there are different tasks which are not based on any language and people like know and love and can get immediately and something which can be tested with the llms because llms we compare it to uh advertised as llm reasoning model so it's meant to be extrapolating knowledge it learns from some game games and then it's supposed to like sort of extrapolate the rules for other games and here we're not even close to this and being being able to extrapolate knowledge is one of the most crucial abilities for natural intelligence right so there's like this is what llm doesn't have so if you take a llm and you teach it to do some math and win IMO and all of this fancy Olympias it's just we have a natural assumption like oh this model is so smart let me give it some code or let me give it some other problems in math and it's going to solve it and reality is not right it's not it's just really good at one thing you're trained for but if you take a child and you force them to like learn some mathematics, they're probably going to be good at mathematical modeling.

26:05Eve Bodnia:They can try theoretical physics, or they can go even further. They can even study the law, which is a logic in language, which I can't do personally, but I have a lot of friends from physics department went to law school in Harvard. That's funny. I didn't realize there was so much crossover there. Yeah, because naturally people are good at extrapolating knowledge across different domains, and that's where the creativity comes from, right? You sometimes get an idea from some other areas which you never dreamed of, and all of a sudden it works. That's fascinating. And, yeah, LLMs, unfortunately, at this stage, they can't do it, and I don't think they will ever be able to do this.

26:45Yeah, that's interesting. What's the next... Where would you say you are right now in terms of your development of Kona? and and like what are your what's your roadmap look like to to what extent you can share i

27:01Eve Bodnia:understand that that might be asking too much um no it's um so the first when we started this company i just had like some theoretical idea right and then i was surrounded by talented engineers who just brought this idea into a form of proof of concept and that was like a few months ago and the natural question was like oh can the proof of concept be the actual like a toy model for the model we have today so the answer was yes but to get there we had to perform like a series of experiments to evaluate like what's right what works what doesn't so when the architecture was fully designed the next step would be oh is it compatible with llms or with transformers in general because it's so fundamentally different we didn't even know like it's possible to do so the first step was to attach transformer and try to scale it a little bit and then kind of shrink it back to the toy model version so we successfully done so and then we're like oh can we like not even have an llm but the most simplest version of something related to llm attached to transformant which is small and we understand.

28:16Eve Bodnia:So we attach that, we also scale it and then scale it back and like, okay, that works. How about we just attach the real LLM to the EBM and see how it is as a user interface. Can it prompt the EBM in the way we want? And the answer was yes. So we again scale it and then test it. Like we have a set of benchmarks, which is related to spatial thinking and hierarchical planning. so we like had baselines for the smallest version of the model then proof of concept then the real version of the model and kind of like compare it back and forth and seems to be working so then we're like okay let's actually try to scale it as much as we can and we performed a bunch of experiments and also we have pretty decent theoretical understanding how it works so we don't see any obstacles um but you know engineering can be tricky so sometimes things work and sometimes things you need to debug so the biggest part for me personally was to how to say it um so the the ebm is not naturally autoregressive because there's no tokens and it's also non-autoregressive so meaning it's overseeing all possible scenarios at the same time.

29:34Eve Bodnia:But when you try to attach it to transformers, transformers are very autoregressive. So you have to take this wild thing and attach to something which is thinking very linearly. One step after another, yeah. Yeah. So you're facing a huge information loss in the middle. And then the same thing when you try to prompt using LLM, the EBM. So there is also a giant reduction of the information on that layer so we were trying to like orchestrate this layer alone which took some time and try to see how it scales so i think that was the biggest difficulty we faced but now the architecture is there it's scalable it's it's already like progressing the way we expected and a little bit even beyond yeah that's where we are that's awesome will it always need the the language model attachment to it as like the interface or is that just like an interim step okay it's just for you know like i said to me a gi is the ecosystem you need to be able to adopt and plan and sometimes you need language because it's for people who might want to speak to you and you as ai i mean and you need to have an ability to speak you need to have an ability to like perform spatial tasks like navigation and so on and here like you don't have to have it but it's nice to have ability to have it for robotics you don't need to have it at all you could just put the ebm to control your like energy grades like how much energy distribute in the town and you can control it like is there's it's going to analyze a giant um chunk of data in real time And this is why the millisecond skill we have is important here because also important for trading.

31:21Eve Bodnia:This is like my soft card use case. I just love this whole idea of like algorithms. And it's like I love game theory personally. And there's a lot of it in trading use case. But yeah, so you don't have to have LLM all the time, but it's nice to have it. well could you do the same thing where you attach like let's say you have this in a robot um could you attach like a vision action model to this in the same way is that how that would work or yeah i don't think we actually need a vision action all you need is like a sensor which takes the visual information so it can be just the cameras which taking like a bunch of pictures for you but it also like a temperature sensors as well um and whatever sensors people use for like self-driving cars or you know self-navigating systems in general and this alone can be all mapped in energy space and it all can be like one space for all because we do have ability to extrapolate knowledge in in abms so that's another part i'm so excited about because now all of a sudden you don't need to have like a separate model just for generating your videos generating your images you could just have it all in one.

32:41Oh, that makes so much sense. You know, we keep thinking about this idea that there is this whole world out there that is eventually going to have to be understood by them, that one of the big differences between like an LOM and us is that we have this experiential knowledge, all of this spatial stuff that they do not have in the same way. Do you see, after saying that, do you see that EBMs are a piece of this, maybe the cognitive hole that becomes AGI at some point? That maybe there are multiple elements to such a thing? I'm sorry if that's a little out there.

33:27Eve Bodnia:Okay. I think it's definitely a step towards something bigger. It's definitely a step and what's exciting, this step is compatible with what we already have. So we don't say like, hey, I'm going to kill the LLM entirely because of EBMs. You actually, like language is important and you know, LLM deserves the place to be. Yeah, so it's going to be that and we'll see how it scales, right? We'll see how it learns, how it adopts. There's going to be a world when we're going to know that EBMs are specifically good at one particular task, but not good at another particular task. And if that's the case, there's going to be some versions of the EBMs like, you know, we don't know of, but people are creative and we're going to come up with something and it's the ecosystem.

34:20Eve Bodnia:Yeah. Right. Same is true of people, I would say. I mean, whether it's people or LLM, it seems like everything has here's what we're here here are areas where we're really good and here's something silly that we're just awful at exactly and this is what makes us as a society right we just like um all together you know function as collective consciousness and collectively contribute to the benefits of all of us and i could see ai to be a part of it and the same the same thing is like ebm is one part of the story right you asking me what's next ebm is one part of the story ebm attached to the LLM is another part of the story.

34:57Eve Bodnia:But then you have a gentic layer on top. We actually do quite intelligent agentic layer. And then you can orchestrate the gentic layers between LLMs, between EBMs, or you can just clone the hybrids of EBMs and LLMs. And then you set up some sort of game theory situation, transitive games, non-transitive games. And this is you have your full evolution of AI ecosystem all of a sudden, because it's going to self-train, it's going to to self-align. It's going to create something we can't even dream of. So that's the exciting part. And I don't know, at what moment do you call it AGI? Do you call it like when you already have agents, I don't know, bringing you solutions for human hypothesis, or it's just ability to control the energy grid and the car?

35:41Eve Bodnia:So what is it? That's a great question. I don't know. The classic is it can generalize, right? So, I mean, at this point, can the energy model generalize can can you give it you're saying you can give it any data in and it can give you an answer it can give you reliable answers usually accurate yeah i mean that sounds like but now we see but you know 10 10 years uh ago people would like if they would know how llms perform right now they would also call it agi so i think the definition of the agi is gonna evolve as well because like we're gonna have a new thing we're gonna see the flaws in the new thing and we're like, oh, no, no, no, this is bulls**t.

36:21Eve Bodnia:Like AGI is something higher. Move the goalposts. Yeah, I could see like textbooks, the course on AI and AGI and the universities, they're going to talk about different areas, what was called as an AGI. All of the different times we've pushed the bar further away on AGI. Exactly, exactly. The encyclopedia of AGI. Exactly. I have so many branching questions on this, but one of the things I wanted to go back and touch on is, let's say you were to have the energy model in the robot example, and you wanted to talk to the robot, would it be a similar thing where, okay, let's say it has sensors, but would it need some sort of language speech model to actually talk to you out of it?

37:03And then I guess the same question for agents, which is, you know, does the agent need some sort of, like, I guess, like... Like for human interaction? Yeah, yeah, yeah. Exactly, yeah. Yeah.

37:16Eve Bodnia:Well, it's always up to people, right, what we need. It's not up to AI model. So the way if somebody would tell me and I were an engineer how to design such robot, I would say, OK, what kind of data we have available? We have probably the language part, which is, you know, people talking to me. And we also have the visual part. We also have sensors measuring the distance, how far you are from things. You probably have like an input from like overall environment around you. so you can adopt your behavior and you're like moving your body around so you can have like a bunch of input information out here and you also like have a purpose so what kind of robot is it is it gonna be i don't know your house cleaner or something else so the purpose is gonna define how much and what kind of interaction you're gonna have so if it's a robot who's taking care of like a sick person and this person just leave themselves and they don't have a caregiver so they they're gonna have a set of routines and they probably should be communicated using language but it's also they should be having ability to like evaluate what's around them is this person in danger should i call nine-on-one you know what i mean it depends on the purpose so this is why it's important to have all of these ingredients in place and be combined in in the way and making sure it's functioning the way it's meant to be.

38:43Eve Bodnia:So LLM alone cannot do this. And it's also like very expensive. And in terms of compute, you have to wait for like a few minutes before it knows is it going left or right. So you need to be able like on the seconds, milliseconds to understand what's around you. It's kind of amazing that humans could do that now. I know, I know. It's like maybe we are an AGI we're looking for and we're just trying to create a version of ourselves but i don't know yeah i'm always joking that one of these days our kids will uh will look back and think you let people drive you in a car like why would you ever do humans um what would what would then be the the scaling paradigm here and perhaps this is more about like how you train one of these things like um is it is it just doing doing the same training run on as many gpus as you can do is it does it work the same way or how to yeah well we're gonna find out i can only speculate how it's gonna be like right now the scaling we doing it's all still one gpu for now um but we haven't like forced the model to be out there in real world enough yet to understand um so for me like when when i was working on it from the theoretical perspective the most important question is, is architecture scalable?