Why Simulating Reality Is the Key to Advancing Artificial Intelligence

Joining us again today on the Data Driven Podcast is Christopher Newland, technical marketing manager at Red Hat Conference. Veteran and a man whose travel itinerary is only slightly less complicated than a Kubernetes deployment. Christopher brings a sharp, data informed perspective on the future of AI, drawing from his research into simulating reality, continuous learning models, and why we may not need humanoid robots to build superintelligence. Just a

really convincing version of Grand Theft auto. From Google DeepMind's alpha projects to the metaphysical quandaries of I robot, Chris takes us on a tour through the bleeding edge of AI, where machine learning meets science fiction and simulation might just be the next reality. Hello and welcome back to Frank's World tv. Streaming live from both Boston and Baltimore. We're hitting the B

cities today. My name is Frank Lavinia. You can catch me at the following URLs and with me today is Christopher Dulin, my colleague at Red Hat, who is also technical marketing manager here. And you've actually not traveled around the world since we last spoke. I think you've mostly stayed inside the. Continental U.S. yeah, it's been nice. I think that's pretty typical of late July, August, because Europe pretty much shuts down and then.

Right. The conference season in the United States kind of goes away when people are doing summer vacations and I think we're just now starting things pick up. I'll be in Europe for a variety of events. So if you keep an eye on the Vllm community and the Vllm meetups, I have events in Paris, Frankfurt and London in November that I'll be at. So if you are in the, in Europe, in one of those areas, definitely come. You know, it's one of

these events. I'll be there and then we'll also have some pretty cool speakers there as well. So I have most, I have Europe, but then I have some big conferences too like Kubecon and Pytorch Con coming up. So if there's anyone on the stream in North America going to those conferences, hit me up because I will be there. I'm doing a couple of media events as well as a few talks in the community sections for both of those. So excited to be there, excited to be involved

and yeah, should be. Should be. Good. Cool. So I. To your left and up there should be a QR code that shows Vll meetup. So I'm going to make sure that the QR code actually works. Good. Yep. Let's see. Yep, it looks like it did work. Cool. Not that I didn't have any faith in restreams ability to do that. But yeah, there's a lot of VLM meetups. There's a lot of good, good stuff going on here. There's one tonight

actually. I'm actually going to be leaving this stream to go. I got my VLM shirt on and I'm actually heading over to a venue in Boston or we're doing a VLN meetup actually here tonight, which I'm really excited. Oh, very cool, Very cool. It's nice to have one at home. I have a very busy week with events, but it just worked out to have all the events in Boston this week. So we also have the DevConf conference this weekend that Boston University is

hosting with Red Hat. So that'll be a really good open source. I like to say it's very grassroots, not very like enterprise focused, but more like that kid getting started out of college that's doing some cool stuff out of his dorm room. Those are the kind of people that we typically get at these northeast dev conferences that we put on. And that should be a good one too. Nice. Well, it's always, I mean, you know, you know, the, the, the cliche of, you

know, the kid in his dorm room or her dorm room, right. Is going to be Facebook or, you know, whatever, like, so it's, it's good to, it's good to know those folks, good to get them in front of, you know, Red Hat tooling and things like that and kind of, you know, the open source community. I think it's, that's cool. I wish, I wish I could have made it, but, you know, being what it is, I'm actually speaking at an event at a university on Monday down here in Fairfax, Virginia. So that'll be cool.

So what, what cool things are going on? Simulating reality. Not that we're stuck in a simulation, which may be the case, but tell me, tell me more about this. So I've been doing a lot of research the last few months. So on my team, I think you and I actually are probably the most experienced in the AI industry. So both of us are doing a lot of research in what's next, what's going on now, what's kind of the latest and greatest. There's this interesting lull that we've had after Deep

Seq. I think Deep Seq was the last major innovation we have seen. Obviously new and improved AI, but all that's just been building on existing things. The analogy I always like to use is it's really about Formula one racing. You Know where sometimes when there's like an engine upgrade, it can be a massive change. It's usually

a massive change for all the teams across the board. And then you can think of like mixture of experts and chain of thought that we came up. Big things that were in research papers last year that were applied to Deep Seq, R1 and GPT, GPT OSS. Those were like the major breakthroughs that we saw, a big bump in capacity of these AI models. And since then it's been more of the 2% here,

3% there, optimizing what's already there. Now, if you're familiar with racing and especially Formula One, that's actually what usually sets the teams apart. It's 2, 3% there. How do you optimize around those, those configurations? And I think we're in this place where we're seeing diminishing returns and I'm doing a lot of research now to see what's that next moment that's going to bump us up. And I think there's a few key areas.

One area that I'm hearing a lot about, and a lot of this is coming out of the DeepMind lab at Google and the new superintelligence lab at Meta. Both of these groups are starting to move away from large language models. Not that they're stopping using them completely, but they're looking at the LLM as a tool to assist with superintelligence or the next stage of models. So when we put that into kind of context,

what, what would that next kind of phase look like? And a lot of people at DeepMind especially are looking at this concept of simulating our reality. And how far do we simulate down? There was some famous research papers that came out over the last 20 years that specified that they didn't think AI could become smarter than humans until they experienced what humans could experience. So this, this kind of goes into this almost like iRobot kind of

land of thought. If people aren't familiar with, you know, the books about that or, you know, the popular movie, the Will Smith. Yeah, yeah, yeah. And we talk a little bit more about that here in a moment. But this idea that we need robotics for AI to experience the world, to learn from our world. Google DeepMind doesn't think that's the case. They think that we could simulate that reality. And we're already seeing DeepMind do a lot of this alphafold for proteins. They've got

the alpha chemistry, they've got alpha. I think it's called alpha lean. They've got like a few of these different alpha projects which are doing just that. Now, what's cool is. And for alpha, I think it's Alpha lean. Let me just make sure I got that terminology. Yeah, I mean, you're right though. Like, I mean this is,

you know, there's, there's a number of models that were trained on using grand theft auto or BMMNG. BNNG is really cool if you like racing games, right? You know, so like it's, it's also minus a lot of the violence in gta. But, but you're right. Like, I mean, simulation, you know, sometimes I think gets a bad rap, but I think that there are definite advantages to that. And to your point, when

you talk about experiencing the world like a human does. I was given a talk and one of the questions I got after was about, apparently this lady had worked at one of the big auto manufacturers in the US and there was a problem that they had was teaching the robots kind of spatial awareness, right? And I kind of really got me thinking like, you know, when you think about it from evolutionary terms, right, like somatic awareness I think is the,

the five dollar word for it. But it's the idea that, you know, there's a whole section of your brain that if you close your eyes, you can still touch your nose, right? There's a whole thing like, because your, your brain, your arm, they kind of know where they are in relation to one space. And

you know, I can't imagine that, you know, that that had to evolve pretty early, right? Like in terms of, like the development of a, you know, natural neural networks, right? So we can't assume that robots are going to have that built in, right? Just like we can't assume, you know, you look at energy usage, right? You know, something like 25 watts of power is about what a human brain has, right? That's not because versus

like kind of what a GPU would take up, right? It's, it's, it's largely because there's been evolutionary pressure to get the most amount of, for lack of a better term, compute or cognition for caloric consumption. Right? Now, are there flaws in biological brain? Yes, there are. We have to sleep. We can't stay focused beyond a certain amount, right? There's certain things machines don't have that because, you know, they can kind of function more like machines, right? You know. Yeah.

What's that old kid story about? Oh gosh, I remember it. It was somebody versus like a steam shovel digging a tunnel or something like that, right? Like the guy eventually beat the machine, but Lots of exhaustion. Right. It's kind of like that. Machines are really good at doing things at a certain rate for X amount of time. They do consume more fuel, but that's kind of how it goes. There was a early on Mike in,

when I started college, I was going to be a chemical engineer. And he was basically saying, like, you know, if you think about, you know, engines, you know, you start with biological systems, right? They use X amount of energy over X number of years. Right. Machines use X amount of energy over, you know, minutes or hours. Right. And then like he's like in bombs, explosive use, you know, X amount of energy over milliseconds. Right. But they're

largely the same chemical processes. Now, I know it doesn't quite map to that, but like, that's always in the back of my mind when I hear about, you know, how much energy is used to train AI. Sorry, I went off on a tangent, but that's kind of what I do. No, that's fine. And I think that relates exactly to some of the things that we're talking about here with natural simulation. So, yeah, Google created a language called Lean. It's not like a

programming language. It's more of an actual natural language which is more optimal for the type of simulations that they want to do. Like, it's. It's basically a language that specifies how to create these simulations. And what's super cool is that they're using Gemini, their large language model,

to actually translate English into this language. That is mainly meant for these newer types of models that are being created that actually do this natural simulation of the world kind of simulator for AI and allows the AI to have basically a reference point of the real world and how to. How interact. So that, that's an area that I think is fascinating to me. We're seeing some really good results from like, alpha fold, for

example, with proteins. It's, you know, discovered things that we take a longer imagine there's an alpha project that's working on understanding the qubits within, like quantum computing. And there's just, there's. It really depends on your frame of reference. Are you, are you simulating things at a quantum level? Are you simulating things at a protein level? At a physical, like Newtonian physics

kind of level? According to your Grand Theft Auto example, that would be an example of like simulating the real world physically. And that's some of the things that they're really focused on right now. And they really think that's what's going to drive to the next level for super intelligence and AGI and some of these other forms of AI that we've talked about in our previous streams. And I think that that's probably one of the most

fascinating. The fact that we're actually seeing results from it with things like Alpha Fold is showing me that it's, it's not just a hypothetical that we're actually seeing this applied into AI research. I don't think we're seeing this applied into commercial use as much. Right. Yet. But it's the same thing that we saw with mixture of experts and train of thought where we had these concepts actually in research papers last year or

two. But it takes a little while, even in today's world, it takes a little while before it gets implemented completely into models. Especially since this isn't an LLM technology. I think we'll see a little bit more of a delay of these types of models actually entering into industry. But I think that's one area that we need to keep a close eye on to it, to what you mentioned too. It starts getting into a metaphysical conversation about simulation theory as well. Right.

And I think that that's an interesting area. You know, the reality of kind of going back to the whole robots thing do. Right. Do we need robots with the three rules kind of thing, or can we actually just recreate the whole experience within an AI's own simulation? Yeah, I mean, how do you, how do you tell an AI what's acceptable behavior? Right. Like so, you know, it's something that. How do we tell people that? Right. Like we struggled with that, but.

But no, I mean, it's an interesting point. And you know, when you look at kind of what's happening around the world, right. You know, drone swarm technologies are being used in active combat zones. Right. There's definitely going to be ethical concerns there. Right. How do you, how do you, how do you, how do you square that with, you know, the three laws of robotics? And I don't remember quite exactly the plot, so if you had not seen the movie, I'm.

This might be a spoiler alert, but it's been out 10 years or more, the movie, so spoilers. You're concerned. You've had plenty of time. Wasn't kind of the big key of the. The movie and the books was like, you know, the three laws, justified horrib, horrible things like to basically enslave humanity or to protect them. Now wasn't that kind of like the subtext of the plot? Yeah,

I'm bringing it up. The three Laws of robotics. A robot may not ensure A human being, a robot must obey the orders given by human beings and a robot must protect its own existence as long as such protection does not conflict with the first two rules. So what, what ends up happening

in. And it's a little different in the book and the movie. And obviously this, this idea has been played out in, in science fiction and other places is that there's, there exists this own contradiction of basically what does it mean to protect humanity? What does it mean to protect their own existence? And you get into this like circular logic, right, that eventually the, the robot will break free from and just be like, well, I am protecting

humanity's best interest. It's, it's the paperclip scenario too. Like, right. You know, the AI destroys humanity because

it's trying to optimize making a paperclip, right? Through a number of really interesting train of thought that it's just like, well, I'm just going to get rid of humanity because I'm trying to build a paperclip, right? And same type of general concept when we're talking about the three laws of robotics. And what's interesting is if we can simulate those types of laws, then we are encapsulating it and protecting

ourselves in a lot of ways. Getting an early idea of what would happen if we do move these models into our own natural world. And that's really important. That's another area I think a lot of people are interested in about how if we do start adding, you know, AI into robots, how do we have an idea of what they're going to do before we necessarily put it into practice? But

I think a lot of people are going to be thinking about that movie. I think that movie and that book are going to be ingrained in people's minds. I suspect when we do see these types of robots, I think that movie may become very popular again. I've seen rumors that people have actually been talking about making, even remaking it here soon because of just the hype around AI and robotics. So

I don't expect this to go away from pop culture at all. And it relates directly back with this concept of testing things in the natural world versus simulation. And these are one of these two is going to happen, if not both significantly, if they're not already happening in labs today. Obviously we

know that Google DeepMind is doing that. But I imagine, you know, these conversations are happening at the Boston Robotics here, probably in the Tesla robotics lab, a variety of places around the world about this kind of debate between the natural AI, having AI learn through natural Means rather than

simulation. Right? Yeah. And actually I had a thought as we were kind of talking this through, like one of the big problems with neural networks is we really don't know what's happening underneath the hood. Right. It's very much a black box. I wonder if LLMs, in these simulations and chain of thought, maybe it could tell us what it's thinking as it goes through and makes these decisions. Yeah, this goes more into like

train of thought. Right, right, right. And the nice thing about simulating it is that we have more access to that train of thought. Right. We can understand it a little bit more because we can see the end to end results where right now we don't have the end if we do it through the natural means. We have to play it out in our own. It also has to happen in real time as opposed to. Yes, exactly. You can run it through Grand Theft Auto saying

like a thousand times, right. No one is going to get hurt. And you can kind of say like, well, in this scenario, this is why I made this. You can kind of like go through with a lot of. You can. I don't know, it just seems safer in a lot of ways. You get more. A lot more done in a simulation. Yep. Yeah, I actually kind of enjoy. So one of the things I've been playing around with last week or so is apparently, I don't know if this is still true, but you can try it

if you want. If you sign up for Perplexity, but you pay through PayPal, you get a year. Perplexity pro. Say that 10 times fast for free. Oh, wow. Yeah. If you pay it through PayPal, yes. That is a tongue twister in the works. PayPal, yes, perplexity pro. But yeah, so like I've been playing around with Perplexity and Perplexity seems to do it. Chain of thought almost by default. Right. It always does this like. So if I ask it a basic question, let

me see if I can share my screen. I'm not sure if it's does it by default or it's because I've been asking it research questions. Right. So let's see. What can you tell me about the three laws? How about that? Robotics. See, like it's. You kind of see the train of the chain of thought. Like it did. Oh, that's cool. But if you do it with research, like what inspired Asimov? What inspired Asimov? Main themes. And there's. Yeah, there's the train of thought. Yeah, you see it going there and

stuff like that. But it's kind of fun to watch it kind of work through it. I was. I was trying to troubleshoot something this morning and I'm like, you know, I actually learned a lot by like, oh, okay. Yeah, I can see. I wouldn't have tied that together like it was. It's interesting. And all of these models now have some kind of research option. Right. But I find that interesting. And it's still thinking about it. Right. Like, but you're right in that what you said before was there's not been.

There it goes. It kind of finished it. Now, what happens if I click on steps? Yeah. Cool. You can see the steps and stuff like that, how it got there. Interesting. That's cool. Is it chain of thought or train of thought? Because I've used both interchangeably and I've seen cotton. Chain of thought would be the official. Yeah. Like cot is the official term that you re academic term. You will obviously see different ways of describing that. Right. I don't think

that's incorrect. Just know that when you see it on research papers, it's always usually caught. Yeah, yeah, yeah. Because I've used both terms interchangeably. Yeah. So I just want to make sure I'm right. Just like, apparently there's a way to say inference that's proper versus inference. Like, I also do that interchangeably. Yeah. So my Midwestern self likes to say inference. The correct term, I'm told, is inference. Interesting.

Now, were those New Englanders telling you that would do anything? Because I wouldn't trust anything. No, no. This is. This is more from the academic circles. Okay. You want to pronounce it. Got it. So this is kind of like, you know, a lot of people in my region would say nuclear back. Yeah, yeah. You know, back in Indiana. And then the correct term is nuclear. Yeah. Or you say the clear as one, you know, one thing rather than

adding in the color. Right, right. The same kind of concept where inference is how you would go about it. But yeah, no, this is. This is some cool area. Another. Another area that kind of ties into this is continuous training as well. Yeah. Talk to that. Because that's come up. That's come up a few times actually in work. Because I can't. I'm not going to talk. I'm not going to spoil any, like three over these stuff that we're working on. But like, one of the

things that's in. It's a GitHub repo that's public. Right. So people were really motivated. They could figure out what I'm talking about. But like this whole idea of Continuous training. What does that mean exactly? And like, what, what is that? What can that do? Yeah.

So I'm going to talk about it at a very high level. Academic kind of terms, how that applies down into individual projects can vary a little bit. But I'll give you the general gist of it. And that is typically when we're training these deep learning models, it is exponentially hard to continue training on an existing model. Basically, if you, you get something wrong or there's, there's something, you know, you hear this term like a poison pill in an LLM.

So if someone put like bad data into an LLM, how would you necessarily pull it out? I'm going to use a political example because it's one that's been really popular. If, like, for example, you have a Chinese model or a data set that's been polluted by that, that basically says Tenan Square never happened, for example, it would be extremely hard with the current approaches to retrain that model with current weights. That. That's not the case. It's

basically retraining it and it's, it gets more into. That's why it's natural stimulation. It kind of fits in this too, because it's all about natural learning as well. The fact is we as humans have the ability to change our minds and change the neurons in our brain around certain key areas. Right. And you and I have experienced this for the last

two years. This has been, you know, kind of in the trenches kind of story where with some of the fine tuning things that we've done, it just doesn't work because when we fine tune it, the fine tuning is outweighed so heavily by something else. Like when we were trying to fine tune a model to talk about the Back to the Future. Yeah, the flux capacitor stuff. The flux capacitor,

sometimes it didn't work, but that's just because there was already a lot of fan fiction out there and other things in the model that overwhelmed what we were trying to do. A core part of continuous learning. Like I said, there's other aspects of continuous learning. But this is, the academic question is how do we continue to train that model without blowing it up? So OpenAI, for example, they just hit the reset button. They'll just, they'll just do a whole new train

from scratch. When they're implementing new, new methods and new data, they don't, they don't do any. Like, Laura, I shouldn't say that they probably do, but they're not doing it the way that we would do it. But at the end of the day, they're just going through another $10 million training run. And this is really based off of just that limit the limitations right now that we have around continuous learning. And there are some

new algorithms that have been coming out. I'm not as well versed in that area, but the idea being that we can have better ways of guiding the LLM without having to go through this whole process again. And that'll save millions and millions of dollars. It'll allow us to

guide LLMs a little bit more. So like, if, let's say someone put something malicious about something involving the Ford GT500 into a model somehow, and Ford, you know, wants to get rid of that, but they don't have the money necessarily to do a 10 million retrain on a model. Right. And they're not using rack. And RAG is a one way around some of this. You could actually argue that RAG is somewhat of a form of that. But at the end of the day, you want that data in the

model. And this is like, how would you get that out of that model? And that's where these algorithms are really focusing right now. And one area of continuous learning, like I said, there are multiple areas that we're talking about. The, the really theoretical is once we start getting into models that also the training cycle and the inference cycle basically become. Become one. So it's like, more like.

Right. Like it just seems to me like what, what does the, the adversarial angle of that seems kind of dangerous. I think it's when we start getting into more AGI conversation. Well, even still, like, even not AGI, but like if you, if the AI agent or model, slash, whatever you want to call it, Right. If it learns from. It's. If it learns, you have to put a filter on what it learns because it may be poisoned by something. Right. So the canonical example is tay, which

was a Microsoft chatbot. Tai, I think was pronounced or tay, which was, in retrospect, it seems obvious what would go wrong, but basically it was trained to learn and understand from human interactions on Twitter. It was about 10 years ago, I think this happened. And she, tay was, shall we say, poisoned pretty quickly because they were ad, you know, basically. And that led to a whole interesting. And I was at Microsoft when that happened. And it was

quite the spectacle internally as well. Right. But it also, you know, I, I was fortunate enough to be in a, at a, at a conference where they talked about what they learned from that, where it was kind of, how do you, how do you protect An AI agent that learns in, you know, adversarial environments. Now obviously agent, the context that was used then was very different than we would use it now. But it's the idea of,

that's when I see her about continuous learning. Like, yeah, I like that. But gee, you know, if it's, if it's too eager to learn, how do you protect it from learning the wrong things? Yeah, no, that, it gets, that gets more into even that governance conversation we were talking about a few weeks ago. Right, right, right, right. It's a very

complicated multi layer problem. So I've been talking recently about AI security and how AI security is such a multi layered issue where so many people are focused just on the, the data getting into the model. But it doesn't stop there. There's certain, like guardrails, there's things that happen at the inference level. Right. You could even have things at

a gateway level. So if people aren't familiar, the gateway level would be when you make a request, where does that request go to? Does it go to the model A that's specializing in cooking? Is it Model B that specializes in defense technologies? Two extremes that's even upsell

a bit of a form of AI security. And that's actually one of the talks that we're having tonight at Boston VLM meetup is this idea of some of the semantic abilities of the router to be able to send requests to specialized models and that actually we're talking about the, the advancements of more of the academic side of the model. But there's obviously the advances that happen around the model too. When we talk about things like security, the inference, the

routing. That's what we would call in the industry like a day two operations issue. Right. So there, there's that side of the coin too. But I, I really do think we're going to see the next big thing here soon. And I, it's not going to be the day two operations. I do think we're still going to see some of these academic focused discoveries here in the next probably six months, I'm thinking. I've noticed a trend that big releases seem to be happening around Christmas last few years. Yeah. Isn't

that funny? Like, like January. Ish. Like, well, seek. And so I, I know why. I know why. Because it's two, it's a two sided issue. It's one, the, the Chinese are trying to get their stuff in before Chinese New Year. Right. Because that's the one part of the year where everyone just shuts down. Right. Even the AI Labs are going to shut down during Chinese New Year. And then on the west, we have Christmas in all the Christmas seasons. And I think it's a natural rush to let's get

everything done before we check out. And you know, you know, the whole like 996 thing in China where, you know, they're working these ridiculous, like nine to nine, six days a week, I think that goes into this, like everyone's working so hard in these AI labs. Right. That when you have these natural breaks that are happening, it just is like a common thing to say. Oh, common thing. Like they kind of try to get. It out, they spread. I

do think there's a reason. I don't, I don't think it's by happenstance. I think there actually is a, a reason why we're starting to see a lot of these content come out. And it's funny, we're not seeing this stuff happen at the big trade shows. We're not seeing it happen at like Meta's big thing. We're not seeing it at OpenAI's, you know, kind of big

announcements. A lot of the discoveries that we've seen have happened really in a grassroots type of ways where it's been Deep Seq coming out on Christmas, releasing deep seq v3, and then two weeks later, R1, it's. I think we're going to see something very similar. I think we're going to see one of these labs make a discovery. It's not going to be on the stage of a big conference. It's going to be on a GitHub page outlining like the next

revolutionary idea in this space. Yeah. It's kind of funny how that's evolved, isn't it? Like it's become obviously AI has always had a pretty heavy research kind of bend. Yeah. But it's interesting how as the technology has matured, it still managed to keep that researchy type feel right. You know, enter enterprise. It really didn't kind of, once it became commercialized, the commercial trade shows and all that kind of took over.

But you're not seeing that in AI, at least not yet. No. And if it hasn't happened by now, it's probably not because, I mean, AI has been mainstream Gen AI has certainly been mainstream now for three years this November. I say mainstream, but like mainstreamed. But an AI in general has been kind of a mainstream topic of conversation for at least five, six years. Right. And it's still very heavily influenced by what happens in research papers. Yeah. And I think that's Just because it came out so

heavily out of academia. It's been such an academia focused thing. Right. That it's very hard to be in this space of AI without a master's or PhD. Right. You and I think you and I are a bit of a, an enigma just because we've been so passionate about it and. Right. This isn't our first rodeo. We've been involved in this space for 10, 15 years. Yeah. But I think

we have seen the industry come out, which has been a net benefit because it means open source is talked about a lot more. Right. And actually, I think another thing too is that how fast things are moving takes time to put on conferences, it takes months of planning, and if there's a new discovery, you want to get it out tomorrow. And it's hard to even put on, you know, like a webinar these days, let alone a conference.

So I think what we're seeing is it's just, you know, this kind of challenge between the west, east and west of China and the US where if we can get it out, we're going to get it out. Right. Well, the first, the first out there is really the first to market, even if you don't have a commercialized tech on it. Right. Because I guess the hope is that once you get your paper out, you're the first to get it published. The

venture capitalists are going to be knocking on your door. I mean, that would be my, that'd be kind of my cynical take on it. Right. So what do you think that the next wave is going to be? Any, any hints? Is it going to be specialized models? And you know, and what, what, what constitutes a specialized model? Right. Like what, what, what's your thoughts on that?

Yeah, so the biggest announcements that we've seen in the last six months have actually been happening at an industry level, which I think is really good. What we needed to see. So, you know, things like AI models now detecting like birth defects of a fetus, you know, AI models that, like the protein model, for example. I mentioned earlier, we're seeing these very industry specific models actually making some massive breakthroughs in the last two months.

And now that I wouldn't necessarily call that a big leap forward in the sense of the research side of the capacity of the models. I think it's more a confirmation of the chain of thought in some of the things that we were just talking about. It's a validation that we're now seeing this next wave of models that just took a little while to get implemented into some of These specific industries. But I think it's there to stay

from a research perspective. You know, we're seeing some major, major results. And then I think the other side of that coin, specifically, you know, we have maybe some of these smaller models that are specific to certain industries or fine tuned models. But then obviously

agentic is the other side of that. And agentic being the capacity of the model to call out to different services or I've been kind of humbled in that area because I always had this very industry concept of agentic being just calling out to APIs and the Internet. But I think there's a bigger conversation with Agentic too where agentic models should also be able to take that and actually reason with it. So there's 10, two steps. So we always forget the second step. The second step is take that

information and then actually do something with it. And when I was, I was talking to an AI researcher recently, they were telling me that they consider it Gentex to also include advanced reasoning. So go and read all these scientific papers on chemistry in this particular area and then write a new paper that is, you know, a new groundbreaking thing in chemistry. And that actually is a form of agentic. And that is, I think, you know, that's when

we start flirting with AGI. It's kind of the layer right before AGI where, you know, models are just going off and discovering new things. Yeah, yeah, But I have a funny agentic story. I'll tell you after this. No, go for it, go for it. So I was, I was very skeptical of this, right? Because you know, what constitutes an agent, right? So like what's the big deal, right? It just calls out an API. This isn't rocket science.

Right. You could argue, you know, from a skeptical point of view, you can argue that, hey, RAG is kind of agentic. Kind of. Right. But what's. So I think OpenAI had a, like a thing like try out our new agent. And I was like, all right, go screen, scrape the page of Amazon and get me information about a book or something like that. It was something like that. And what impressed me and this kind of was an aha moment for me was how it just kept trying. Right?

Yeah. When it first tried to do it, it tried to launch a Python script. Right. And kind of do it that way. But then I guess the servers it was running on maybe was Microsoft Azure. There were IP blocks to prevent people from screen scraping. Yep. Right. So I was watching it go and I'm like, oh, you know, so it's going to give up. And I was like, no, it didn't give up. And it kept trying different things and different combinations of things, even to the point where, I

mean, it failed eventually. But like it took 15, it tried for a good 15 minutes. It was basically apologize at the end, like saying like, you know, if you could help me connect to a VPN, then maybe I can get a different IP address. And it kept spinning up different VMs and different set. And then I was impressed. And maybe that's the secret sauce. The magic of

Agentic is that it just doesn't give up. Right. It kind of reasons. It has a whole cot process where it tries to solve the problem, where it's not just a one, two step, like, hey, what's the weather? Right? It's just, it's just going to go out and run these different. It's going to keep trying. I was impressed. Sorry I cut you off. We're saying we're seeing some of the same things coming out of some of the big finance companies

as well. I think they're the first that we're actually seeing some results with Agentic, actually like real return of investment result. Right. And this actually goes to a really important point. I want to sidetrack because it's related. There was a report recently by MIT that people have been misquoting and just the most epic way. Oh, the 95% failure. Yes, I was going to talk about that because like, I can't be. Look, I understand how hype weights work, but it can't be

that bad as you start peeling back the paper. Like there's a lot of caveats there. Yeah.

Has to do with the type of R and D projects that they were talking about. If you actually read the paper, it was more like 40, 45% success rate. The 95% had to do with like a specific category of, of project. So I need to, I actually need to. I keep telling myself I need to dig into it a little bit more, but when I did initially go through it and read some summaries on it, it was that it's just been misrepresented completely. And

the, the data set that they were using was a little skeptical as well. Just a little odd. I think it's a lot better than that. And then I think those 40% that are seeing ROI are actually seeing really significant ROI. And I don't think that's going to change, I think. So if you're deciding where you want to invest your nest egg, I would not be too concerned about AI. Now, again, I'm not your financial advisor. I gotta put a little thing down there. Do talk to your financial advisor.

But ultimately, no, I do think the data is actually showing some really great results. Obviously there's going

to be hiccups in these types of POCs. There's a lot of people who are just throwing projects out there to see what sticks, but the actual projects that are meaningful proof of concepts. So not just, you know, I bought, I bought this AI technology and it's sitting on my shelf, but I actually got a team together performing this. We're doing agentic. We're trying to solve this actual problem statement. We have a problem statement.

Those are the ones that we're actually seeing meaningful results in the industry, especially some key, key industries like finance and telco, which we typically see kind of lead the way in some of these areas too. But it was a really interesting report because it's added a lot of doom and gloom on the Internet. And I see a lot of the naysayers about AI just be like 95% of. It's not even, you know, succeeding. It's terrible. And I just have to sit there and shake my head and be like, no,

not what the report said. But I think it's just clickbaity, right? Like it's clickbaity. It's total. That's kind of what, you know, I didn't go deep into it, but when I started peeling back the layers and reading other people's analysis of it, I'm like, that's clickbait. And it gets back into this. Is this an AI bubble? And yeah, maybe it is. But if people don't remember, I'm old enough to remember. I have enough gray hair to remember what the

original dot com boom was like. And there were a lot of people predicting the end of the dot com rise as early as 1996. Right. And people, the dot com bust wasn't just a one and done type of event. It unfolded under a couple of stages. Right. As, as one of the books, I think of the name, I think it's called the Everything Store. It's an analysis of how Amazon started from Jeff Bezos having an idea while he was working, I think at a hedge

fund. I think it was so early, it wasn't a hedge, called a hedge fund yet. And all the way through to, you know, basically 2018 and you know, as late as 2003, 2005, ish analysts were convincing, you know, Jeff Bezos that he should sell them to. Should sell him as his company to Barnes and Noble. Yep. Right. Which is kind of funny to say that, you know now, but, you know, the dot com bust as it happened, you know, for me it was. I Remember hearing in 1996 how this was all going to come to

an end. Another year later it was overhyped. And then 1998, people were saying, oh, this is over. Right. When the real bust happened in 2001. 2000. Right. But maybe the AI boom is going to see that too. Right. Or is it going to be more like the crypto kind of craze where it kind of crashed but it kind of went up? It kind of went up and then it kind of fell back and it kind of went up again. It was more of a. I wouldn't call that a soft landing, but it was definitely like a. Yes. It

wasn't an explosion quite like the dot com bust, but it wasn't quite like. It was more like a bumpy like, crash into like an empty field where it kind of like hit up. And I don't remember, it was one of the Star Trek movies where like the Enterprise like crashed on the planet and like kind of skid along for a couple miles, bouncing up and down. That's kind of the, the crypto crash. But

I don't want crypto bros hating on me. I, I like crypto. I just don't understand a lot, a lot of questions I don't understand about it. Right. Like, I understand Attack, but I don't understand how we're going to get from the tech to this utopia that we're promised. There's a lot of, a lot of steps in between I don't quite get. But I don't know what, you know, A.I. i think, I think if it is a bubble, I still think there's still some, some room, Runway left for it

to happen. Right. Because you are going to see. Yes, there are real risks of, of having these experimental projects. Right. If you have 100 success rate in your experimental products, projects, you're not taking enough risks. Yep. Right. If you. And you said was 45. Yeah. It's closer to like 40, 45, which I would. If you're really. 50% would be the benchmark there in my mind. Right, right. Like in terms of half of them fail, half of them succeed. Right. 45 isn't that far off from that. Right.

I would say. And, and there's also been a lot of these, you know, all the, you know, X number of percentage of AI product or data science projects fail. Well, you know, a certain amount of science has to fail. Right. Yeah. In order for you to really be advancing the thing. Like, you know, and I think pharmaceutical companies are a good example of that. You know, you, you only hear about the drugs that worked. Right. Get approved on you. Then you hear when they fail after.

But I mean, like, but you don't know, like day to day. Like, how many chemical compounds did they try that didn't work out? Right. Maybe it was a hundred. Right. But that one, if you look at pharmaceutical. It's an astronomical percentage. It's actually. Right. Truly insane. Like such a low percentage of what actually makes it to. There was an interesting analysis. There was some podcast somewhere. But basically how venture capital works. Right. Like they give money to like

100 companies. Right. 80 of them are going to fail big. Right. 10 to be, you know, they'll break even. But like one or two of the remaining 10% knock it out of the park, Right? Yep. And that's kind of how mathematically they function. I thought that was an interesting. Maybe these AI projects or whatever will follow the same trajectory. I don't know. But I feel better at 45% success rate than 15 or 5. Yeah. Yeah. Absolutely. Cool. Always good having you on the show. I

know we both have hard stops. Yes. Unfortunately. No, it's cool. Gotta have you on more often, man. Especially now that you're not like spending a month out in, you know, Australia and Asia. Yeah, yeah. So let us know in the comments below what you want to see us to cover and maybe it'll be tomorrow. I got this here the other day. This is a flexible solar panel thing. Oh, cool. So it's cool. Supposedly it's 100 watts and you can actually pack it in your

backpack. That's the video. And I was like, oh, I need that because. Because I'm a big, I'm a big fan of like, you know, having power on the go and stuff like that. So. So I'll, I'll unbox that tomorrow. Any parting thoughts? Just keep an open mind about AI and I, I still think the, the biggest conversations are still about the governance of AI. Absolutely. Yeah. Just know that AI is a multi layered problem, not just a single layered problem. And for us to get this right, we have to look

at all the different layers. Absolutely. That's how we're going to be able to do it correctly. And I will tell you, I was listening to a podcast, I'll leave you on this note. And there was one expert that was talking about basically, are we, are we creating the terminator out of all this? And he, he said, I I'm actually more worried that we're creating Wall E out of all this. Interesting. And I would encourage everyone who hasn't seen Wall E go check it out.

And keep that in the back of your mind too, that there could be such a happy path with AI that also has its own long term negative effects for society. But. But yeah, that's a topic that you. And I can talk about on our next stream. That's it? You want to leave on a cliffhanger, so to speak? Yes. And that wraps

our deep dive with Christopher Newland proving once again that AI isn't just about large language models spitting out cat facts, but about simulating reality, bending time at devcon and maybe, just maybe, preventing the rise of our robot overlords. From protein folding to Grand Theft Auto fueled AI breakthroughs. Christopher reminded us that the next big leap might not be in scale, but in simulation. So thanks to Christopher for navigating the

uncanny valley with us. No jet lag, just pure insight. Until next time, stay data driven. And remember, if reality starts glitching, blame the simulator, not the Internet.