AI Accountability, Reasoning Standards and the Future of World Models

Hey everyone, happy Wednesday. You know what day it is. You know what day it is, right?

It's... Well, I'm not going to go there. Mike.

Anyway. Mike, Mike, Mike. Yep, Mike.

We got a Mike. But welcome to Techstrong gang. We've got a great Techstrong gang to present to you today with, as usual, a whole bunch of AI stuff and other stuff, and we'll jump right into it.

Let me introduce you to our gang for today. Looks like you got a little haircut there, my friend, JP Morganthal. JP, how you doing?

I'm doing great. And actually, I haven't had one in a while. It's just growing in nicely.

Oh, good. They did a good job. They did.

Joining JP is the woman in front of the red picture in front of the brick wall now, our friend Terry Robinson. Terry, good to see you. Good to see you.

Our man about Spain, our international gang member, the one and only Andy Mann. And joining us two days in a row, it's not Mickey Mouse, is it? No, it's Steamboat Willie.

I have no idea who that is. Steamboat Willie, which was kind of a Mickey, but joining Steamboat is our friend Chris Velask. And then finally, as I mentioned before, Mike, Mike, Mike, Mike, Mike.

Really in a good mood after his Yankees walloped those Sox last night, Ben Rice-a-roni. Mike Vizard. Guys, welcome.

Let's have a great gang today. We got some good stuff to do today. Mike, I slept soundly last night, not just because the Yankees win, but I'm so happy to know that everyone in AI are patriots, and they love the Earth.

And that they are morally bound to do safe AI for us. We can all rest easy. I think we'll be celebrating this day, September 29th, yesterday, every year from now on because- Absolutely ...

it marks the signing of the great Superintelligence Accords. Jason- Today is our Independence Day. But JP, I got to say, when I look back in history, every time someone has signed some great accord, whether it was Neville Chamberlain or the Russians and the Germans, a war broke out.

So, what's your sense of what's going on here with this thing, and how viable is it, and how realistic? Yeah, I'm not liking it at all. I'm sorry.

I am really, really not. You can call it fox in the henhouse, all that, right? All the great analogies about self-policing and let's face it, this is not an area where I wanted self-policing at all.

And I can understand that, in the US, Republican administrations tend to move away from regulation. It's just part of their platform. This is an area, just like airlines, where you want regulation.

Regulation is key. Now, the other part of it is that I didn't like of the story is we had a bunch of great tech executives come and join us and sign this thing. I don't know if anyone has read the book, "The Nerd Reich" by Gil Duran, Silicon Valley Fascism and the War on Democracy, but I recommend it as a good read at least.

You don't have to agree with it, but you should understand the information in it. Our Silicon Valley culture has shifted from a very liberal-oriented, let's save the world, to how much money can I get from these coffers, and how do I keep it for myself? This is that movie with Matt Damon.

The tech execs are the ones in the sky. " I'm sorry. No, you don't.

I don't disagree. The fact that they're morally bound by their duty and honor. What does that even mean?

" But no, what it calls for is each company sets up its own board of directors subcommittee to watch over that company. Yeah, nothing. So now you got a board of directors that's keyed into shareholders, and now that board of directors has to make a decision.

Do we release something that might be a little edgy, or do we delay it and our stock takes a hit? No. Right?

Give me a board with a CEO of General Motors, of AbbVie- No, but you want what we used to call a blue ribbon committee. We don't do blue ribbons anymore. We just- No, because they stand in the way.

Nobody wants-- We can't stand in the way of... Don't let anything get in the way of my greed. I mean progress.

Yeah. And then the other thing that I found, if it wasn't so ridiculous, it was ludicrous, is you got JD Vance and Mike Johnson facepalming for the cameras, just basking in Trump's glow, as if they would do anything otherwise, right? Like Mike Johnson is going to go get Congress to pass some law.

" Oh, man. So that man's trying to remain relevant somehow because the Dems take the House, Mike Johnson just- Goes back to whatever. Yeah, no, I get it ...

right. Goes back to my podunk town or whatever he does. But he hasn't distinguished himself as a speaker in any other way except to carry water for the president.

I had to laugh at the morally bound thing, though. Can I get back to that a minute? Sure.

Because to have moral come out of Donald Trump or anybody's mouth in that group maybe is a little bit of a stretch. Oxymoronic. Oxymoronic, yes.

But what does that mean in terms of, well, somebody breaks the moral code or the moral boundaries, what happens to them? Well, they're no longer an officer and a gentleman. Yeah.

They got nowhere else to go. " Yeah. That's right.

Yeah, they get- And this is not just morally binding, Terry, but also it's almost like a constitution. And we all know how respected and bound to the Constitution the administration is, too. I think it's a very interesting choice of words.

I love the article itself because it did highlight those specific words as quotes, and it's like, wow, they deliberately chose to use those words. And it actually just makes me wonder if that was just a deliberate thumb of the nose. Well, I think it's deliberate on, first of all, I think on the administration's part because they're playing to a base that is very concerned with morals, right?

So now we've got people that are executives who are morally bound. It's the same way I feel about the name change to whatever it is this week. Superintelligence, Supreme Intelligence, I don't even know what it's being called.

That's all optics and theater to appease a certain group and keep them voting in the way that you want to, in my humble opinion, and take the sting out of the word artificial and whatnot. I know everybody has their political opinions, and I really don't want to see this become politicized because it is not a political issue. It is a safety issue.

And I stand at the point of actually putting in the necessary controls to ensure that civilization doesn't get destroyed. And I understand that it's like firearm Firemen understand and respect fire. Electricians understand and respect electricity.

You got the guys... I've seen an electrician go into an outlet with no gloves, no nothing, and touch a wire because they know what they're doing. I do that, I end up halfway across the room, right?

So I want people who respect and understand the technology and what it can do, and know how to utilize it properly, and make sure that it doesn't hurt the people who are consuming it. Sure. And right now we don't have that.

I don't want to be the gloom and doomer. I think it has fantastic properties. I think it is extremely helpful in many regards, and can really improve society in many ways.

It certainly gives us a way to understand so much information so quickly and in ways that the human mind just can't comprehend. That's true. And Andy, I think you make me want to feel...

You make me want to be a better person, because that's a really grown-up take on this. And I look at what it really is for me, and it is meaningless, and it is theater, and it is a reaction against the polling that says that people really do want the government to have teeth- Right ... and take control of AI in ways.

But you make me want to be a better person, and I look at this, and I do think, you know what? At least the heads of the real movers and shakers in the industry are actually getting together, and they are actually agreeing on some basic framework. And for me, the really important point is that that will ideally flow down to the individual practitioners.

You know I love my ops people, right? And my dev people, the people who are actually working on creating new agents, using it to create code, using it to create potentially destructive systems, have some at least backing to go to their immediate boss and go, "No, no, no. " That actually is potentially a very positive thing that's going to come out of this.

So trickle down to them, is that what you're saying? Yeah. I would be more comfortable if I, to somebody's point here, if we had the CEOs of the companies using AI, whether they're Ford or insurance companies or whatever, signing on to this thing, because at least then I would feel like we'd have some actual moral weight behind it from folks that have something at stake, rather than somebody who's more concerned about their stock price, per se.

So maybe there's hope that this is a beginning of a larger conversation, but I think it goes beyond what a few tech bros are saying. I hate to beat the dead horse. We all have talked about the Hugging Face breach in ad nauseam.

However, we need to keep it. The one point about it we need to keep in front of us at all times here is that no human was able to consider this as a potential outcome of this in their planning. And I'm sure they thought about, "Oh, well, we got this locked down on the network.

" But the issue with this is all our fears are based in edge cases, and the problem is, they're day zero, and we don't know about them until they occur. Mm-hmm. Right.

Chris, what do you got? So this morning, I was on a United Nations AI focus group call for an hour and a half with 130 brilliant people, and we're talking about the same thing that we did right before we went live in the green room, right? There's super techie people trying to figure out how to share the screens, have the microphone work without echoing, and so forth.

That's where we are. And my joke in the chat is that the fact that we can't use our own systems well at this point is just the only proof we need that the internet is only moderately mature at this point. And this whole topic, look, I agree with the general sentiment of the panel, but the reality is that we have not finished writing one internet security package.

" And the characteristic that all six of us live this every day, but we talk about it on the show every week, is that there's a lot of popcorn because we told you you needed to take these issues more seriously, but you didn't budget for it, you didn't regulate it. And the flip side of that is that we, as the security experts, haven't really thought it all the way through. JP, your last comment, right?

We're finding things rapidly. " And so there's a lot of learning to be done. Look, to the topic of this block, this moral constitution thing, yeah, I expect that to work as well as everything else that's not working.

But there's more to it than them just being wrong, right? So can I tell you some advice that I get from my fireman friends all the time here in New York, and they will say... " So Very true.

Very true. I had not heard that, and my father-in-law was a fire chief. That's a good one.

I'll have to run that by him. Look, let me tie a bow on this one, and we'll move on to our next segment. This transcends cybersecurity.

Really does. This is an issue, not that cyber isn't important, it's very important, and critical infrastructure is, as the name says, critical infrastructure. But if we don't get this stuff right with the AI, we have only ourselves to blame when something goes terribly, terribly wrong with it.

And this tech bro... I did see a woman. It looked like Lisa Su from AMD was there.

But this tech bro gathering of patting on the back, attaboys, and we're morally bound, sounded like Ashley Wilkes in "Gone With the Wind," and why he was joining the Confederate Army. Oh, that's it. It's not.

What is it from the scene in "The American President"? "They've had their 15 minutes of fame. " All right.

Mike, let's move to the next one. Yeah, because I'm amazed that there are five people who remember "Gone With the Wind," but that'll do. Well, I'm on now.

I'm going to remember it, but... All right. So let's do shift the gear with something that might be, well, a little more hopeful, but we'll see.

There's, LandGrant is a company that's out there, and they and 16 or 17 others have at least put out a call saying we need some standards for reasoning as applied to AI models so that we can understand what these AI agents and models are actually doing, because if you've ever watched them have a conversation or do something, you can see that they're thinking, we just don't have any formats for capturing that and kind of understanding what it is they're doing and analyzing it and thinking it through. So I'm going to start with Andy on this, but really, at the end of the day, is this kind of like the next level of observability that we kind of need to understand these AI agents in a reasoning framework? And if we put these things in place, well, maybe, darn tootin', we can control these AI agents.

What do you think? Yeah. Look, absolutely.

I'm sort of excited for this one, actually. Fair disclosure, I work at an observability company, right? We're doing monitoring and observability and so forth.

Get out of the way, this is really positive stuff. This is an open source initiative, and looking at the way that AI engines are creating their results, not just looking at whether they process quickly enough, but whether they're processing accurately. This is my day-to-day job, is looking at this sort of stuff, and it's actually quite vexing.

You can have a model that processes in the right time, it's got the right lag, latency, you're meeting your SLAs, but it can still come out with a completely false response or something it made up. It could go and use the wrong model. The model can drift.

It could just hallucinate. It could just say what it wants you to hear. There are lots of failure modes in an AI, and especially agentic AI, which are non-observable And so that's what this initiative seems to be to me, is making a lot of these non-observable things observable.

And observability is all about collecting the passive signals, so anything the system wants to tell you, you know about. From logs and metrics and traces and events, met data. And what we're doing right now is trying to figure out how do we get further beyond that?

How do we get active signals? How do we get signals that the system doesn't tell us dynamically, we have to go and ask for? And that's what I see here.

So asking about how is it doing the reasoning, where are the assumptions coming from? What domain is it using? What is the workflow and is it processing properly?

These sorts of signals are not observable, but you can find them out if you ask. This seems really positive. Look, it's actually a part of what is going on in a lot of different locations.

The Linux Foundation, big shout-out to them, they've got now the Agentic AI Foundation. They have an observability and traceability group, a working group working on similar kinds of ideas. So you can always understand why it's doing what it's doing, not just that it does it in time, but that it does it accurately.

This is a great initiative. I'm very buoyant about this one. So the part that I'm less thrilled about is the fact that there are no providers of AI models participating in this thing.

And it occurred to me as I looked at this, I said, well, maybe they have a vested interest in not wanting to have this level of visibility in their reasoning processes. And frankly, JP, I'll toss this to you, but I think that they know that this is an issue, It's going to be an issue, so I'm a little surprised with giving all the saluting of apple pie and motherhood, they're now taking a more aggressive lead on this whole play. So what's your read on what's going on here?

I'm curious as to why you think they would take a more active lead. Because they're patriots, goddammit. It just makes sense really badly.

There's no money to be made in standards. We learned that from Oasis. And so, right?

We went through this early on with SOA and XML and obviously there's a underlying requirement here that, first of all, I see something like around reasoning and the first thing I say is, okay, you can standardize all the reasoning mechanisms you want. Doesn't mean that the LLM is going to reason anyway because it's a generative AI, it's not a reasoning tool. Now, future generations could improve reasoning, that's great.

And we move to AGI, definitely includes reasoning. Our current LLMs are not reasoning tools. They're generative content tools.

They work differently. They work at deciding what you put in means and what the next word should be based on that. So, reasoning comes as almost a weird side effect of how the information in the knowledge base is distributed based on language.

And so all right, good. Go for it, man. Now you know what you got?

You created a way for people to standardize their business processes into models, and now they can ask questions about their business processes. Good on you. That'll help, especially around business process re-engineering.

But it's not going to tell you if Sarah has five apples and her sister has three, how many does Timmy have? It just doesn't do that stuff. Well, is reasoning maybe the wrong word here, though?

Because Andy, I'm listening to what you're saying too, and I was thinking how cool this kind of is as a step from observability to explainability, which is what the industry is really trying to shift toward now too. Because observability's great. I love it.

I'm a huge fan and have recently gotten really interested in it. But you have to get to the explainability part when you're talking about especially AI agents, and I just feel like this might be a step toward that. Yeah.

And we're seeing this in healthcare decision-making. We're seeing companies using AI to make healthcare decisions and going to court having to defend it. And so just from a purely practical point of view, having this level of, I'll call it observability, I love that, Mike.

Having this level of observability into how these decisions are made, whether they're reasoned or not, is a very fair point, JP. But I think that's a really positive thing, again, for the people who have to front to court and say, well, why did you support this healthcare decision? How was it made?

Show me your working. I think that you're exactly right, Terri. Yeah.

It's better than going... Yeah. I'm going to tell you the right- I think I misunderstood it.

When I see reasoning, again, I immediately go to people are trying to get this thing to be intelligent. It's not intelligent. But you're right.

I may have misread it, and when you look at it, certainly then it becomes just a common structure for the way all LLMs to report back. What did you do and how did you come to this conclusion? And that's what people want, right?

I read this and my thought was the road to perdition is lined with the best of intentions, right? What we really want is we want transparency. How the hell is this thing coming up with this answer?

Well, how did we get from here to there? Why should I do that because it said so? I don't know if you ever saw, there was an old "Twilight Zone," I've mentioned this before.

There was an old "Twilight Zone" episode where it was a nuclear holocaust, and now it's post-nuclear apocalypse. Right. And this one town is still alive, and the people have managed to stay alive because every question they have an old man in the cave that they ask the question, and the old man in the cave says, "No, this canned good is probably contaminated.

Don't eat it. Eat this. You can do that.

" Some roving people come by and they want to take all of the good stuff that these people have, and they want to kill the old man in the cave. And they go into the cave, and what do you think that old man in the cave is? It's a computer.

Mm-hmm. And they unplug the computer and then they die. That, too, was on Channel 11 30 years ago.

Yes, it was. " So I'm reminiscing today. But the point here is there's something inherent about what we're seeing with AI that make people feel, I want to see into that black box.

I would feel better if I understood why, how. So let me ask this, though. JP, do we really understand how these AI models work?

And if we don't, how are you supposed to stand in court to defend a decision? Because you can't really definitively say how the thing works. Do we understand how they work?

We understand how they work theoretically, yes. I can tell you about encoders and decoders, and I can tell you about how headers work and how they hunt down different parts of the input and compare and vectorize and say, oh, my job is to look for what the word 'it' means, and this next thread is focusing on what the word 'dog' is in the sentence and what its role is, and that all comes together. Yeah.

And we can describe that. And so now, you take that and you put that on top of a big old database of vectors that are data that was collected from all over the place. Some of it cold, some of it not cold.

Right? And you have no idea what that thing is going to find in the bucket. So that's where we start to run into problems.

If you take that and you stick it on top of your bucket, you're going to be a lot less worried. Hence small language models that sit there and go, it's my model. It's trained on my data.

It doesn't know anything about the outside world. I'm not worried about what it's going to find in there because it's my data. Now you open that up to OpenAI and what it's trained on, and you don't know what's going to come out of the box.

It's surprise. It's Christmas time, man. We open up the present.

What did I get? So you need- Secret Santa We might be able to. So what we look for is, show me how you came to that conclusion.

What sources did you use and which embeddings did you use within the model? And so yeah, that helps a little. I think that could definitely support- But that's a road to hell, JP, because then the next thing is, I understand how the model came up with that based upon how it was trained, but where did it get that data that it trained on?

And we've got- We've been here before, right? So this, yeah, this is the one. So I got this pullover 10 years ago at a Kaspersky summit, right?

And I met a guy there who was the recent ex microcode security person for Intel, when he had to basically go tell the executive suite that we can't anymore. It's too much, too big. It's physically, economically, scientifically, mathematically impossible to check every possible implication of everything in the microcode embedded in chips by that point.

And it's the theory, the explainability, right? The obvious thing in this issue, and I think JP was getting there, is we could make observable data in telecausality model. At a certain point, we got to compress it down to what do we actually need.

In this case, it's much like the microcode issue or in supply chain. You're not literally reading all the dependencies in every open source library. Can't be done.

So we look for the effect of, what are the consequences of using this? And if I go further, I'll use it in the next segment, so I'll shut up. So are you guys really saying that the patron saint of generative AI is Forrest Gump because it's a box of chocolates, and we don't know what the hell we're going to get?

You can't know everything. It certainly can be. One of the things that I'm trying to work out right now is how we can use synthetics and some of the old technologies coming back, synthetics are new, baby.

Trying to do this with known knowns to check the work. We know if I put this in, I'll get that out, and that works to an extent. There are challenges, though, because as you said, JP, sometimes it digs in, and it finds something else in the bucket that you didn't expect.

We actually did a demo of this the other day, where the demo did something we didn't expect, but the system handled it. That was a fun moment. Normally, demos are exactly the opposite.

So yeah, I think there is explicability, there is predictability if you can find the right models and do it within the right boundaries. But ultimately, I don't think we want to do that for everything. There's going to be breakout.

And that's where I think some of these standards, it's the XKCD thing, right? One more standard. It's like induced demand.

One more lane, that'll fix traffic. The one more standard will fix our problem. Look, I think it's a good thing, but you're absolutely right, Mike.

This is not the panacea that we all will ultimately want. Okay. Well, I think we should jump to our next topic.

Alan, what do you think? I think it's about that time. All right.

Here we go. So we have now also seen some merger and acquisition activity picking up in the land of AI, and AMD kind of surprised some folks by buying World Labs, which is a provider of one of these world models that are out there. And so folks are thinking, in some regards, that world models might be a bigger deal than gen AI models eventually because, well, the world models give us the physical world, and they can kind of connect the dots between events and things of that nature.

So Chris, as you look at this, what's your take on this whole merger and acquisition and world models? Because on the other hand, my mind was like, well, if world models are so big, then they're going to be so cool, then how come this company is selling? Well, I'll leave the acquisition and finance side to you and Alan.

But on the topic itself, welcome to my world model. I am, as we speak, physically, electronically building a digital twin of the physical property that I'm on, and we sort of have our campus on is 200 acres. So things are in places, they're in relation to each other, and it matters.

In the systems we build, the relationship literally between the computers, how are they related, what are they doing, and so forth, that matters. And I think as we go down this path, we get to JP, what you were saying a minute ago about, yeah, large language models are word prediction things. As you start looking at world models or whatever you want to call it, you start to get into consequence prediction, right?

Sure, the next word in this sentence for an LLM may be thus and such, unless the sentence is going to be consumed by something that has a consequence. And to do that, you had to have world model development of some sort. Call it what you will.

And Andy, one more standards, I am to blame for being involved with lots. I'm involved in at least four right now. And I can't say enough, this is how the internet got built.

People got together, and they collaborated. But I think even that's coming to a conclusion. And I think world models and what I'm about to say are related, even though it may not sound that way, but I think protocol evolution is how standards get developed more.

It happens out in the world. It gets adopted canonically at various levels, and it's less about eight or hundreds of people getting together once a week for a month. Mm-hmm.

Because it has to happen in relation to everything else and getting down into the details of every field and every protocol that every computer on earth is all going to agree on or AI guardrail, that's like trying to check all the microcode. We're past that point. So I am going to look at it maybe from a business finance point of view, but I'm also going to look at it from a world point of view, no pun intended.

At the end of the day, make fun of artificial intelligence, super intelligence, superior intelligence, I don't give a care what you call it. A lot of the people who really work in AI, who did the seminal work in AI, want or think that we have the capability of achieving super intelligence, AGI, whatever you want to call it, whatever you want to say it, that it exceeds the intelligence of a human being. It exceeds the intelligence of the smartest human beings.

At the same time, so how do you get there? Well, we've done large language modules, basically. That's generative AI and now agentics, and that's been it.

But people are starting to feel that that's only going to take you so far, that if you want to get to this super intelligence, AGI, you need a different mousetrap. You need a world model. The folks behind this particular acquisition was Dr.

Feifei Li, I believe, is one of those people. There's the other guy who's in London now. I wrote a special report on this.

It's up on Techstrong AI. It's about super intelligence and how we get there. But there are a lot of people within the AI world who believe that we need to adopt these world models if we're going to achieve super intelligence, number one.

Number two, we are going to need this for physical AI, for robotics. The idea that a robot knows that when I take these glasses and drop them, they're going to break, or how hard they're going to fall, or what's going to happen. That's a world model, versus just kind of putting words in order next to each other.

And so these world models, if you buy into this, these world models are really, really important. My question for you is, what's AMD going to do with it? Is AMD going to move out of the chip world and into the world model robotics world?

ASICs. AI- I love ASICs. Everything's moving to ASICs in this.

So the question is- Okay, but can you put a world model onto an ASIC? Is that what you're talking? You can put things to accelerate ASIC's ability to process into a world model.

Do I need to own the world model to do that? And one more question on that line. Am I doing that in a way that's open, or am I trying to replicate the NVIDIA stack for world models, and I'm basically going to have- Maybe that's the game Well, you got to remember these are just words.

World model can mean a bunch of things. Think, know, understand, all these words we use in conversations like this, you can either take it to a standards group, you literally have to define them, but a world model is relative. This is what I meant by digital bill of materials back in 2019, and you see this happening in supply chain now.

I'm this company here. I have these dependencies across organizations and institutions. That's my world model.

I need to understand that I need to have systems that know- No, but Chris, I don't think that's what they mean here. I know. I know.

Their AI model is really a visual-- They're a company who specializes in pictures. Yeah, but look, again, trouble with words. The entire world, if you want to centralize this, I think to the commercial point of this, if they're trying to be the world model for the world, that's not how any of this works.

Right? You can- No, I don't think that that's what they're saying. They're saying you will build multiple world models.

I don't think there's one world model to rule them all kind of thing. No, no more than there's one LLM. Right.

But I do have a question for JP. For me, so could I create a world model that tracks the LLM, so now I can use the world model to see what's going on in the LLM? So can I use one type of model to kind of manage the other type of model?

What do you say? Well, I think the concept is yes. I don't think it's the world model that helps you there.

I look at world models as kind of an extension of something we've had early on in the industry called large action models. And large action models are heavily oriented towards automation, like RPA type automation, computer use automation, because what they do is they model how people interact with applications so that when you run a computer use scenario and you say, "Log into this app," it's never seen before. Well, it knows typically in the past, it's seen 5,000 actions of people who have logged into apps in the past, and it can look on the screen and say, "Do I see a username?

Do I see a password? " Right? So, and the world model is really an extension of that large action model, where it's looking at repeated activities and then-- See, the piece with world model that gets tricky is there's some piece that it's never seen before that it's going to compute.

And that's what we don't see in generative AI. We don't have that computational aspect to, I've collected these three pieces of information, and my response is based-- And now I'm going to add in this computational aspect, like the glass is breaking, right? I don't know.

I have watched 500 videos of people dropping their glasses and seeing them break. What I do know is through physics, that glass, I go and retrieve the thickness of the glass and the impact of how far it's falling, and then I can apply some computational physics to determine whether the glass will break or not. We don't do that today in so much as without tools.

I have seen a generative AI build up some information and then go write some code, Python code, and execute the code as a tool to get the answer, which is kind of its way on there, but the world model will actually be able to do that as part of its action. Well, the world model, the physical world model I mentioned at the beginning of this section, so this 200-acre property, we're building a physical world model where things like the Raspberry Pi-driven RC car test unit we have here can drive around the property and do its thing. And and and.

Right? Everything's in relation to everything else. So to the topic of literal physical, visual, generated, managed world models, again, to me, it's the same thing.

It's not about whether it's physical or whether it's relational across supply chains or cybersecurity. It comes back to, I think, our last topic, right? Yes, an LLM by itself, a chatbot by itself will do whatever the hell it does.

However, systemically, in the context of something like a 3D world model or a complex supply chain, it's a different beast. We're dealing with the system and not the engine. To me, this all sounds a lot like digital twins when we were kicking around that idea.

That was the term I really wanted to use for this because that- ... we're really talking about digital twins. For real.

And- But it's digital twins to you guys. The game here is robotics. Right.

Well, and again, I think for architectural purposes, it's the same stuff all the way down. So whether it's Raspberry Pi, remote units crawling around my property, or agents navigating your HR database, it's the same relational architectures. All right.

So is there a new phrase in the world? Can I say things like, now the AI world is your oyster? Come on.

Digital pearl. The AI world model is your oyster. All right, Mike, I can see we've-- I don't even know what the word is, but we're done.

I believe the word was exhausted, but there you go. Yes, we've exhausted this, and we are descending. Let's call a wrap on this version of the Techstrong gang.

Chris, Andy, Terry, and JP, thank you so much for joining Mike and I. We will be back tomorrow. It's Thursday.

Actually, we've got a special announcement tomorrow. Don't miss tomorrow's show. I'm going to leave it right there.

But until then, thanks for joining us. Of course, we're live every Monday to Friday from noon Eastern Time. tv, TechstrongTV YouTube channel, TechstrongTV OTT channel for iOS, Android, Roku, Amazon, and Apple TV.

But until tomorrow, on behalf of the gang, thanks for watching, everyone. Have a great day.