AI Safety Rules, Data-Scraping Filings, and What Really Blocks Enterprise AI

I'm off a few days, and I forget how the whole thing works. Hey, everyone. Welcome to Techstrong gang.

It's Tuesday, and we are glad to have you here with us. We're glad to have a great gang. We've got a lot of stuff going on here.

I could hear the referee whistles going as you started that. Yes, they were. The red lights were blaring.

But hey, you know what? It's just Tuesday, just another Tuesday. Let me introduce you to our gang today.

We've got our friend Chris Blask. Chris, how are you? We've got Kate.

I don't know if she's just sans glasses or went to contacts or what, but a whole- They're here ... new look for Kate. Oh, there we go.

That's the Kate Scarsella I know. Hi, Kate. And joining Kate and Chris, we've got Sid Nag.

Hello, hello. Hello. Jack Gold.

Jack, always a pleasure. And of course, Mike Vizard. I will give you a tell.

It's one of those things that most of my family members know, but if I take off my glasses, it means I stop listening to you. Is that what it is? Is your hearing- You're looking for the pen ...

gone again or what? Yeah. As if he was listening to start.

Well, I've got to tell you the truth. There are times these last couple of weeks where I feel like just taking off my glasses and stop listening. There's just so much AI stuff flying, and it's not AI slop I'm talking about, mind you.

Just everything going on with this AI. I read an interesting article today that the economy just passed a very serious inflection point. We're actually spending more money on AI infrastructure right now than we are on the housing infrastructure.

And housing, if you remember it, well, that was the downturn of 2008, right? 2009. Housing was the engine that drove the economy, right?

Housing starts and housing construction and so forth. Well, AI has displaced it now. And- It's crazy.

Well- Well, it's still housing, Alan. It's just housing for chips instead of housing for people. Yeah.

That's one way of looking at it. But we do need to jump into the topic at hand, which is related, but everybody is talking about AI safety, and we've been talking about it on this show as an ongoing series of things, and now OpenAI is proposing some sort of safety standard around the notion of fully autonomous self-improvement and/or recursive self-improvement, RSI. And I guess what they're trying to say is that, well, that's the thing that's going to lead to the downfall of humanity, apparently, and that this is the thing that we need some standards for and some international agreements with.

And of course, everybody and his brother who's involved in this conversation seems to be in New York for the AI meetings this week that are happening at the UN, among other things. I guess they're trying to find a new Secretary General, but that's a secondary consideration, apparently. But Chris, what did you read on what's going on here?

Because you also have an article talking about the control point. Is that the same as the trust? And I feel like we all don't really have a good handle on exactly what it is we're trying to make safe.

Well, as I find myself saying every week on this show, this is a chance for me to step back once a week. And I'm pretty sure I've mentioned the Standards Council of Canada/ISO 42000 technical specification working group that I've been part of since the spring or whatnot. And it's specifically working on systems of AI systems and a couple other things, but I love that one to help think this through because what you need is the trust point.

As I argue in that article, in the working group where I and our folks have been contributing there. Yeah, that's great. Fine.

How do I know? How can I go back later and look and see, right? You look at this block today, it's the same sort of thing.

Should there be controls and so forth? Great question. There's standards and regulations, but what are we trying to achieve?

And more importantly, how do we not just get the controls, where's the trust? Why would I, by any definition of I, trust entity ABC? What is it that I can see?

In the same way that we build standards and regulations, you shall not this, that, and the other. How about you shall not give me something that I can't verify, right? Look, we're human beings.

We keep talking about what AI can replace and do and whatnot, and no, it's not going to end the world. But we're all on this show right now. I'm on this show right now because I trust Alan, right?

Alan, you and I crossed paths a long, long time ago. You started this stuff, and I benefit by doing your stuff. I like you, so I come on and do things.

That's actually how the world works. We actually need to build this into our systems now that we've built language systems that read all our documents and act on them. So on the- Chris, if only the world were as simple as you and I trusting each other, right?

Well, it is, and I try not to get pitchy here, but look, I joke or I try to, as a metaphor that I live AI governance. That's what we do. But it's not really that.

We build systems based on the idea they need to work this way, right? Just like I said, systems need to have the same mechanisms to trust each other. And I think for this topic, when we're talking about humans and human organizations and institutions, it's the same thing, right?

Now, is that trust is different than control. Yes, there should be controls. Who has them?

How do I know? Where's that? And I think there are models like the ISO group that really show how that can be done.

Sorry for interrupting, but Chris, isn't trust also an issue... Or sorry, let me back it up. Isn't verification also an issue of trust?

Verifying what I'm trusting is critical for us to trust what's actually going on. And part of the problem is that, especially with AI now, the volumes have gotten so huge with all of this stuff coming at us, that how do you verify it? It's surprisingly easy when you start from the bottom up, right?

So verify, again, let's think about trust. If I can't verify, there is no trust, period. We all know it, right?

Right. So controls then say, "Trust me," that's not trust. That's a statement.

I need to know that from my perspective, I mean, the information that I as an individual or I as a company or even quite large organizations, the amount of information we actually care about for ourselves, it's not that much. It's not everything trained into every language model in the world. " And you may choose not to give me that information, or I can make a decision on that.

I can tell you that's where most of the big model providers are. They're not providing the information that I would require to trust them. That's true.

So if I use them, I know that I'm using them without that. And sometimes you can, but you just know what you're doing. Yeah, I think it's beyond the...

I mean, you can put all the controls and guardrails and all the constructs in place, but I'm glad that OpenAI is waking up now and talking about this. I'm glad Dario wrote the paper on basing AI and all that. I think it's a good thing, but I don't know that just building controls and the frameworks is sufficient.

In other words, who else is going to comply with this, right? I think I talk about this in an article on LinkedIn, right, on creating an international artificial intelligence safe proliferation agency, right? It sounds a bit bombastic, but I think at the end of the day, you really need a consortium of participants, willing participants, and there's always going to be rogue participants who will not come to the table, but you almost need that, right?

Because how do we allow increasingly powerful AI to proliferate across the world without allowing its most dangerous capabilities to proliferate without any safeguards? So yeah, safeguards are the controls, but how do you make sure that people are actually coming to the table, agreeing to comply, and actually executing on that, right? So I don't believe the answer is to stop AI from spreading.

Quite the opposite. I mean, AI should reach our critical hospitals, our universities, our scientists, manufacturers, all of that. But again, we need some kind of a ...

gathering of minds. And it reminds me of there is an analog of the nuclear proliferation problem, right? We all came together as humanity and built an agency that would allow that framework where people come to the table and talk about it and agree to do things in the proper way.

And I think that's what's needed. Hold on. Let me jump in here because I know she's waving her hand.

Yeah. Now I found my glasses too, so. Now I'm going to at least listen to myself.

No, but you know what is so bothersome is that we've been doing this for a really long time, and what I mean this is, I mean cybersecurity as a whole. We have in place controls that we don't use now, and we're certainly not using with AI. And it just reminds me, as we're talking, I don't know if anybody was around with organizations when the C-level started using their phones.

And all of a sudden, all their phones came in, and they could do whatever they wanted- Yeah ... and the other folks couldn't. And I just remember how stupid.

Even cybersecurity. We became idiots. We got a phone in our hand, and all of a sudden we were just idiots.

The things that we were accessing and everything else, and not that it was bad, but it was just... And I feel like this is the same story repeated over and over and over and over again. Now here we are with AI.

Same controls, same mitigations, and we are facing the same questions, and yet we don't act. We don't really do anything different, and it's crushing me. I'm just like, this is like Groundhog's Day.

So how will this work, though, at machine speed? " And that's the way most people are, right? Most transactions, most relationships I have, it's not all black and white, right?

So how would I trust and verify, to quote Ronald Reagan, at machine speed when I have all these AI agents bouncing around? Chris? I'll go back to Sid's point, right?

Because this is the kind of evolution we're going through. In the post-war, you go back to the late 1800s, right? The IEC, one of the oldest industry standards organizations that's around today.

And we've had consortiums, right? One, sometimes as many as three for the entire globe and the entire world population. You have consortiums now.

io. I was at the Linux Foundation, and it was Maddie, and Desari, and I. Had one conversation led to that, and the premise is just exactly the same now.

At that time, we were Unisys. We bought piles of Dell computers, turned them into mainframes, sold them to Citibank to run the global financial world, right? " And they would say, "Yes," too.

So that consortium, and thick and thin is the time before we took that into the standards world. It's like, we can just do this. Don't have to have a standard.

We are a consortium. Those are three huge organizations. But now, whoever you are, you're a small company, you buy from whoever you buy from.

It's a finite short list. You have a consortium with them. It's your contracts, your agreements, your regulations.

Working up from there, you may find that you may interface with some mass government corporation, whatnot, and they only provide thus and such, and you just choose not to. Or you just know that working in that jurisdiction, here's what you can do. I don't think there is a UN ISO single global standard.

Again, we did these things to make the internet work, to make the whole post-World War II order work, because that's the best you could do. But the internet specifically was never supposed to have centralized things. It's just evolved that way because to date, we had to have global centralized things.

But everyone is a consortium already. Yeah. I've got to jump in here.

Chris, I understand where you're going with this, but historically, it just doesn't work. To build on what Kate was saying earlier around phones, before that it was the internet, before that it was PCs, before that it was mini computers. And what ends up happening in most organizations, from my historical view at least, is that you have some critical event or events that happen that force organizations to deal with this stuff.

Before that, they just don't deal with it. There are always people, especially execs, C-level suite, that go out and say, "Hey, this is a really cool technology. " And I think the same thing is going to happen with AI.

We're going to have some disaster somewhere. Hopefully, it's a minor one. Yeah.

And we're going to have to go through this all over again. And it's based on money. It's really based on money at the end of the day.

How much is it going to hurt us? And how much is it going to cost the organization if we're sued? At the end of the day- Or lose customers.

Even worse, lose customers. But it's money. But it's money.

But it's money. But also, I don't think you can compare the internet analogy with the AI analogy. You know why?

Because the internet started initially as a- Resource ... sandbox or a toy for technical guys. Eventually- No, research ...

research, yeah. Eventually, it became things that the average consumer used, but that took time. This AI tsunami is moving so fast and is impacting humans so dramatically.

I think you almost have to jumpstart the regulation in some ways. And I'm not in favor of big UN type of bodies either, but I feel like there's got to be some way for the thinking minds to come together and agree. So there's a technology conversation, but there's also a policy conversation.

And that policy conversation is where I think this belongs. I know we're at the end, but two points here. So Jack, to your point, I see the last 50, 60, 80 years as a very tiny period of time at the beginning of an era, right?

Right. So the patterns we... The internet not living up to the initial, as you say, centralization fault, is what it is.

And to your point, Sid, yes. Right? And again, I and many of our viewers and people here, we are the standards community.

Whether it's the civil society, UN people, or the standards group meeting, there's a handful of thousand people in the world who have been doing it for decades. We're still talking. We're working through it.

But we're inside these rooms, we realize it doesn't move at that speed anymore. Right? And if it is economic value, competitive value inside organizations that drives them to put out the standards we want or the practices we want, that's just a mechanism.

Yeah. So guys, here's where you missed it. Let me tie a bow on this so we can move on.

The real fundamental question here is do you believe that we're dealing with a technology that could potentially cause upwards of the extinction of humanity, as we've heard in the last couple of weeks? Or do you think we're dealing with the internet may go down. Oh my, oh my.

Or my phone has a virus in it. Oh, geez. Right?

It's orders of magnitude. If you think this thing rises to the level of human extinction, you're talking about nuclear non-proliferation type treaties and organizations, but with the caveat is it's not just countries that need to buy into this, but corporations and even individuals that need to buy into it. And you know what?

I'm glad you bring up the UN. Look what's going on at the UN today. It's become a bit of a farce.

Right? It's become a bit of a farce, and we're responsible for it Yeah, we are. Yeah.

Yeah. And so is that the model we're going to use for AI because that works so well? Well, one could say we haven't blown ourselves up since Hiroshima and Nagasaki.

But apparently we just came close, so who knows? Well. Well, maybe UN is not the right analogy.

Maybe it's the IAEA, right? Maybe that's what- Perhaps ... we should all think about.

Perhaps. That could rule humanity, right? You posed it well, Alan.

You talked about whether it rises to the level of obliterating humanity versus the internet, which is a technology conversation. I think the answer is we don't know. We just don't know, right?

So- I think we do know. I think we're making a choice, and I think that there are individuals and companies and governments all making choices. " No, everybody's making a choice right now.

And- And for their own self-interest, right? Exactly. You don't know who the good guys are in this plot.

Yeah. We'll have to wait till the end of the movie, but we've got to go to the end of this segment. I'm making a choice.

I'm buying a small little house up in the Adirondacks just in case. Okay. And you won't be connected in case the internet goes down.

All right, Mike, let's move on. What do we got for B here? All right.

B is we're returning to this topic of, I guess, the lifting of content to train AI models without permission. And there's been some recent unredacted court filings that show that the big tech companies were at least advised that this was going to be an issue when they were training their models, and they knew about it beforehand, and are still standing behind the notion that this is fair use of content to come up with something that is different from the original. But the trouble is, it seems like a lot of the things that get generated as output look a lot like the original, so it's hard to say one or the other.

But this is still kicking around the court cases. But I know, Alan, you've been following this closely. What's your take on the latest here?

Because Microsoft would say that the people who were quoted in here were advisors and not stating official policy. Well, Satya's, I think, an official Microsoft guy at this point. He's earned his badge.

Mm-hmm. But here's the deal. The big case that's still kicking, right-- Well, there's a few big cases, but the instant case here is the case against OpenAI by the Times and several other big-time publishers.

The Justice Department has weighed in with a, I think, amicus curiae brief, friend of the court brief, that basically says it's okay. AI is a national security goal, and it's too important, and it's okay for them to take your data. And Satya weighed in and said, well, if something was behind a paywall, that signifies that you really shouldn't just suck it up to use it to train your model.

Right? Because it's behind a paywall or a regwall, and therefore, you know that it's not just open to you to use whatever you want. I say horse feathers to that.

Nonsense. Right? This goes against the entire open source-- And what do you expect?

A Microsoft guy to know about open source? I don't want to bring up that old story. But you know what?

There's free as in freedom, and there's free as in beer. " And I'll tell you something. " I wrote a column about free as in freedom and free as in beer.

Just because something is free, it's not behind a paywall, doesn't mean that you get to own the IP when you read it. Yeah. That you can go use it any way you want.

There's licenses, there's copyrights, there's the ability to read something, the ability to use something, the ability to build something on it. It underscores the entire open source world that we live in now in terms of how software is done and what we use. It's basic copyright.

Just because it's free to read doesn't mean it's free for you to use any way you want, and it's certainly not free to use to come back against me. And I'm talking now as a publisher. Right?

We choose not to put most of our content behind regwalls or paywalls because we want people to see, to disseminate. We want to get that news out there. We want our point of view.

That doesn't mean we relinquish our IP rights to that content. Yeah. It's our IP.

Yeah. Free to read doesn't mean free to train on. I totally agree with you.

I think we need to separate access from permission, right? If Copilot is scraping all this information and giving an answer to the prompt user based on that information and not crediting that information, whereas in this example, I think I was reading the article, Bing, their search engine- Search engine ... is crediting the source and Copilot isn't.

That's a big problem, right? Right. Well, the bigger problem is three people in the world use Bing, but everyone who's using Microsoft Office uses Copilot.

Here's the part that's unconscionable, though. So they knew that this was an issue. Yes.

" But no. What they decided to do was just start hoovering all this content and then using it to train without permission, and then they ran behind fair use as kind of like, "Well, that's okay. " Yeah, but think about it, Mike.

I don't disagree with what you just said, but think about it. If I were, pick one, Microsoft, Google, OpenAI, Perplexity, whoever, had to go and do a deal with each and every publisher of content out there, it would take me 15 years to train my models. It's impossible.

You couldn't possibly do it. So I'm not saying it's right. Or you could've done it this way: A, you can negotiate with the big players.

But if they had said, I won't speak for Alan, but if somebody had said, "There's this training program that we want you to sign up for, and we're going to kick you back a couple of bucks for your trouble," I probably, we would've joined. Or create a hub and use MCP- No, this is- ... for all these guys to subscribe to.

This is- And for the AI agents to sign up. That's all. No, they didn't have MCP per se, but this has been a problem that's been dealt with before.

You make a blanket deal and let people opt out. But tech doesn't move at that speed, Alan. That's part of the problem.

Yeah. I'm not suggesting that it's correct. I'm not suggesting that what they did was kosher.

Right? But if you had to go through all of the legal processes to make this happen- Oh, wait. But wait ...

it would be six versions behind an AI. Time out, Jack. I'm going to vociferously disagree because this is the same b******t that's going on with the data centers, right?

They damn well knew that all those things were going to be violating things. They definitely knew it was an impact on the environment. They signed non-disclosure agreements with these people.

It is part of a pattern. Sure. This is a callous pattern.

Damn the torpedoes, full speed ahead, and if they say anything, wrap yourselves in the flag and blame China. So I think the long-term answer, sorry, Chris, I just want to make a quick point here if I may, is probably a licensing architecture, not an- Sure ... endless creeping war.

It's an endless creeping war today, and that's got to stop, right? Yeah. To me, the scale- And here's-- But Sid, let me just go one step further.

I'm sorry to interrupt you. Yeah, no, go for it. But here's how you know they're full of s**t.

What do they do as soon as the Chinese come out with a new model? Dario Amodei gets out here with his little cry rag of, "These poor people stole my stuff. " Hey, you didn't cry when you knew you took it from me, did you?

Yeah. That's the difference here. That's hypocrisy.

So then I think- That's hypocrisy ... I agree. I think the answer is a machine-readable licensing and the favorite word that we've been using in the last few shows, a provenance layer for AI data.

Right? So that's what we need Sorry, I'm not going to do business Let me see if I can take this from the courtroom to the lab. Because putting together what everybody said, Jack, you're exactly right.

The mechanisms didn't exist to achieve this. It just wouldn't have happened. We'd be five years behind where we are now or whatever.

And Alan, yeah, the players, their intentions, you have to debate that in court. But the reality is, look at what are the mechanisms we have on controls right now. We have legal jurisdictions and laws that literally have controls like a reasonable person would assume.

That's not automatable, that's not enunciable, that's not machineable. So there was no way. And the way this works is that if I, as the copyright holder or whatnot, don't like it, I can engage in a legal battle that will take days and weeks and months.

That's not the speed things are working at. So it turns out we have not had a instrumenting level. How do I make a decision at scale about what I can or cannot read and use?

We have not- Chris, I'm going to call b******t. Chris, I'm going to call b******t. Let me give you an example.

Let me throw a word out at you. Napster. Anybody remember Napster?

Oh, boy. I'll say the same thing I said to Jack. That we confuse all of human history and thermodynamics with the last 50 years.

Napster was a fascinating experiment in this issue. We have not worked out how to do this at scale. Yeah, but it shows you how quickly it took.

It took the music industry weeks, maybe months, to shut down Napster for ripping off people's content. Actually longer, Alan. We're back in the courtroom, right?

I'm an engineer. Sometimes you got to go to the courtroom, Chris. I love courtrooms, but there has to be a technical solution to this, and I can tell you how the technology works.

We do not have the provenance and verification evidentiary layer in our published content at all to make really informed, really good decisions about this. We don't. We have skipped that step.

It's not wired in. Chris, I'm going to, again, look, I'll point to open source, licensing, IP. This is not from 50 years.

Copyright law is a little bit older than 50 years, my friend. It's been around a really long time, and it's very clear. It goes back to Cain and Abel times.

Don't take what's not yours. Each of us are going to have to agree on what that means now. There's always a copyright statement on the bottom.

" It's there. And the other thing I'll add is, okay, why when they know that it's wrong, they're going to court, how much is this court case going to cost? Why don't they just pay people?

These are things I don't understand. Why haven't- Well, and Traffic did settle. 5 billion to start, which is big money.

These guys are making trillions. This is worse than Napster, actually, because Napster wasn't making money off of copyrighted material. These guys are making money off of copyrighted material.

Yeah, just pay. Just pay. Forget about the court.

I mean- Are you kidding me? Agreed. But what's the mechanism to make the payments is the other problem.

So you do need... Jack, you're right. You do need a framework around all of this.

The problem is, as I sit here as the editor-in-chief of Techstrong, publisher, whatever, I don't even know who took my stuff and used it where. Let's start there. Right.

Right? Are you going to agree with me right now? Excuse me, Chris?

Look, on the emotional level, I agree with everybody here. The actors, are they being good or not? I don't think so.

But my point is, though, to your point, we have no idea to see what's been consumed. Yeah, we got little copyright labels, but I'm telling you, we do not have the ability. No, but Chris, the AI labs know what they consume.

Plus- Yes or no ... plus, we set up with the Hugging Face incident, we set up a bunch of AI models to figure out what the AI agents did. Pretty sure we can use AI to figure out who's looking at what.

Yeah, exactly. I'm going to get back to what I stated earlier. We can create a machine-readable licensing and provenance layer for AI data.

What is so difficult about that? Well, I'll tell you, because again, I spent the last year or so doing exactly that. Because I think without that, you can't build anything.

But it's not quite as simple as you think. It's not that complicated. We have all the parts laying around, everybody's touching on the right bits.

But when you look at the individual parts, how do I take the human consumable, the legal system, and make it machine consumable? We need to actually live up to our standards. You say you're going to be transparent and provide receipts and so forth, build the systems to show that.

And we're getting there. But we didn't have these tools when these boats started. I mean, the generators of the copyrighted material would have to do some work.

There's no doubt about that. So you got to create content that could carry, for example, I don't know, metadata, whether it can be indexed, retrieved for RAG, used for inferencing. All these metadatas have to be created along with copyrighted content.

So that requires some cooperation. If they have interest in protecting that data, they should do that. So it comes both ways, right?

What you're really saying is that because it was inconvenient, it is okay to commit a crime. Which is roughly like me saying, "It's inconvenient for me to line up at the cash register, so I'm just going to take all this s**t home with me. " Well, yeah, I'm not going to let any of these folks.

They were trying that in San Francisco, I thought, for a while. They did. Hey, this is a little bit heated, but we need to take it down.

Stephie, take it outside. I say we'll see. We're over time.

We got to jump to the next segment, Mike. I'm sorry. Let's go to C block.

All right. C block is going into this, and this might wind up being contentious, but there's been this ongoing tension between employees and execs, and the execs think that the employees sometimes are dragging their feet about how to use AI and drive processes and deliver value, and employees think that the technology is, shall we say, not quite as robust as the boss tends to think. " And they're interviewing them, and they're coming up with answers, and just about everything on the index kind of suggests that the issue isn't the employees, the issue is the lack of leadership from the execs.

And I think when you're presented with that, the execs may see that more. But their number one at the top of this hit parade is handing out AI tools, and nobody has any instructions, and they're basically being told, "Go use this for something," but nobody has a plan. Jack, does this sound familiar?

This is right back to the discussion we had earlier with Kate about cell phones, and smartphones, and PCs, and internet, and email, and all of the rest. If you look at historically, and I'm sort of a history buff, but I can't make a living at it, so I'm doing this instead. If you look at it from a historical perspective, technology always rolls out in big companies well ahead of the processes and the controls that are put in place, and AI is no different.

The other problem, of course, is that you're going to have, in many organizations, you're going to have some people that are really good with technology and some that aren't. And so they're going to be struggling with how to use it and how to make it useful for my business. The third piece of it is that new technologies always bring about new business models within organizations.

It's just the way things are done. And most companies today that are looking at AI, except in maybe certain targeted areas like sales or customer service, are throwing AI tools out there, or letting people access them at least, without a clear vision of how that's going to change my workflow, how that's going to change how my company operates. And basically, it's a free-for-all.

It's how do I put processes in place and workflows in place, which we've been establishing forever, when you have these new tools in place. So it's a whole series of things. The other part of it is, similar to the discussion we had earlier, is where do I get information within my organization to make these AI tools worthwhile?

Most organizations have databases that are scattered all over the place. There's duplicative data. There is data that you can access that I can't, or I can that you can't.

It gets really, really messy. And then when you start rolling agents on top of all of this stuff They go off and do things on their own once we put them out there. It gets even more complex.

So the bottom line is that we're still in a very early learning process for AI in most organizations. We haven't trained people how to use it properly. We haven't put the proper tools in place to manage and control it.

We haven't put security in place to make sure that it's not doing bad things to us. And I think this is just going to be a process that's going to take a couple of years before corporations, or big enterprises at least, really understand what's going on. It's even worse with consumers, by the way, which we probably don't even want to get into.

So my next question then, if I look back in time, and since we've been all looking back in time this whole show, but every time there's a new technology, it does take years, and we wind up using it to do the same thing we always did slightly faster. But it takes years to figure out that there's a whole different business model or process or something that we can do entirely different that takes us a while to wrap our heads around. So Sid, is this just one more of these examples where AI will have this profound effect, but it ain't going to be tomorrow?

Yeah, it's all about the IT adoption model, right? So if enterprises could adopt IT themselves, they would have. However, they didn't.

They relied on the systems integrators and the managed service providers, all of this because there's a skill issue. However, I think the challenge with AI has been the, Gs, we talked about this, I think in the last show, the GSIs haven't stepped up and really done anything about it. And the model builders, like the Open AIs and the Anthropics, are too busy developing technology for technology's sake and shoving them over the wall to enterprises to adopt.

And that's where the big divide is. So we're seeing some changes there, where we're seeing the model builders either buy companies that can do this stuff, and these are not the traditional SIs that we talked about in the last show. So I think that's where the challenge is.

I think you're going to see more and more of the shift towards adoption, where, as Jack said, there are company-specific-- Somebody made a comment, Jamie Dimon has more valuable AI token than Sam Altman does. Because when you start to implement AI within a particular enterprise environment, it's not about just taking a frontier model and start to use that and succeed, because there's a lot of company-specific data that's sitting in CMDBs and databases of records that will never be scrapable by these frontier models. So putting all that together and creating those specific company and industry-specific models that can be used effectively is damn hard.

That's sausage making. Nobody's focusing on that, and I think that's the real problem. Oh, I disagree, Sid.

There's a lot of focus on vertical markets for AI. There's a lot of focus on small language models that are targeted towards specific companies now. If you look at what Google's doing, AWS is doing it, Microsoft is doing it.

I'm not sure SIs are going to be the right way to go about this either, but there's a lot of effort to try to make this much more in tune with what I'm trying to get done in my organization on a vertical scale than any other way. And so it's a messy problem that we've gone through this with all kinds of different industries in the past. This is just an amplification factor with AI because it's so pervasive or going to be so pervasive.

But that's my point. Google may be doing enough in small language models, but they're not doing enough to retrofit that to a particular environment, and that's where the role of the SI is going to be big. Because if you read the news recently, OpenAI acquired a company called Tomorrow, T-O-M-O-R-R-O.

Anthropic did a partnership with Blackstone to create a new company called Ode that is specifically designed to do integration for AI deployment within an enterprise, and that's where the real challenge is. So let me just make one quick comment. Sorry, Kate, but I've got to jump in here for this.

Sid, the problem with SIs is they all go to the cream of the crop. They deal with the 10 or 20% of the top layer of organizations because they're charging billions of dollars to do this stuff. That leaves 80% of the companies to fend for themselves, and that's a whole different model.

Jack- That's classic SI. So Jack, what you're saying, the problem with the Fortune 500 is there's only 500 of them. No, but also, Jack, the point is- How many companies are there out there, Alan?

Six million. Let's be real about what GSIs really do, because they show you the 10 people who really know something, but then who shows up is a bunch of college kids on a bus. That's right.

And they move in for two years trying to figure it out. Right. Yeah.

And charging you $1,000 an hour to do it. So the classic GSI model is broken, which is what Kate was saying. Which we talked about actually last week, and it is, that's a true statement.

Hey, do you know what the difference, I think, and where, again, the Pollyanna in me is so darn hopeful, is when it comes to the AI jargon. I don't remember. There was a few terms that we put out there, like shadow IT and things like this, but now we have a really good list that talks about all the different problems, and I think that's the difference.

We are listing the problems, and that is going to actually help us as we move forward. And it may sound silly, but it's words that we understand. We understand.

I keep saying, we've been a part of this rodeo for a while now. And when I say the rodeo, I mean the entire cybersecurity rodeo. So shadow agency, unchecked output, panic-driven experimentation, pilot graveyard, reality resistance, token maxi.

I love these words because they're really-- And it's a list of words. This is a great list of words. I love it.

Hey, Wordle, here I come with some of these. History repeats itself. In times of change, there's standard risks.

And we touched on a whole bunch of them in this exchange. So one risk is to say everything's going to change, and that's never the case. Some things from the past keep working in the future.

And another risk is to say, well, since then. And we have to recognize that perspective is my life, my career, that's about it. We think things are inevitable because that's the way it's been done.

That's where the risk of thinking the new models go forward forever comes from. But Sid, to your point, and Kate, models. Let's talk about models.

Well, we need to train models for small businesses and small models. No. Maybe.

What is a model? It's a set of information. What does an individual company care about?

Their set of information, which they already have, and some large language model, some AI system to process that. Yes, it matters what the model is, but not that much. So some things will be the same.

Big companies keep existing. Most of the mechanics keep existing. But the words we use over the next-- And Jack, back to your point, cybersecurity regulations and laws and so forth could not be written in 1998.

I was there. We were arguing amongst ourselves what that even meant. AI regulations don't exist today.

The controls do not exist today. " We're arguing. We're arguing on this show.

So there's no way it's going to be done. But it's trailing technology, Chris. It's always trailing technology.

That's the way standards are. It has to. Yeah.

How are you going to build the standards before you've built the thing, used it for a while, and realized don't do that? Sure. Yeah.

That would be cold comfort to me when I wake up one morning and all the power in my town is gone because somebody wrapped up an AI agent and turned it loose for grins. Just go do your background checks. You have a backup plan.

And no one's talking about measurable business outcome as a consequence of deploying AI. Oh, they are. Sure they are.

Yeah, but the model builders are not talking about it. Oh, not model. The model builders are model builders.

Guys, and pundits are pundits. We're out of time. We've got to end this one.

But for those, and I sense some of our gang members, it's cathartic to have these conversations. Mike, we should offer off hours. What do they do on Bill Maher after the show?

They go on YouTube or whatever. Right. We might have to do like the after-hours show or something to continue these conversations, but we got to call it a day.

So you're saying Techstrong Gang after dark? Where are you going? After dark.

There you go. Jack, Sid, Kate, Chris, thank you so much. Thank you, Kate.

Thank you. Thank you, Alan. Thank you.

What a great conversation. Great conversation. Yeah.

Well, it's because none of you have any passion for the things we're talking about. Care. Mike, we'll be back tomorrow with more.

Very well. So I'm trying to think. I actually think I'm at a conference tomorrow, but I'll talk to you about that.

But thank you for joining. It's been another great Techstrong Gang. Of course, we're here live Monday to Friday, noon Eastern.

tv, Techstrong's OTT app on Android and iPhone or iOS and Apple, Roku, Amazon. Or if YouTube's your thing, check out our Techstrong TV YouTube channel. Until tomorrow, though, on behalf of the gang, we're out.