Splunk’s IT Unification Bet Meets AI’s Energy Reality

Hey everybody, happy Friday. Welcome to the Techstrong Gang. We have a full lineup on a Friday, which tells me summer is definitely over.

So let me welcome our members today. John Schwartz, good to see you as always. Good to see you, Mike.

Everybody. All right. Tracy Reagan, I see you are still at home.

You've been at home almost every show that we go to. Yeah. Lately, I've been staying at home.

Going to stay home through Christmas. Lucky you. I'm just back from Denver, which was lovely.

But, I still like to be at home. Speaking of Denver, Kimberly Bates, we have not seen you in a while. How are you?

Haven't seen you in a while. I've been traveling a lot. But thanks, I was able to fit it in this week, so there we go.

Excellent. Good timing as always. Jack Palmer, how you doing, my friend?

Doing very good, thank you. All right. And Jeff Reich, what day is it?

In case you couldn't guess, it's National Cheeseburger Day, Mike, and I think everyone should go celebrate. It's about what you need to be. All right.

Well, sounds good. But let's just jump in here. I don't have a transition from cheeseburger to the next topic, other than maybe it's fully loaded.

How's that? I think that's a natural. But, I went to the Splunk conference this week, and as usual, this business unit of Cisco announced a billion different things, and it made your eyes cross after a while.

But there was a theme, and the theme was that we are going to finally maybe unify all the data that we use to manage IT, especially the telemetry data, and we're going to do that using something called the Cisco Data Fabric. And that's going to allow us to finally unify all these little IT silos that have cropped up over the years so we can have IT ops, dev ops, sec ops. We can all be one happy family working on the same thing at the same time.

And the thing that's going to enable all of that are going to be these AI agents that we have working for us that are going to be an overlay across all of that, and they are going to communicate with each other, and we are just going to happily supervise them. Tracy, does that sound good to you? Is that where we're going to wind up with all of this?

Well, don't get me started on agents, because all of us know that we're not even going to go down that road. The way I see it is that SIM and observability are working for what they're designed to do, collect events, detect suspicious behavior, show how systems are performing. But the problem is that visibility does not equal remediation when it comes to vulnerabilities and real problems.

We can have enormous amounts of telemetry data and still struggle to answer basic questions like, is a new vulnerability out there, or where's it running on the software, and how do I fix it? So it feels to me like Splunk is trying to take a different approach to pull some of this data together to make this happen and to make the data better. And siloed data is never totally useful.

You can have the data, but if it's not doing anything for you, how helpful is that data layer? And the data layer is extremely valuable when it provides context, but we haven't really seen a lot of context coming from these tools. And then my bigger issue with all of these tools is that these tools are not affordable for smaller teams.

And even in large enterprises, only some applications are using it. So large enterprises can justify SIM and observability platforms because they have these large environments and dedicated security teams, and somewhat sizable budgets. But smaller organizations and less critical apps face the same problems, but cannot afford the licensing and infrastructure and integration work that these tools require.

So I don't know if it's helping that much. I am always suspicious right now with anybody running towards anything agentic until we can get the problem solved. We get new information about agents going rogue every day, and it's happening more often.

So are we really solving cybersecurity for everyone, or are we increasing the cost, and are we just opening ourselves up for more problems at this point in time? So I think in this industry, we have a lot of problems to fix. Unifying it could help, but it's probably just going to be more costly and less companies are going to be able to afford to use it.

There is all of that. Kimberly, you've been in this thing as long as I have, and as I look at it, all these people that I've talked to all these years in these various departments that own all these different fiefdoms, they kind of like it the way it is. " So where's the pressure to unify going to come from, do you think, at the end of the day, when all these folks that are in these places kind of like it the way it is?

I'm not sure if they like it the way-- I understand the fiefdoms. I understand fiefdoms in different areas and that kind of thing, but I think in the area that you guys are talking about in terms of consolidating the information for looking at observability and the traction and everything else, especially that's going on in the AI agents, I think it's probably a move to make sure that happens because there's so much exposure that's going on, and there's so much concern. I think my bigger issue is that when anybody says, "I've got something to solve all the data management problems, I can consolidate this thing easily, and all you need to do is plug in the tool," my little propeller head goes up and goes, "Well, let me see you pull that one off.

" Because usually there's data in all kinds of different formats, and having a single tool to rule them all has, so far we have not been able to find that God box. Mm-hmm. But maybe Splunk is pulling it off, and maybe I don't understand what they're totally pulling off, in terms of what they're all pulling together here.

Well, I don't know if it's going to work or not. I can tell you what they said it will entail, because they're saying instead of having to move all this data into some massive repository somewhere, which would be difficult and expensive, then it'll be a more federated approach, and we'll be able to analyze the data where it already resides and wrap together some context around that. And as part of that, they were saying, "We're going to have more affordable plans for Splunk," because while you're not going to have to put everything in a repository per se, and they're moving to something that at least they say is going to feel a little more outcome-based now.

Every time I hear a vendor say that they want to charge for something that's outcome-based, my alarm bells go off because it's like, well, who determines what that value is, and how we're going to pay for all of that and how that all comes together? So I believe that to be very problematic. John, let me ask you something.

Does any of this sound familiar? Are you hearing similar noises from different vendors out there? Are they all saying that?

Yes. As a matter of fact, Dreamforce was this week as well. We talked about that yesterday on the show, and the one thing, and I think you all have mentioned this, there's this kind of undercurrent of resistance to change despite all the beseeching and the megaphone coming from Splunk, ServiceNow, and Salesforce in particular the last couple of days.

As a matter of fact, ServiceNow's going to be announcing something in a couple of days in this vein. And to me, it's like they're rationalizing, they're trying to oversell, they're pushing the AI agents narrative a little bit too hard. I think that the difference between what happened with Splunk, which seemed to me more substantive than Salesforce, is that Salesforce does these flashy shows where we have people like Jensen, Dario, Sam Altman, Katy Perry, are all there.

It's a lot of salesmanship, and I'm not sure how it's resonating within people within enterprises. And I think you brought that up, Mike. Yeah.

Jack, they are pointing to security as the reason that all of this is going to happen, because while the vulnerabilities are so many and the exploits are coming so fast that we need to respond as a unified team in milliseconds, and that requires everybody to get on the same bus. Is that something that resonates with you? Is that a hope or a dream or a noise?

It's a little bit more of a hope and dream and painting a picture of the future, which I think is great, and that's what we need to do. And I think there's a lot of the details of what they did that I really like. But from the big picture point of view, it feels to me a lot like a solution to a problem that doesn't exist.

Mm-hmm. Which is let's throw AI at everything. Now, when you dive into the details, there's sort of two things that I think they're doing that's very important.

One of the biggest challenges, and I think both Tracy and Kimberly sort of highlighted this, one of the biggest challenges with SIMs in general is it costs a lot of money to take your telemetry and throw it into the cloud so that the SIM can analyze it. And in fact, what people do because of that is they limit the amount of data that they throw at the SIM, which limits its usefulness, because the more data you have, the more cross-correlation. The longer you can keep that data, the better likelihood it is that you can find a slow-moving, long-lived attack from a very sophisticated actor.

So if you're now being able to point the analysis engines and the AIs at data you store locally on your local system, where it's a lot cheaper than shipping it into the cloud and being charged for it on a volume basis, that's a very interesting component that they're adding. The other thing that they're adding, which is very important, is the ability to throw AI at natural language expression of what you're searching for. And I think one of the things AI is now very, very good at is being able to correlate different structures of data, right?

So you don't have to normalize all your data into one format to be able to analyze it. So if you take a log message from this type of solution, and you have some telemetry that isn't in a standard format from some other solution, you can still point the AI at it and do cross correlations and make it understand things. So those are really good from a security perspective.

But if you think about it and say, "Okay, what I'm going to do is paint this big picture of the future. We're throwing AI agents at it and all of that stuff," then you get into the Tracy territory, which is this stuff is scary because AI agents right now, from a security perspective, are very scary. This is true.

I don't understand why everything has to be solved with natural language. Whatever happened to data engineering and machine learning? What happened to MLOps?

And this is a perfect example of where MLOps alone could solve this problem. I really don't understand why AI has to be the answer to absolutely everything. It's like we got a hammer, we're going to just start beating everything with it, even though it's not a nail.

I just don't get it. I really don't. I do, but it's a conspiracy theory.

Here's the thing. Yeah. When you go talk to the business leaders and IT leaders of companies, they complain about the cost of labor, and the cost of labor goes up every time I have a specialist.

So what they really want is, can I get away with running this with an IT admin making maybe $90,000 a year throwing natural language at this thing versus going out and having to find a DevOps engineer or any of these other so-called specialists. Or a data scientist. Or a data scientist.

So I don't know, Jeff, I'll throw that to you. Is this a conspiracy theory, or is this what's really the end game here? However, I have to say, if you were to say this has to happen, and you were asking me to, on a particular market, say which organization could pull it off best, Splunk probably wins that bet based on their history.

However, what I think we're hearing is, if you can use an analogy, you know how I like looking at things, that delivering all this stuff is too expensive for me to do. This truck is too expensive. Splunk is saying, "Oh, don't worry, we have a low-cost delivery vehicle for you to get.

It runs on diesel, sorry," to use today's analogies. Because even though we're not going to pay by data size anymore according to this, to Jack's point, which has always been a limiter with Splunk, now you're saying they don't tell you how many agents are running there. They do talk about token metering, but when they say we're going to use AI agents to actually get it all together, I really think there's a pretty big hidden cost there in two ways.

One, the actual what are you paying for the tokens to run, and two, to Tracy's point, what are you paying to fix the data or the results that come out of that when you discover that there was flawed logic or lack of a guardrail that got you there in the first place? So I think it's a good idea because we're going to be looking for give me a single pane of glass rather than, I have to look at nine monitors at once. However, I'm not certain we're quite mature enough to be there yet.

And I don't know exactly what problems they really want to solve with it. What is their problem statement? Which I really didn't hear in the articles that you presented.

I didn't hear what they're trying to solve. And I get the data being siloed. I totally get that.

But I don't understand why you would want to connect those with AI, considering what we're going through right now at this point in time. Well- I mean, the quantity of data that we're talking about to parse through. One of the things that I read through there, and Tracy, I don't know this area that much as like you do, the log files and everything else, but the amount of log files, et cetera, we're talking about being generated by the activity that's going on significantly has had to increase.

And for something to spin through that at the speed that we're looking at, yes, we do have machine learning capabilities out there to spin through it, but is the AI technology more efficient? And efficient is a very broad term right now, what efficiency is, because of all the stuff that we're using to do this. But is it more efficient, and can it be more effective at finding potential risk?

Mm-hmm. Right. Can I throw out another conspiracy theory?

Sure. This might be totally crazy, but I'll do it anyway. Do you think, given what we've been hearing of all these AI security scares, and they're accelerating every day, these companies are almost tripping over each other talking about their responsibility or breaches or acknowledging things.

Does that in any way prompt IT folks to then more seriously consider using AI? Does that actually help the greater ecosystem? No.

I'll answer that question. It should. Which is it should, but it doesn't.

Right now, what I see in talking to a lot of people is the current perspective from the C-suite is that AI is so beneficial that it's full steam ahead, damn the torpedoes, let's just go. And there's a level of accepted risk today that is much higher than it used to be, and we haven't yet had a major catastrophic event that will cause people to rethink what's acceptable risk and what isn't. Of which we're aware.

Of which we're aware. But yeah, at some point, there'll be a public event where people will finally clue into- Yeah. I ask that- ...

accepted risk. I mean- Right. Yeah, Jack, I ask that because, and I'm sorry to interrupt.

" Right. It's just like part of a marketing plan sometimes, I think. It's a P&L play.

But I think you would need to divide the kinds of use cases into different categories to be able to talk through that. Because John, you're talking about those use cases. On the other hand, sorry, I'm stumbling on my words here.

There are applications that are getting pulled back because they don't think they're ready for the enterprise space. Got it. And they're scared to deploy that, and that's an absolute that's happening.

And then there are ones like this one that maybe has been tested by Splunk, Cisco, et cetera, as a contained case that can be deployed with a level of confidence and less risk. So you know what, I would be excited if they were saying that we're going to build a small language model based on this data, as opposed to using foundational models. This is the biggest problem I'm having with AI right now.

We don't need a foundational model to determine what's happening in SIM or observability on our systems. So they are- So what we're doing is we're connecting AI into our production systems, which is what we don't want to do right now. Yeah, they are- It's not about securing AI.

It's what it's connected to. They are talking about partnering up with NVIDIA for small language models, but they're not quite fully baked yet, but at least they have the concept. But I come back to what Tracy said earlier, and then I'm going to end this conversation here, but yeah, what are we going to spend?

How many millions of dollars on platforms and the consultants to configure the platforms to save what, exactly? And what would be our ROI? And if the ROI is just that we're going to be more efficient on IT, well, how many years has to go by because I am now using fewer IT people who are admins versus specialists before I get to an ROI on the thing?

I think that this story and the math don't quite play out fully, but maybe someday soon. But for now, wait and see. All right.

I'm going to shift the gear here to AI and power. And you'd be surprised, but there's yet another alliance in the world of IT. I am shocked.

It's just like these things seem to be just in and out every other day, almost like somebody just wrote it up on a piece of paper and shipped it out, and it didn't really mean anything yet because there's not really anything behind it. But anyway, we have a new initiative by multiple members of the industry to create a more efficient energy data center setup where, if I read it correctly, the goal is to create data centers that can more dynamically plug into grids, maybe not plug into grids at all. And of course, if you look around in Congress, they are trying to solve this problem with some legislation.

But I'm going to go with Kimberly first. Can the industry solve this themselves? And for that matter, if they can, then why the hell didn't they do it before they started this mess?

Well, it depends on what you mean by the industry. If you mean OpenAI, Anthropic, and Google, no, because they don't own the grid. Mm-hmm.

They have to engage with the people that own the grid. " It appears this might be driven partially by the Duke report that came out, that is talking about how, and I'm going to mess up the words here, some variables of use of the energy they found. And I think this is the classic situation where you're not always using all the energy that's getting produced, so can you capture some of that leftover energy?

And they said, yeah, there's about 100 gigawatts there that we believe that we can capture by variables. And some of it has to do with storage, but you have to have a new redesign in order to do that. So if you read through that, and I didn't read through the entire thing because I only started looking at it this morning, there's possibilities there, and that I think that this group here is saying, okay, so they're coming together to say can...

I believe coming together to say there are other ways to look at the grid to keep it as reliable as we need it to be, use whatever extra is already out there that we should be capturing, and feed that in with other systems for storage, et cetera, and feed that into the data centers. There's a bunch of problems with it, in terms of classic last mile kind of stuff that's going to be going on. But, you put a bunch of good heads together, maybe if we can't capture 100 gigawatts, maybe we can capture 25 gigawatts.

That's a lot. Or 50 gigawatts. And do you think they're going to do that for free?

Because it sure smells like, buy my product so I can help fix your thing that drives my product. Well, I think the free stuff is that there's more tech systems and technology like storage and that kind of thing that's going to have to be used by the data centers. And since they have the need- Mm-hmm ...

and they're going back to the municipalities and the power companies, et cetera, saying, "I need this," they're looking for every which way they can get it at the same time that as we're looking at the government saying, "Don't take this stuff away, but it's from the personal. " And you can't raise our rates and build. So they're looking for new options, and the cost is going to be sitting on, I believe, it's primarily going to be sitting on the AI folks, because I don't believe the power companies nor the local municipalities are going to permit anything else.

Oh. John, let me ask you this. Is this just part of an ongoing wave of PR aimed at making data centers look better?

Yes, absolutely. So, interesting. So interestingly enough, the House- After all that analysis, that's what you're going to say, John?

Yeah. That's it. Now move on.

No, so here's the thing. So the House unanimously and by partisanship, 417 to three. When's the last time they voted agreed at that magnitude?

So they passed this along. They have this Ratepayer Protection Act, so it's to protect everyday consumers from utility price hikes. It goes to the Senate, and the Senate has this long tradition of debate and scuttling or saying no to House legislation.

This is nothing new. But in a sense, the measure trying to force tech companies to pay for the cost of new energy infrastructure used for data centers, rather than pass it on to consumers. I don't know the details of this, and I haven't written about this bill, but there's already opposition in the Senate from the Senator Martin Heinrich from New Mexico, who I think interestingly enough is the top Democrat on the Senate Energy and Natural Resources Committee.

He says it doesn't go far enough, so he probably knows what he's talking about. " Meanwhile, on the other side, we have the embattled Ohio Senator, JD Vance, who's up for re-election against Sherrod Brown, and he wants to approve the legislation under unanimous consent, just rush it through. So in the Senate, it becomes partisan because I think Heinrich has a better understanding of what this really means, and he wants a better bill.

So that's my take from 37,000 feet. All right. Tracy, for those of us who do not know your hometown senator, what's his story?

Well, think about who our state is. Mm-hmm. We are pretty much a laboratory for this debate.

We have a lot of renewable energy potential in this state, and there are companies who want to come here and build these enormous data centers. But we're also a water-constrained state, and I've talked about that before. So with a grid that wasn't really designed for individual customers consuming hundreds of megawatts, by the way.

So the question for us isn't whether we want data center investment, it's who builds the infrastructure to pay for it, and whether our small local communities benefit from the resources that they're giving up. Which is probably, if you think about the Jupiter Project, which is out where they grow a lot of pecans, those folks are realizing they may not have water for their farms anymore, and that's a big change in the culture down there. So if a data center uses as much electricity and water as a small city, should they not be treated, literally, should they be treated like an ordinary commercial customer?

And I think that's what Martin Heinrich's saying. He didn't oppose the idea of making data centers pay. He argued that it didn't go far enough.

Right. We hear a lot about it, believe me, and he wants the requirements to have real teeth from the federal side, rather than simply a bill that asks states to consider it. Mm-hmm.

So I think that he's making an argument that it's got to be a stronger bill to make it real. In all fairness to this, and that happens in the Senate, right? He considers it, and maybe they amend the bill.

It's salvageable. Well, he proposed his own legislation- Yes ... called the Grid Saving Act, which we heard a lot about here, which would direct FERC to establish these rules requiring these large electricity users to pay for facilities and upgrades.

It always comes back to, and we always talk about this theme that in Congress, they really don't understand tech. They really haven't for a long time. And in a sense, we have this guy, Heinrich, who really knows what he's talking about in terms of this topic.

So in a sense, I think he's trying to be a responsible voice. And so maybe they do get it right, but it is heartening to have some sort of opposition, not rubber stamp and rush something through. There you go.

Jeff, you want to say something? Yeah. The downside of going for unanimous consent is all it takes is one senator to say no, and it stops.

Right. And that's exactly what happened here. And the issue with, Kimberly, you make a good point about there's wasted energy.

It's not like you can run around with a bucket and capture it, and I'm not suggesting that's what you were saying. Exactly. But you need the capacitors or batteries- Mm-hmm ...

to store it after you redirect it, which means there's transmission lines, both local and long distance. There's storage capabilities, there's batteries, and then there's redistribution transmission lines. None of that's in place.

Yes. So even if we were to say, yes, we can do this, I'm not sure it'll even happen in my lifetime as described. Yes.

So there's going to have to be a better solution, and the The traditional method of utility rates that in many municipalities and utilities give greater discounts to larger users, it's really an incentive for them to use more. That needs to be flipped on its head before we even look at recapturing or anything like that. So there's a whole bunch of fundamental fixes that I think need to take place before we can consider here is a bill that's going to universally say, AI data centers pay for themselves.

And maybe, just maybe, think positively here is I think the frontier large language model people are looking at ways to say, are there ways to work with the grid system to change how things are and put the investment in? And are they motivated enough to do that? Mm-hmm.

And it seems that some of these data centers are going in, they are looking at changing how they're doing things and looking at other ways to do it because they want those data centers to go up and realizing this is the biggest deterrent for them to be putting those data centers up is the grid issue right now. Yeah. Here's my problem with the whole thing.

You basically have a bunch of companies and vendors who damn well knew that this was going to be an issue. And now they're turning around saying, "Yeah, trust us to regulate this amongst ourselves. " And this is a theme that's going on with all this AI stuff.

Mm-hmm. And it ranges from how they got their content to the amount of pollution that they generate. " Right.

That was part of the AI speed initiative, right? It's like, let's get this through as fast as we possibly can. We'll worry about the repercussions later.

" I think it's Nvidia and Meta right now, and they've been more and more vocal about it of late. Anyway, I just think that they've all shown their true colors. So if this problem's going to get solved, it can't be by the people who created it in the first place because, well, frankly, they can't be trusted.

All right. I'm going to shift the gears because, well, King Charles had something to say about AI, and he was in Scotland where I guess some royal society or another was having a meeting about AI. " At the end of the day, when you think about all this stuff, though, and I guess I'll go to Jack with this one.

Jack, it's all about safety at the end of the day. And is that achievable in your mind? Or have we reached some sort of point now where no matter what we do, the AI agents, because of the way they're designed and programmed, are just going to end run every guardrail and every safety thing we can possibly come up with?

Well, I would argue that your premise is incorrect. It isn't about safety at all. It is all about...

And now it's my turn to let my cynical flag fly large. You got your- Do we have a white flag that I can wave up here to signal? But there's a lot of posturing that has been going on in the last three or four weeks that has to do with regulation.

In the history of capitalism, when companies say, "We want to self-regulate," it has to do with putting up barriers for the competition. We'll set up artificial guardrails so competition can't get in. And then if you look at a lot of this ethics conversation around here, the people putting the ethics conversation or the safety conversations around AI are doing the same thing using different words.

It's the same group of people funded by the same group of people. It's the people who have the money that can lobby. And as an aside, when you look at the who was in attendance with this, among other people mentioned was Jensen Huang.

Mm-hmm. When does Jensen have time to actually run Nvidia? Can I ask you this, Jack?

I have a question. Does Jensen have a body double or does he use a transporter? He has an agent.

Seriously. That's why he wears the same clothes all the time. Yeah, exactly.

He never changes his clothes when he's... So, and he's in LA when President Trump calls in to his appearance, then he's at Dreamforce, where he's leading Benny off along like a little puppy, and then he's in Scotland. How is this possible?

Anyway, I'm sorry, Jack, go ahead. Well, no, like I said, it's a Friday. We're all letting our cynics fly, right?

That's all I can say. And I'm sure Jensen ran home after hearing the King's speech and made major changes to his video. Right.

I think that's- I look at Jensen's role, particularly Jensen's role in Nvidia. Nvidia, AMD, and Intel are small arms merchants. They're selling their stuff to everybody they can, say, "Push chips out the door.

Use them however you want. I'm just a chip supplier. I'm just going to put it out there," right?

Same thing with TSMC. We'll build anybody's chips. Then you look at Meta and OpenAI and Anthropic and Google and Microsoft and Oracle.

Those are the guys who are the big engines pushing AI, and they're the ones who need to impose guardrails on their own stuff or have guardrails imposed. And then there's the next question is when you, and Mike, I think you mentioned this earlier in another aspect of it, but you're talking about who gets to define something, right? Who defines the ethics of this and what is safe and not safe?

And coming from the Brits, who have a huge history of lack of free speech, I find it rich for them to be commenting on ethical standards on AI. The British have arrested more people for saying something on the network than any other country in the world. They're worse than Russia right now in terms of arresting people for saying something that they find objectionable on Twitter or on Facebook.

So, I take it with a very large dose of salt. Yeah. The other thing is King Charles, without any binding incentives or legal friction, he has this moral appeal, right?

It's not going to lead anywhere. They've got something called the AI Security Institute- Mm-hmm ... that enforces guardrails, but eh, this is all talk.

I have a lot of English friends who still listen to the monarch. For them, it's the spiritual leadership of the country, and they still lean in. And they are perfectly rational human beings as a result, but that's still where their mind is culturally.

So I think when the King speaks, it matters more in certain sectors than others, or geographies at least. Well, from a culture perspective, remember almost all of the AI companies are US-based, and here in the United States, we have a very long history of saluting the monarchy with a single digit, so Yes. I think the King's speech was- I think you just said, "He's not my king.

" Is that what you said? So I think the importance of it is that the King took the time to say something about it. And in that, you've got somebody says, "Yes, this is on.

" Okay, so from there, do you trust the UK people that they're actually looking at it? But at least you know that, hey, Charlie and his wife are doing... I think he wanted to just be able to say, "Hey, I told you so," in the future.

Yeah. Wasn't there a representative from the Vatican there too, right? Their AI head was there.

Yeah. Well, from what I understand, the Pope is working on a, whatever you call those things that he pushes out, a paper on AI, which will be interesting given what is going on in terms of Especially if you haven't looked at the Anthropic's constitution and how they treat AI almost a consciousness being. And so I'm very interested in hearing what he's- Yeah, they do come out with a statement every once a year at least- Yeah ...

about responsible use and humanity versus AI. The word humanity is being used quite a bit, and it was used in his speech as well. Mm-hmm.

And I think it's telling that the word humanity has to be pushed when we're talking about something that's not human. We need to keep that in mind. And Jack, I'm just going to grab that cynic flag and wave it higher because to me, although, it's aspirational and nice what he said, humanity and good for the community interest, but he's essentially saying, if you're British, don't pee in the pool.

And it's not a problem because the Brits aren't doing it. But the pool is filled with people that are not all British. So we're all going to get the effect of that until we have...

Cambry, you don't know me yet. No, I'm thinking how this goes along with your cheeseburger behind you. I'm sorry.

No, do not conflate the two. One may come after the other, but no, they're only loosely coupled at best. " And maybe it's the UN, I can't believe I'm saying that, but maybe it's the UN that says as part of human rights, we come up with how human rights are protected by use of AI, and find a way to get some sort of enforcement behind that, and UN may or may not be the right body for it.

But until there's something universal like that, it's great that any individual country, monarch, president, premier, whatever, makes these statements. It's good- Mm-hmm ... in the right direction.

But until we all get there, it's difficult to make it work because even if you don't pee in the pool, there's someone right behind you maybe. We should salute the whistleblowers, right? We should also salute the whistleblowers who spoke up in the last couple of weeks, because before then, we weren't hearing the word humanity or responsible AI use or development slowdown.

We were not going to hear any of that unless Jacob Cox and some of these other guys spoke up. So I'm going to- Really? salute them.

John, you guys talked about this I think on the show yesterday, but this is the part that is totally cray cray. So you had the head of AI from Microsoft saying, "Hey, you know what? " Well, guess what?

We can go back and retrain those AI agents using frameworks that are a little better and more responsible, and certainly not the ones that Anthropic is messing around with. " And it's just- Can I push back on that humanity thing? Yeah.

And the reason being is because I'm a member of the American Society for AI. We've been around now for about four years. Three years ago, we were debating, does AI have a persona?

Mm. Does it have a personhood? And that society is there for the goodness of humanity, the AI for the goodness of humanity, and we've published a huge anthology that has to do with those things.

So I think maybe, and I hear what you're saying because you're right too, that there is a section of us like me that hasn't really thought about as much of the humanity piece except for that I am part of this very, very diverse group. Oh, yeah. No, I don't- Of lots of technology people.

Okay, no, I'm kind of pointing the blame at the big players who are in such a rush to cash in, especially as they get close to going public. And it took some people within those companies to actually speak up, and it kind of reminds me of Facebook and Meta and how those voices were suppressed or ignored altogether, or they were forced out. And so I think it's healthy to have a degree of skepticism about where this is headed.

" That's what it always comes to. Trust us. We can self-regulate.

We know better than anyone. And that's what scares me because they do know, but they also kind of keep these things under the covers until they're forced to talk about it. Well, there's that, but there's one other thing, because Mike, to your point about retraining, don't forget that a lot of these particular models that we're talking about that went, quote unquote, "rogue," that was their training.

Mm. These are models that are cybersecurity specific models whose whole goal, and it's a goal-seeking model, is to go and break into something. So don't be surprised when it actually follows its training.

Exactly. All right. We got to end this conversation.

But I would just point out, maybe AI has some sense of being conscious or not. That's debatable. But what's not debatable is, A, machines and AI agents are not people.

They don't get to vote. They're not really a part of the society. They are impacting society, but they are not members of society.

So let's keep this in mind to remember who's in charge of this stuff. I'll leave it there on a Friday. Thank you all as always for sharing your insights and you guys are awesome.

And I want to thank everybody else for watching the show and staying with us. And please stay tuned for the rest of the Techstrong TV replay. And of course, we have a little note from an upcoming conference that we're hosting on Quantum.

Stay tuned for that too. We'll see you all Monday.