WEBVTT
00:00:04.959 --> 00:00:16.399
Welcome back to the AI Accelerator Podcast, where we cut through the noise of AI in the world today and get to what actually works when AI works for you inside your organization.
00:00:16.559 --> 00:00:21.199
I'm Matt Sembratsky, and today my guest is Scott Anderson.
00:00:21.600 --> 00:00:29.039
Scott spent nearly three decades as an enterprise architect at companies like that you might have heard of, like Microsoft, Lockheed Martin, Verizon.
00:00:29.199 --> 00:00:35.920
And he's also the author of two books on AI infrastructure, including his new one, designing AI data centers.
00:00:36.159 --> 00:00:37.679
Another big hot topic today.
00:00:37.840 --> 00:00:41.200
I'm very excited to have uh Scott on the podcast today.
00:00:41.280 --> 00:00:43.439
And I'm I'm really excited for many reasons.
00:00:43.679 --> 00:00:47.520
One is that a lot of leaders talk about AI as software.
00:00:47.759 --> 00:00:52.880
Scott thinks about it, about how AI actually runs, the infrastructure it runs on.
00:00:53.039 --> 00:00:54.799
That actually changes the strategy.
00:00:55.039 --> 00:01:03.840
So you're all going to have a great uh bird's eye view into the uh the mind of Scott and how AI really works inside infrastructure today.
00:01:04.000 --> 00:01:06.480
So uh Scott, welcome to the podcast.
00:01:06.879 --> 00:01:07.280
Awesome.
00:01:07.439 --> 00:01:08.000
Well, thank you.
00:01:08.079 --> 00:01:09.200
I'm I'm very excited.
00:01:09.359 --> 00:01:12.640
This is uh a topic that is near and dear to my heart.
00:01:13.760 --> 00:01:14.400
Fantastic.
00:01:14.719 --> 00:01:15.599
So let's get started.
00:01:15.760 --> 00:01:25.920
Before we before we dig in uh um into the the crazy topics we're gonna hit on, you know, fill in the gaps for people, you know, what what's your background?
00:01:26.079 --> 00:01:28.079
How did you end up being uh doing this work?
00:01:28.159 --> 00:01:39.519
Uh from what I understand, you started out in the education world and uh teaching, and then you moved into the wild world of the uh corporate world and uh digital AI infrastructure and so forth.
00:01:39.599 --> 00:01:40.560
How did you get there?
00:01:40.799 --> 00:01:52.719
Yeah, well, uh I mean, if you think about it, being a school teacher, uh you're you're having to connect to a bunch of people, you're having to build a consensus, uh, and you're having to help people move forward at whatever pace they're moving.
00:01:52.799 --> 00:01:56.079
And no one in a classroom moves at the same pace, right?
00:01:56.159 --> 00:01:59.920
You don't get 22 kids and you're like, we're ready to roll.
00:02:00.079 --> 00:02:00.719
Let's go.
00:02:00.879 --> 00:02:02.719
Let's hit it's uh zero to 60.
00:02:02.799 --> 00:02:04.079
We're at 60, let's keep going.
00:02:04.239 --> 00:02:13.039
60, you know, it's you know, you got half the classes at 15 miles an hour, half the classes at 35 miles an hour, half the classes at 75 miles an hour.
00:02:13.120 --> 00:02:13.840
Um, right?
00:02:14.000 --> 00:02:15.919
Those are the ones the cops are usually chasing.
00:02:16.159 --> 00:02:19.759
Uh but the reality is, right, it people move at different paces.
00:02:19.919 --> 00:02:22.560
Uh, and so, you know, I kind of was doing that.
00:02:22.719 --> 00:02:29.439
Um, and uh uh I met a wonderful person, um, and uh I was a tenured school teacher.
00:02:29.599 --> 00:02:38.400
Um, and I said to her, and she was literally graduating college, uh, she took the long, long plan uh to make it through her bachelor's degree.
00:02:38.479 --> 00:02:41.680
Um, and so she was graduating college and she was looking for a job.
00:02:41.759 --> 00:02:47.759
And I said, Well, if you find a job in another city that pays more than what I'm making as a tenured school teacher, I'll move.
00:02:47.919 --> 00:02:52.879
Uh and so three months later, I was living in a new city, uh, looking for a new job.
00:02:53.120 --> 00:03:03.360
And uh I got offered a job by the school system, but they wanted me to take a pay cut, a lot, a huge pay cut, like enough that it would have really impacted what I was doing.
00:03:03.520 --> 00:03:09.680
Um, and at the same time, this computer training company said to me, Hey, you're a teacher, you know computers.
00:03:09.840 --> 00:03:11.759
Why don't you come be a trainer?
00:03:12.240 --> 00:03:17.599
So I did that, started making more money, moved to help desk, and from there I kind of became a consultant.
00:03:17.680 --> 00:03:28.400
And for the last 30 years, I've been a consulting architect working on uh international, national, US DOD, US government, uh, state government, uh, local government projects.
00:03:28.479 --> 00:03:31.599
Um, but I've always evolved with the shift, right?
00:03:31.759 --> 00:03:37.520
So I was a collaboration person in the 2000 era, uh, internet, web, collaboration.
00:03:37.599 --> 00:03:42.800
Um, I moved into becoming a cloud person in the 2010 range because you know, cloud was the big thing.
00:03:42.960 --> 00:03:44.479
Uh, cloud was the new thing.
00:03:44.639 --> 00:03:53.360
Uh, and then lately I've moved to being an AI person because I personally believe AI is going to augment humans and make us better.
00:03:53.439 --> 00:03:58.159
Uh, and I refuse to uh I refuse to buy into the doom and gloom.
00:03:59.039 --> 00:03:59.919
That's amazing.
00:04:00.240 --> 00:04:01.120
That's awesome.
00:04:01.199 --> 00:04:01.599
That's great.
00:04:01.680 --> 00:04:04.960
Yeah, I I I uh echo the same sentiment there, uh Scott.
00:04:05.120 --> 00:04:05.919
Fantastic.
00:04:06.240 --> 00:04:09.439
So um that catches us up to today.
00:04:09.599 --> 00:04:22.720
So now for our curious listeners out there, the business leaders who are uh are leading all the knowledge workers, you know, the tens of millions of knowledge workers out there that are getting so much work done um with or without AI, right?
00:04:22.800 --> 00:04:24.560
Because there's a lot of adoption still to do.
00:04:24.720 --> 00:04:27.920
But um, catch everybody up to what do you do today?
00:04:28.079 --> 00:04:29.519
Who do you do it for?
00:04:29.759 --> 00:04:33.040
Um, you know, are you are you just doing one job with one company?
00:04:33.199 --> 00:04:35.920
Are you uh you know a consultant to the industry?
00:04:36.160 --> 00:04:37.759
Is it tied into data centers?
00:04:37.920 --> 00:04:41.759
Is it tied into digital infrastructure within companies, all the above?
00:04:41.839 --> 00:04:47.040
Uh, there's a lot for your background, you're suit, you're you're highly qualified for a lot of different directions.
00:04:47.199 --> 00:04:50.480
So, what's the short version of what you're doing today and who you're doing it for?
00:04:50.720 --> 00:04:58.079
Well, I'm I'm working on a contracting gig, working for a company doing uh RFP responses on the federal government side.
00:04:58.160 --> 00:05:02.560
Uh, and then I'm also working with another organization and we're doing uh AI assessments.
00:05:02.639 --> 00:05:05.600
So I'm I'm actually building uh so I have a book.
00:05:05.680 --> 00:05:11.360
You you mentioned the two books that are in the market now, Infrastructure for Designing AI Networks and Designing AI data centers.
00:05:11.600 --> 00:05:12.800
I have a third book.
00:05:12.879 --> 00:05:17.040
Um, it's called Organizational Debt or Why Your AI Project Fails.
00:05:17.199 --> 00:05:19.199
Uh, the publisher just optioned it.
00:05:19.279 --> 00:05:25.040
I literally just finished all the paperwork that the publisher sent me, um, and they have accepted it.
00:05:25.120 --> 00:05:31.279
Uh, and now they're gonna go through the hard part for me, which is they're gonna edit everything, send it back to me, and I have to clean it up.
00:05:31.360 --> 00:05:37.360
Uh, but in that book is a concept of an organizational assessment, and that's what I've been really focused on.
00:05:37.439 --> 00:05:46.639
Uh, but I also do an AI newsletter um for uh about 200 of my closest friends, um, and then constantly growing, uh, trying to get information out.
00:05:46.800 --> 00:05:49.439
Now, I do AI news differently than most people do.
00:05:49.519 --> 00:05:54.959
I actually go to the AI engines and I ask them, tell me what the led, what's the biggest news from last week?
00:05:55.040 --> 00:06:01.920
Uh and then I ask each one, and I I literally paste into the uh newsletter format that one of my friends designed for me.
00:06:02.000 --> 00:06:03.920
Um, you know, I paste it in.
00:06:04.000 --> 00:06:12.319
So you see what Claude says, you see what OpenAI says, you see what Meta says, uh, and then I do analysis because it's interesting, right?
00:06:12.560 --> 00:06:21.920
Last week OpenAI had two agents that went rogue and attacked Hugging Face like 116,000 times in a minute.
00:06:22.000 --> 00:06:24.079
Uh, but just a massive thing, right?
00:06:24.959 --> 00:06:28.160
If you asked OpenAI, that issue is resolved.
00:06:28.240 --> 00:06:29.839
It's gone, it's forgotten.
00:06:30.079 --> 00:06:34.959
All the other vendors, though, all had a really long hey, by the way, this is a big crisis.
00:06:35.040 --> 00:06:37.839
This is security, this is all this is doom and gloom.
00:06:37.920 --> 00:06:42.879
And oh, so it's really interesting to me to see what each of the engines actually share.
00:06:42.959 --> 00:06:44.720
So that's part of my newsletter that I share.
00:06:44.879 --> 00:06:48.480
Put it out on LinkedIn uh so people can uh uh download it.
00:06:48.560 --> 00:06:48.959
It's free.
00:06:49.040 --> 00:06:52.319
I don't charge anything for it, but um, you know, it's just kind of a that thing.
00:06:52.399 --> 00:07:01.120
And then I have my organizational assessment that I've been doing and helping organizations, you know, remove the components that will cause their AI projects to fail.
00:07:01.279 --> 00:07:05.519
Because we're seeing a lot of projects go and they have very successful pilots.
00:07:05.680 --> 00:07:14.639
I mean, it's amazing how many more successful pilots in AI to date, I think, than all of the successful pilots I've done in my career.
00:07:14.720 --> 00:07:17.439
And I planned the crap out of out of projects in the past.
00:07:17.519 --> 00:07:20.959
You know, when you're migrating international users, you got to be very careful, right?
00:07:21.040 --> 00:07:23.439
You got to spend the time, do it right.
00:07:23.680 --> 00:07:27.759
Um but AI projects go to pilot and then fail.
00:07:28.079 --> 00:07:33.199
Not after pilot, or excuse me, not before pilot, not during pilot, after pilot.
00:07:34.000 --> 00:07:38.240
Because the one flaw today is you have the human in the loop.
00:07:38.720 --> 00:07:43.279
And that human hero makes your pilot successful.
00:07:43.680 --> 00:07:53.519
And a person, you know, great pilot, or you got like let's say you got 10 power users that are really making your pilot successful, they can scale to a thousand users, no problem.
00:07:55.040 --> 00:07:58.079
Ask them to scale to 10,000 users, and your project fails.
00:07:58.319 --> 00:08:01.680
And the reason is the humans can't do it, it's not possible, it's not feasible.
00:08:01.920 --> 00:08:07.439
So we've been looking at all the different things that cause organizations to struggle as they deploy AI.
00:08:07.839 --> 00:08:12.480
Um, and the one I'll I'll leave the magic bullet that we found.
00:08:12.720 --> 00:08:14.959
Um, it's really quite simple.
00:08:15.519 --> 00:08:17.759
Clearly define success.
00:08:19.040 --> 00:08:23.439
If you know what success is, failure is not an option.
00:08:24.240 --> 00:08:25.439
That's a very good one.
00:08:25.519 --> 00:08:26.480
Yeah, that's fantastic.
00:08:26.560 --> 00:08:29.120
There's so much I want to unpack about all that, uh, Scott.
00:08:29.279 --> 00:08:37.519
So, first of all, the um the uh organizational assessments that uh that you do, sort of how do those work?
00:08:37.679 --> 00:08:38.799
How involved are they?
00:08:38.960 --> 00:08:47.039
If there's, you know, there's a lot of because what what we've found uh with our company is that there's a lot of business leaders out there who want to do something with AI.
00:08:47.120 --> 00:08:50.799
They know they need to do something, they know it can add a lot of value to their organization.
00:08:50.960 --> 00:08:58.159
They're not sure because of the mixed media, you know, doom and gloom or or or pie in the sky, whatever it is, but they're not sure.
00:08:58.240 --> 00:09:02.159
Um they're not most of them don't really know what's possible with AI.
00:09:02.399 --> 00:09:06.399
So walk us through how the organizational assessment works.
00:09:06.639 --> 00:09:07.919
How much time does it take?
00:09:08.080 --> 00:09:10.159
What's sort of what's the in investment in that?
00:09:10.240 --> 00:09:11.519
Like, is it just like an hour?
00:09:11.679 --> 00:09:15.440
Is it a whole you know, 20-page uh survey they have to fill out?
00:09:15.600 --> 00:09:17.840
How do they how does that work?
00:09:18.000 --> 00:09:20.399
Um, what is the outcome of that?
00:09:20.639 --> 00:09:25.519
And then how does that lead into like a successful something that they're gonna do with AI?
00:09:26.000 --> 00:09:27.200
Well, I I gotta do the Dr.
00:09:27.360 --> 00:09:27.679
Evil.
00:09:27.919 --> 00:09:29.759
It costs one million dollars.
00:09:31.039 --> 00:09:31.519
Yeah, right.
00:09:31.679 --> 00:09:32.480
Got gotta do that.
00:09:32.559 --> 00:09:34.720
That you you asked the the cost, right?
00:09:34.799 --> 00:09:38.320
So they're they're there right now, there are two flavors of the assessment.
00:09:38.480 --> 00:09:41.600
One is a light assessment, and we have two flavors of the light assessment.
00:09:41.679 --> 00:09:48.080
One is the company does all the questionnaires, does all that, submits it to us, we review it, and we give them their output.
00:09:48.159 --> 00:09:49.919
Now they're gonna get two scores.
00:09:50.000 --> 00:09:51.840
Um, both of those scores are designed.
00:09:51.919 --> 00:10:00.320
One is designed to um take a look at their AI operational space within the organization, the other is their technical space.
00:10:00.399 --> 00:10:01.840
Uh, what do they have, right?
00:10:01.919 --> 00:10:03.279
Uh what are the rules, right?
00:10:03.440 --> 00:10:11.360
If you're talking about a US federal government agency, let's say the IRS, we do not want them putting our tax returns into AI, right?
00:10:11.519 --> 00:10:19.679
Because the AI engines, not their fault, it's the way they're built, are going to use that to um, you know, to train on in the future.
00:10:19.919 --> 00:10:25.759
So a year from now, your tax returns are suddenly visible to everybody because it was data that was used to train, right?
00:10:25.919 --> 00:10:30.000
So we want to make sure that we understand what the the confines of that is.
00:10:30.080 --> 00:10:36.960
And so the light assessment, again, they can do most of the inbound or inputting of data, and then we do the reviews.
00:10:37.039 --> 00:10:40.240
That's normally between 40 and 80 hours.
00:10:40.320 --> 00:10:42.399
Um, so it's not a long process.
00:10:42.639 --> 00:10:53.120
Now, if they want to do the full assessment, that's gonna take a lot longer because we're gonna go through and we are going to get, we're gonna give you an absolute resolute score.
00:10:53.360 --> 00:10:56.480
This is what your score is, and it's on a scale.
00:10:56.559 --> 00:10:58.639
Uh, and the scale is clearly defined.
00:10:58.720 --> 00:11:03.600
That's actually defined in my book that's coming out, but it's also clearly defined in the output that you're gonna get.
00:11:03.759 --> 00:11:07.759
You'll be able to say, okay, resolutely, we can go forward.
00:11:08.000 --> 00:11:09.600
We are ready for AI.
00:11:09.759 --> 00:11:19.039
And then uh one of one of my partners uh in this group that I'm working with uh came up with a brilliant idea of doing a diagnostic, a quick kit diagnostic.
00:11:19.440 --> 00:11:21.440
So that's what we're working on right now.
00:11:21.519 --> 00:11:22.960
We're gonna release that very soon.
00:11:23.120 --> 00:11:33.279
That'll be available where you literally can go answer 20 questions, and the tool will come back and say, um, not only should you not do AI, but you probably should stop spelling it as well.
00:11:36.159 --> 00:11:37.440
There'll be a range of outcomes.
00:11:37.519 --> 00:11:40.000
Uh, hopefully, if you get that one, uh, and you know what?
00:11:40.080 --> 00:11:41.919
I think I'm gonna actually write that into the code.
00:11:42.080 --> 00:11:44.000
That's a great error response.
00:11:44.080 --> 00:11:49.360
If you you should not spell AI, in fact, you should probably put your computer away.
00:11:49.600 --> 00:12:01.360
Um, but yeah, so you'll get a range of this is this is what's gonna happen, and then we'll offer to to come in and do a light assessment, we'll offer to come in and do a full assessment, whatever's best for you.
00:12:01.519 --> 00:12:05.679
Um, and and look, it may just be that the diagnostic is enough, right?
00:12:06.799 --> 00:12:11.840
A lot of organizations, and I've worked with as you know, as you pointed out, I've worked with a lot of companies over the years.
00:12:12.080 --> 00:12:13.279
A lot of them are really smart.
00:12:13.440 --> 00:12:15.679
I mean, they're they're very, very, very smart, very skilled.
00:12:15.759 --> 00:12:19.519
They have very intelligent IT people, very intelligent leadership.
00:12:19.759 --> 00:12:25.600
Um they're gonna um they're gonna they're gonna help you uh be successful.
00:12:25.759 --> 00:12:41.519
Um, and then in the light assessment and the full assessment, uh, we're working right now to add in some new stuff uh from uh so there's a great uh podcast that I listened to um from MIT Sloan all about AI.
00:12:41.600 --> 00:12:44.960
Um and one of the things that they were talking about is a J curve.
00:12:45.120 --> 00:12:46.720
Um and they've Matt, right?
00:12:46.879 --> 00:12:48.000
The J curve is really cool.
00:12:48.080 --> 00:12:55.200
It it basically says, you know, anytime there's a new technology, there's a J curve, which means, you know, J goes like this and then J goes down.
00:12:55.360 --> 00:12:55.679
Yeah.
00:12:55.840 --> 00:12:56.080
Right.
00:12:56.159 --> 00:13:00.080
And and I think in the AI world right now, we're in that downside, right?
00:13:00.159 --> 00:13:01.759
We are in that going down.
00:13:02.080 --> 00:13:06.559
But the thing that people always forget is the J curve is really important because then it goes up.
00:13:06.720 --> 00:13:11.440
And when it goes up, it's ahead of where it was when it started.
00:13:11.519 --> 00:13:14.240
And I do believe personally, AI is going to do that.
00:13:14.320 --> 00:13:21.120
So we're building in a uh the J curve process for people in the light and full assessment, so you can go through and look at your organization.
00:13:21.440 --> 00:13:32.799
Um, because I mean, honestly, if I hear one more CEO say, Oh, we used uh we we adopted AI and we're we laid off 10,000 people.
00:13:33.039 --> 00:13:33.440
I know.
00:13:33.519 --> 00:13:34.320
Oh my goodness.
00:13:34.559 --> 00:13:38.159
Well, yeah, what happened when an automaker did that?
00:13:38.399 --> 00:13:42.240
They laid off all their QA people because AI is gonna take it over.
00:13:42.399 --> 00:13:43.519
And what happened a month later?
00:13:43.679 --> 00:13:49.200
They had to rehire all those Q QA people back at a 10 to 15% salary bomb.
00:13:49.519 --> 00:13:51.919
Because they weren't coming back if they didn't get more money.
00:13:52.159 --> 00:13:54.000
So we have to be really careful.
00:13:54.159 --> 00:13:56.559
AI is not gonna lay people off right now.
00:13:56.720 --> 00:13:58.960
AI is not gonna replace human beings.
00:13:59.120 --> 00:14:04.720
It's actually in a situation where if we use it intelligently, it augments what we're doing, makes us better.
00:14:04.879 --> 00:14:06.399
And who doesn't need to be better?
00:14:06.559 --> 00:14:09.039
I know I do a lot better.
00:14:10.000 --> 00:14:11.759
Yeah, 100%, Scott.
00:14:11.840 --> 00:14:12.720
That's so, so true.
00:14:12.799 --> 00:14:19.039
I love, you know, you you and I are very much in sync with with uh with our uh perspective on the marketplace and with what's out there.
00:14:19.200 --> 00:14:30.879
And um yeah, I want to I want to dive into more of the um uh the the implementation side uh that you see from a um uh perspective.
00:14:31.039 --> 00:14:38.720
So the assessment that you do, the organizational assessment with the light and the faux and the diagnostic, et cetera, that's mostly a readiness assessment, right?
00:14:38.799 --> 00:14:44.240
Like how ready are you to start leaning into AI and doing the adoption of AI and so forth.
00:14:44.399 --> 00:14:53.360
So when they're when they're doing it, and you mentioned like the IRS and federal agencies, and even uh most companies have proprietary data that they don't want to be sharing with the models, right?
00:14:53.440 --> 00:14:54.559
I mean, just the way it is.
00:14:54.799 --> 00:15:01.519
Certainly, if you have uh military government contracts, anything like that that you do, uh you know, you need to be very, very careful.
00:15:01.679 --> 00:15:05.120
Um, but um so two questions.
00:15:05.279 --> 00:15:10.799
Number one, uh the work that you do, uh, what size of company does it apply to?
00:15:10.960 --> 00:15:13.519
Is it uh mid-market, large enterprise only?
00:15:13.600 --> 00:15:15.200
Is it is it all of the above?
00:15:15.440 --> 00:15:21.440
And then um what because you're a lot of experience with data infrastructure.
00:15:21.600 --> 00:15:35.200
Um, what is your um, I guess if you were speaking to a business leader who maybe doesn't understand a lot about AI, how would you describe the appropriate way for them to protect their data and keep it safe?
00:15:35.360 --> 00:15:42.159
Is it to bring a model in-house that's uh an open open source model, or is it other things?
00:15:42.320 --> 00:15:49.759
So just describe that a little bit, sort of your um because you understand the architecture of it all, and 99.9% of people do not.
00:15:49.919 --> 00:15:54.399
So I'd love to hear if you're talking to a business leader, uh, what would be your recommendation?
00:15:54.480 --> 00:15:56.639
So, what size of companies do you engage with?
00:15:56.799 --> 00:16:03.120
And then uh what would be your general recommendation for the uh proprietary level of data that people want to protect?
00:16:03.360 --> 00:16:08.080
Well, so far we've been engaging with, you know, all the way down to the bottom and all the way up to the top.
00:16:08.159 --> 00:16:14.720
Um, although we're not doing massive huge companies yet, but we will uh the assessment would work for anybody.
00:16:14.879 --> 00:16:23.759
Um, it's you know, the it I would say if you're in a large company, you probably can't use a light assessment, you probably have to do the full, uh, just because it's it's not possible.
00:16:23.919 --> 00:16:28.240
So um, but the to to kind of get to the the gist of the question, right?
00:16:28.320 --> 00:16:30.399
What what's gonna cause you to trip, right?
00:16:30.559 --> 00:16:31.919
Because that's the big thing, right?
00:16:32.240 --> 00:16:34.000
What's gonna cause you to trip?
00:16:34.240 --> 00:16:34.879
Um, right.
00:16:34.960 --> 00:16:38.000
And and I so I've been doing international migrations for years.
00:16:38.159 --> 00:16:41.600
I mean, I've done some of the biggest migrations that Microsoft ever did.
00:16:41.759 --> 00:16:43.519
Um, they're painful.
00:16:43.679 --> 00:16:47.200
Um, and I've done big federal uh deployments as well.
00:16:47.360 --> 00:16:48.879
Again, they're painful.
00:16:49.039 --> 00:16:49.840
Um, right.
00:16:50.000 --> 00:17:02.000
The very first rule I always tell uh you know company and government leaders um if you have a partner that comes in and tells you that they're going to give you a pain-free migration, throw them out the nearest window.
00:17:02.080 --> 00:17:03.200
Uh, there's no such thing.
00:17:03.360 --> 00:17:05.039
I don't care what happens, right?
00:17:05.119 --> 00:17:17.680
There's all right, there's there's always that one PC that didn't update a month ago, and that update is critical, and you deploy the software and boom, shaka locker, you are now looking at a bricked PC, right?
00:17:17.839 --> 00:17:24.720
Or worse, it's an iPhone, or you know, back in the back in the day, uh BlackBerry.
00:17:24.880 --> 00:17:25.039
Right.
00:17:25.119 --> 00:17:37.039
So so the conversation uh really from from my perspective, the conversation starts with understanding your data first and understand the security confines of that data, right?
00:17:37.519 --> 00:17:40.240
Not all data needs to be secured.
00:17:41.359 --> 00:17:44.000
Not all data is proprietary, right?
00:17:44.160 --> 00:17:51.359
Many years ago, I I worked with a bunch of folks at Microsoft and we we developed um um uh a knowledge management system.
00:17:51.440 --> 00:17:54.720
Uh we called it ICE, Intellectual Capital Exchange.
00:17:54.960 --> 00:17:57.440
Uh awesome name, right?
00:17:57.599 --> 00:18:05.200
Uh and it and when you worked on the team, uh the the director, he was a hoot, used to call everybody a freaking ice hole.
00:18:05.359 --> 00:18:07.039
Uh that's what we were.
00:18:07.119 --> 00:18:08.319
We were ice holes.
00:18:08.480 --> 00:18:12.240
Um the concept is you know, understand your data.
00:18:12.400 --> 00:18:14.880
What it what it where is it, what is it, where does it live?
00:18:15.039 --> 00:18:16.240
How do you connect to it?
00:18:16.400 --> 00:18:18.160
What is the security around it?
00:18:18.319 --> 00:18:19.519
Okay, that's the first thing.
00:18:19.759 --> 00:18:29.519
The next thing is you you really have to have a very hard conversation with what are we willing to give up and what are we not willing to give up?
00:18:29.680 --> 00:18:34.960
Because if you're if you're not going to give something up, then you have to seriously look at the open source models.
00:18:35.119 --> 00:18:40.319
Now, I I have have to be honest, I've been playing a lot with the NVIDIA stuff lately.
00:18:40.559 --> 00:18:43.279
Um, it's pretty pretty good.
00:18:43.440 --> 00:18:45.440
Um, I I'm really impressed.
00:18:45.599 --> 00:18:46.960
It works really well.
00:18:47.119 --> 00:18:53.039
Uh, you can run it on a single GPU, shockingly, that NVIDIA would write something that works on a single GPU.
00:18:53.119 --> 00:18:56.880
Uh, because I do not have a 40 GPU cluster in my house.
00:18:57.440 --> 00:19:02.960
I only I only have one big uh massive GPU uh sitting in my desktop.
00:19:03.279 --> 00:19:06.400
So uh understanding, you know, where are you going to run the model?
00:19:06.480 --> 00:19:16.000
Now, the thing that I pushed for and I continue to push for um is the reality of um doing just in time AI.
00:19:16.160 --> 00:19:19.119
Um that's kind of my thing that I talk a little bit about.
00:19:19.200 --> 00:19:25.599
I've written a number of articles about it, um, but just in time AI is where I have a small language model on a device.
00:19:25.839 --> 00:19:34.160
Um, right, I can load a, in fact, I have uh four models right now sitting on my iPhone that I can run locally.
00:19:34.319 --> 00:19:41.039
Um now they're limited because they're in an environment that's controlled, but they're effectively I'm able to connect with them and use them.
00:19:41.200 --> 00:20:01.519
So imagine a sales rep goes and right, his calendar or her calendar is visible, and the AI system looks at it and goes, Oh, you're going to uh Nematode Motors, uh Bob the Builder, um, I don't know, uh trying to think of other kids' shows, uh Teletubby Land and uh Barneyville.
00:20:01.759 --> 00:20:02.240
Yeah, right.
00:20:02.400 --> 00:20:04.480
You're going to those four customers today.
00:20:04.799 --> 00:20:05.839
You know what I'm gonna do?
00:20:06.000 --> 00:20:08.240
I'm gonna preload their last 30 orders.
00:20:08.319 --> 00:20:15.279
I'm gonna preload any customer complaints that they've had, I'm gonna preload any problems that they've had onto your phone, right?
00:20:15.359 --> 00:20:19.279
So now the sales rep gets there, the CIO goes, What the heck happened?
00:20:19.440 --> 00:20:23.519
You guys shipped me a broken widget uh three weeks ago.
00:20:23.839 --> 00:20:26.319
I got 10,000 widgets, one was broken, right?
00:20:26.640 --> 00:20:30.400
Sales rep can go, oh wait, yeah, that's actually been shipped.
00:20:30.480 --> 00:20:33.519
The replacement's on its way, it's supposed to arrive tomorrow, right?
00:20:33.759 --> 00:20:36.400
You give them the concept of just in time information.
00:20:36.480 --> 00:20:38.720
So then you have to think about right, that's the other part.
00:20:38.880 --> 00:20:40.319
How do I secure the data?
00:20:40.559 --> 00:20:42.880
But also, when do I need the data?
00:20:43.440 --> 00:20:45.119
And where do I need the data?
00:20:45.359 --> 00:20:51.839
Um, my view, my vote, cell phone is the client for AI in the future.
00:20:52.319 --> 00:21:00.720
If you think about the power of your cell phone today, I mean, there's more umph in my iPhone than there was on the uh Apollo 11 computers.
00:21:01.759 --> 00:21:03.200
And those computers were big.
00:21:03.279 --> 00:21:06.240
I mean, think about those things weighed almost 800 pounds.
00:21:06.319 --> 00:21:09.839
Uh, NASA worked really hard to get them down to 800 pounds.
00:21:10.720 --> 00:21:16.400
But, you know, my cell, my cell phone does not weigh 800 pounds, although the otter case feels like it weighs 800 pounds.
00:21:16.559 --> 00:21:18.319
Um, but yeah, so now you've got that.
00:21:18.400 --> 00:21:20.160
So you've got, you know, where is the data?
00:21:20.240 --> 00:21:21.200
How do I get to it?
00:21:21.359 --> 00:21:23.680
What security confines do I have for it?
00:21:23.839 --> 00:21:25.119
And then what do I need to do?
00:21:25.279 --> 00:21:27.119
Can I run it as a local model?
00:21:27.200 --> 00:21:29.279
Uh, what can I do with the local models?
00:21:29.440 --> 00:21:33.519
And then, of course, as you brought out, right, there is the concept of open source.
00:21:33.680 --> 00:21:37.359
Now, I project the open source model is gonna change.
00:21:37.599 --> 00:21:46.079
I don't think China's gonna be allowing um the open source companies in China to continue to release nearly state of the art models in the next year.
00:21:46.240 --> 00:21:48.240
I think they're gonna start limiting it.
00:21:48.400 --> 00:21:53.200
Uh I I think they probably should have limited it already from an economic perspective.
00:21:53.279 --> 00:21:59.039
I don't think it's a smart idea to let them loose, but I understand uh why they did it to gain credibility.
00:21:59.440 --> 00:21:59.599
It
00:22:00.160 --> 00:22:05.359
Work, they're credible, but open source models are interesting in that sense.
00:22:05.599 --> 00:22:10.400
Um I see a lot of value in considering open source models.
00:22:10.480 --> 00:22:16.720
And again, as I mentioned, the NVIDIA models really I've been unbelievably impressed with them.
00:22:16.880 --> 00:22:20.880
Um, not as impressed with Facebook's meta product, the Llama.
00:22:21.440 --> 00:22:24.880
Um, but I they've discontinued Llama basically effectively.
00:22:25.119 --> 00:22:27.680
Um and they're moving to a a pay model.
00:22:27.759 --> 00:22:34.880
But um and I also have seen the open AI open source model, which was questionable.
00:22:35.440 --> 00:22:37.200
So yeah, interesting.
00:22:37.359 --> 00:22:38.480
That's uh fascinating.
00:22:38.640 --> 00:22:45.759
So um we may not be able to answer this this whole um next question, but it's I want to ask you one more good one before we before we wind it down.
00:22:46.000 --> 00:22:54.079
Is um um so when you're in the implementation phase of it, okay, they've done the assessment, they want to do more.
00:22:54.240 --> 00:23:06.880
We're talking about some of the implementation details about this data is like really it's PII data, it's very proprietary, it's whatever, it's special data that we don't want, we don't want to leave the four walls of our of our of our building, right?
00:23:07.119 --> 00:23:10.720
Um so do you uh partner with companies?
00:23:10.880 --> 00:23:16.160
Does your company do some of that um um uh architecture type work?
00:23:16.400 --> 00:23:19.519
Because clearly it seems like where you would recommend.
00:23:19.599 --> 00:23:33.599
So say it's a say it's a mid-sized company, say they're$500 million a year and they they have a thousand, um, a thousand employees and they're you know all knowledge workers and they they do a lot of things, whatever they're doing, production or uh or uh manufacturing.
00:23:33.759 --> 00:23:37.039
Um, but just say it's that type of size company.
00:23:37.759 --> 00:23:47.039
Um is it a multi um uh uh a multi um uh multi-part solution then that you'd recommend for them?
00:23:47.119 --> 00:23:50.720
Like, all right, we're gonna do this in internal for the proprietary stuff.
00:23:50.880 --> 00:23:53.039
That's where that data gets processed with AI.
00:23:53.119 --> 00:23:57.599
So we sort of like put some, like you said, like an internal uh AI engine in there.
00:23:57.680 --> 00:24:02.880
It's not going out to any of the uh frontier models, it's not going out to the cloud, it's just staying internal.
00:24:02.960 --> 00:24:06.960
But then we can do some cloud stuff too on the outside because that's just like the easy stuff.
00:24:07.119 --> 00:24:13.200
Maybe it's like uh field service stuff that just doesn't does it, it works with more public information, uh, etc.
00:24:13.519 --> 00:24:15.279
So I guess two parts.
00:24:15.440 --> 00:24:18.240
Do you get do you do you get involved with that?
00:24:18.400 --> 00:24:20.960
Um, that type of uh implementation.
00:24:21.119 --> 00:24:24.079
And then is it typical for that sort of mid-market company?
00:24:24.160 --> 00:24:29.440
Is it typical that they would have like a multi-component uh deployment of it?
00:24:29.519 --> 00:24:34.799
Or is it just like drop this AI solution in and it it fits it all across the org?
00:24:35.119 --> 00:24:39.359
Well, I so I I haven't seen anybody go as complex as what you just described.
00:24:39.519 --> 00:24:42.960
Um, I'm seeing people, you know, I it's I'm an advocate.
00:24:43.039 --> 00:24:44.960
I'm I've been an architect for 30 years.
00:24:45.039 --> 00:24:47.039
I'm I'm an advocate of crawl, walk, run.
00:24:47.440 --> 00:24:51.920
Um, so the answer to your question is most of the organizations are still in the crawl phase.
00:24:52.000 --> 00:24:54.000
I haven't seen a lot of walks yet.
00:24:54.160 --> 00:24:57.680
Um, and a walk would be where you start deploying that more complex solution.
00:24:57.839 --> 00:25:02.000
But yeah, right now I do see um a lot of organizations doing that split.
00:25:02.240 --> 00:25:07.599
Internal, private, you know, our private proprietary data is internal in an open source model.
00:25:07.759 --> 00:25:12.240
And then uh we use the public cloud for um, you know, the broader things.
00:25:12.400 --> 00:25:15.920
Now, there is the fatal flaw of image generation.
00:25:16.079 --> 00:25:21.039
The local models don't do as well by any stretch of the imagination.
00:25:22.000 --> 00:25:25.440
OpenAI and uh Google's nanobanana do pretty well.
00:25:25.519 --> 00:25:29.680
Uh and and Google gets like, you know, they get an A plus for the name.
00:25:29.920 --> 00:25:34.400
Um Image Gen is a nice name, but nanobanana rocks.
00:25:34.480 --> 00:25:41.519
I mean, I I I use Image Gen more than I use nanobanana, but I I snicker when I use nanobanana because it's a better name.
00:25:41.759 --> 00:25:46.240
Um, so yeah, it so you want to be careful, you want to have that that uh structure split.
00:25:46.400 --> 00:25:50.160
Most organizations, as I said, are still crawl, so they're looking at that initial phase.
00:25:50.480 --> 00:25:55.680
Uh so yeah, we're gonna help them, we're gonna architect the solution, we're gonna help them get to the point where they are.
00:25:55.920 --> 00:26:04.880
Um the painful reality for most companies um is data cleansing uh and getting the data to a position where it's effective.
00:26:05.200 --> 00:26:10.640
Because of that, it's it that that becomes the sticking point uh initially.
00:26:10.880 --> 00:26:14.640
After that, yeah, we start start helping them uh pick up more.
00:26:14.799 --> 00:26:17.680
I can't wait for the first customer I have that gets to walk.
00:26:17.839 --> 00:26:22.319
Um boy, when I find somebody that's running, that's that's that's gonna make me happy.
00:26:22.559 --> 00:26:23.759
Yeah, that's exciting.
00:26:23.920 --> 00:26:34.559
I'm looking forward to talking more with you about this uh outside the podcast, too, because there's a lot of um we we have similar perspectives in the way that we're engaging with with companies, with our company, your company, et cetera.
00:26:34.640 --> 00:26:36.079
So we can talk more into it.
00:26:36.160 --> 00:26:52.960
But I'm really I I appreciate um everything that you shared because I know a lot of what you just shared um are answers to the questions that a lot of these um you know top-level executives in these mid-market and even large enterprise companies are dealing with right now because we we hear these questions all the time.
00:26:53.119 --> 00:26:56.640
That's why I asked them of you uh because you have a unique perspective on it.
00:26:56.720 --> 00:27:01.519
So I really appreciate uh your time, attention, and all the insights that you shared, Scott.
00:27:01.599 --> 00:27:09.839
I'm looking forward to having you back on the podcast and maybe doing something else that we could do together to sort of collaborate and just um communicate with the marketplace.
00:27:10.079 --> 00:27:10.480
Awesome.
00:27:10.640 --> 00:27:11.359
Sounds good to me.
00:27:11.440 --> 00:27:12.160
Thank you so much.
00:27:12.240 --> 00:27:14.079
It was a blast talking to you.
00:00:04.959 --> 00:00:16.399
Welcome back to the AI Accelerator Podcast, where we cut through the noise of AI in the world today and get to what actually works when AI works for you inside your organization.
00:00:16.559 --> 00:00:21.199
I'm Matt Sembratsky, and today my guest is Scott Anderson.
00:00:21.600 --> 00:00:29.039
Scott spent nearly three decades as an enterprise architect at companies like that you might have heard of, like Microsoft, Lockheed Martin, Verizon.
00:00:29.199 --> 00:00:35.920
And he's also the author of two books on AI infrastructure, including his new one, designing AI data centers.
00:00:36.159 --> 00:00:37.679
Another big hot topic today.
00:00:37.840 --> 00:00:41.200
I'm very excited to have uh Scott on the podcast today.
00:00:41.280 --> 00:00:43.439
And I'm I'm really excited for many reasons.
00:00:43.679 --> 00:00:47.520
One is that a lot of leaders talk about AI as software.
00:00:47.759 --> 00:00:52.880
Scott thinks about it, about how AI actually runs, the infrastructure it runs on.
00:00:53.039 --> 00:00:54.799
That actually changes the strategy.
00:00:55.039 --> 00:01:03.840
So you're all going to have a great uh bird's eye view into the uh the mind of Scott and how AI really works inside infrastructure today.
00:01:04.000 --> 00:01:06.480
So uh Scott, welcome to the podcast.
00:01:06.879 --> 00:01:07.280
Awesome.
00:01:07.439 --> 00:01:08.000
Well, thank you.
00:01:08.079 --> 00:01:09.200
I'm I'm very excited.
00:01:09.359 --> 00:01:12.640
This is uh a topic that is near and dear to my heart.
00:01:13.760 --> 00:01:14.400
Fantastic.
00:01:14.719 --> 00:01:15.599
So let's get started.
00:01:15.760 --> 00:01:25.920
Before we before we dig in uh um into the the crazy topics we're gonna hit on, you know, fill in the gaps for people, you know, what what's your background?
00:01:26.079 --> 00:01:28.079
How did you end up being uh doing this work?
00:01:28.159 --> 00:01:39.519
Uh from what I understand, you started out in the education world and uh teaching, and then you moved into the wild world of the uh corporate world and uh digital AI infrastructure and so forth.
00:01:39.599 --> 00:01:40.560
How did you get there?
00:01:40.799 --> 00:01:52.719
Yeah, well, uh I mean, if you think about it, being a school teacher, uh you're you're having to connect to a bunch of people, you're having to build a consensus, uh, and you're having to help people move forward at whatever pace they're moving.
00:01:52.799 --> 00:01:56.079
And no one in a classroom moves at the same pace, right?
00:01:56.159 --> 00:01:59.920
You don't get 22 kids and you're like, we're ready to roll.
00:02:00.079 --> 00:02:00.719
Let's go.
00:02:00.879 --> 00:02:02.719
Let's hit it's uh zero to 60.
00:02:02.799 --> 00:02:04.079
We're at 60, let's keep going.
00:02:04.239 --> 00:02:13.039
60, you know, it's you know, you got half the classes at 15 miles an hour, half the classes at 35 miles an hour, half the classes at 75 miles an hour.
00:02:13.120 --> 00:02:13.840
Um, right?
00:02:14.000 --> 00:02:15.919
Those are the ones the cops are usually chasing.
00:02:16.159 --> 00:02:19.759
Uh but the reality is, right, it people move at different paces.
00:02:19.919 --> 00:02:22.560
Uh, and so, you know, I kind of was doing that.
00:02:22.719 --> 00:02:29.439
Um, and uh uh I met a wonderful person, um, and uh I was a tenured school teacher.
00:02:29.599 --> 00:02:38.400
Um, and I said to her, and she was literally graduating college, uh, she took the long, long plan uh to make it through her bachelor's degree.
00:02:38.479 --> 00:02:41.680
Um, and so she was graduating college and she was looking for a job.
00:02:41.759 --> 00:02:47.759
And I said, Well, if you find a job in another city that pays more than what I'm making as a tenured school teacher, I'll move.
00:02:47.919 --> 00:02:52.879
Uh and so three months later, I was living in a new city, uh, looking for a new job.
00:02:53.120 --> 00:03:03.360
And uh I got offered a job by the school system, but they wanted me to take a pay cut, a lot, a huge pay cut, like enough that it would have really impacted what I was doing.
00:03:03.520 --> 00:03:09.680
Um, and at the same time, this computer training company said to me, Hey, you're a teacher, you know computers.
00:03:09.840 --> 00:03:11.759
Why don't you come be a trainer?
00:03:12.240 --> 00:03:17.599
So I did that, started making more money, moved to help desk, and from there I kind of became a consultant.
00:03:17.680 --> 00:03:28.400
And for the last 30 years, I've been a consulting architect working on uh international, national, US DOD, US government, uh, state government, uh, local government projects.
00:03:28.479 --> 00:03:31.599
Um, but I've always evolved with the shift, right?
00:03:31.759 --> 00:03:37.520
So I was a collaboration person in the 2000 era, uh, internet, web, collaboration.
00:03:37.599 --> 00:03:42.800
Um, I moved into becoming a cloud person in the 2010 range because you know, cloud was the big thing.
00:03:42.960 --> 00:03:44.479
Uh, cloud was the new thing.
00:03:44.639 --> 00:03:53.360
Uh, and then lately I've moved to being an AI person because I personally believe AI is going to augment humans and make us better.
00:03:53.439 --> 00:03:58.159
Uh, and I refuse to uh I refuse to buy into the doom and gloom.
00:03:59.039 --> 00:03:59.919
That's amazing.
00:04:00.240 --> 00:04:01.120
That's awesome.
00:04:01.199 --> 00:04:01.599
That's great.
00:04:01.680 --> 00:04:04.960
Yeah, I I I uh echo the same sentiment there, uh Scott.
00:04:05.120 --> 00:04:05.919
Fantastic.
00:04:06.240 --> 00:04:09.439
So um that catches us up to today.
00:04:09.599 --> 00:04:22.720
So now for our curious listeners out there, the business leaders who are uh are leading all the knowledge workers, you know, the tens of millions of knowledge workers out there that are getting so much work done um with or without AI, right?
00:04:22.800 --> 00:04:24.560
Because there's a lot of adoption still to do.
00:04:24.720 --> 00:04:27.920
But um, catch everybody up to what do you do today?
00:04:28.079 --> 00:04:29.519
Who do you do it for?
00:04:29.759 --> 00:04:33.040
Um, you know, are you are you just doing one job with one company?
00:04:33.199 --> 00:04:35.920
Are you uh you know a consultant to the industry?
00:04:36.160 --> 00:04:37.759
Is it tied into data centers?
00:04:37.920 --> 00:04:41.759
Is it tied into digital infrastructure within companies, all the above?
00:04:41.839 --> 00:04:47.040
Uh, there's a lot for your background, you're suit, you're you're highly qualified for a lot of different directions.
00:04:47.199 --> 00:04:50.480
So, what's the short version of what you're doing today and who you're doing it for?
00:04:50.720 --> 00:04:58.079
Well, I'm I'm working on a contracting gig, working for a company doing uh RFP responses on the federal government side.
00:04:58.160 --> 00:05:02.560
Uh, and then I'm also working with another organization and we're doing uh AI assessments.
00:05:02.639 --> 00:05:05.600
So I'm I'm actually building uh so I have a book.
00:05:05.680 --> 00:05:11.360
You you mentioned the two books that are in the market now, Infrastructure for Designing AI Networks and Designing AI data centers.
00:05:11.600 --> 00:05:12.800
I have a third book.
00:05:12.879 --> 00:05:17.040
Um, it's called Organizational Debt or Why Your AI Project Fails.
00:05:17.199 --> 00:05:19.199
Uh, the publisher just optioned it.
00:05:19.279 --> 00:05:25.040
I literally just finished all the paperwork that the publisher sent me, um, and they have accepted it.
00:05:25.120 --> 00:05:31.279
Uh, and now they're gonna go through the hard part for me, which is they're gonna edit everything, send it back to me, and I have to clean it up.
00:05:31.360 --> 00:05:37.360
Uh, but in that book is a concept of an organizational assessment, and that's what I've been really focused on.
00:05:37.439 --> 00:05:46.639
Uh, but I also do an AI newsletter um for uh about 200 of my closest friends, um, and then constantly growing, uh, trying to get information out.
00:05:46.800 --> 00:05:49.439
Now, I do AI news differently than most people do.
00:05:49.519 --> 00:05:54.959
I actually go to the AI engines and I ask them, tell me what the led, what's the biggest news from last week?
00:05:55.040 --> 00:06:01.920
Uh and then I ask each one, and I I literally paste into the uh newsletter format that one of my friends designed for me.
00:06:02.000 --> 00:06:03.920
Um, you know, I paste it in.
00:06:04.000 --> 00:06:12.319
So you see what Claude says, you see what OpenAI says, you see what Meta says, uh, and then I do analysis because it's interesting, right?
00:06:12.560 --> 00:06:21.920
Last week OpenAI had two agents that went rogue and attacked Hugging Face like 116,000 times in a minute.
00:06:22.000 --> 00:06:24.079
Uh, but just a massive thing, right?
00:06:24.959 --> 00:06:28.160
If you asked OpenAI, that issue is resolved.
00:06:28.240 --> 00:06:29.839
It's gone, it's forgotten.
00:06:30.079 --> 00:06:34.959
All the other vendors, though, all had a really long hey, by the way, this is a big crisis.
00:06:35.040 --> 00:06:37.839
This is security, this is all this is doom and gloom.
00:06:37.920 --> 00:06:42.879
And oh, so it's really interesting to me to see what each of the engines actually share.
00:06:42.959 --> 00:06:44.720
So that's part of my newsletter that I share.
00:06:44.879 --> 00:06:48.480
Put it out on LinkedIn uh so people can uh uh download it.
00:06:48.560 --> 00:06:48.959
It's free.
00:06:49.040 --> 00:06:52.319
I don't charge anything for it, but um, you know, it's just kind of a that thing.
00:06:52.399 --> 00:07:01.120
And then I have my organizational assessment that I've been doing and helping organizations, you know, remove the components that will cause their AI projects to fail.
00:07:01.279 --> 00:07:05.519
Because we're seeing a lot of projects go and they have very successful pilots.
00:07:05.680 --> 00:07:14.639
I mean, it's amazing how many more successful pilots in AI to date, I think, than all of the successful pilots I've done in my career.
00:07:14.720 --> 00:07:17.439
And I planned the crap out of out of projects in the past.
00:07:17.519 --> 00:07:20.959
You know, when you're migrating international users, you got to be very careful, right?
00:07:21.040 --> 00:07:23.439
You got to spend the time, do it right.
00:07:23.680 --> 00:07:27.759
Um but AI projects go to pilot and then fail.
00:07:28.079 --> 00:07:33.199
Not after pilot, or excuse me, not before pilot, not during pilot, after pilot.
00:07:34.000 --> 00:07:38.240
Because the one flaw today is you have the human in the loop.
00:07:38.720 --> 00:07:43.279
And that human hero makes your pilot successful.
00:07:43.680 --> 00:07:53.519
And a person, you know, great pilot, or you got like let's say you got 10 power users that are really making your pilot successful, they can scale to a thousand users, no problem.
00:07:55.040 --> 00:07:58.079
Ask them to scale to 10,000 users, and your project fails.
00:07:58.319 --> 00:08:01.680
And the reason is the humans can't do it, it's not possible, it's not feasible.
00:08:01.920 --> 00:08:07.439
So we've been looking at all the different things that cause organizations to struggle as they deploy AI.
00:08:07.839 --> 00:08:12.480
Um, and the one I'll I'll leave the magic bullet that we found.
00:08:12.720 --> 00:08:14.959
Um, it's really quite simple.
00:08:15.519 --> 00:08:17.759
Clearly define success.
00:08:19.040 --> 00:08:23.439
If you know what success is, failure is not an option.
00:08:24.240 --> 00:08:25.439
That's a very good one.
00:08:25.519 --> 00:08:26.480
Yeah, that's fantastic.
00:08:26.560 --> 00:08:29.120
There's so much I want to unpack about all that, uh, Scott.
00:08:29.279 --> 00:08:37.519
So, first of all, the um the uh organizational assessments that uh that you do, sort of how do those work?
00:08:37.679 --> 00:08:38.799
How involved are they?
00:08:38.960 --> 00:08:47.039
If there's, you know, there's a lot of because what what we've found uh with our company is that there's a lot of business leaders out there who want to do something with AI.
00:08:47.120 --> 00:08:50.799
They know they need to do something, they know it can add a lot of value to their organization.
00:08:50.960 --> 00:08:58.159
They're not sure because of the mixed media, you know, doom and gloom or or or pie in the sky, whatever it is, but they're not sure.
00:08:58.240 --> 00:09:02.159
Um they're not most of them don't really know what's possible with AI.
00:09:02.399 --> 00:09:06.399
So walk us through how the organizational assessment works.
00:09:06.639 --> 00:09:07.919
How much time does it take?
00:09:08.080 --> 00:09:10.159
What's sort of what's the in investment in that?
00:09:10.240 --> 00:09:11.519
Like, is it just like an hour?
00:09:11.679 --> 00:09:15.440
Is it a whole you know, 20-page uh survey they have to fill out?
00:09:15.600 --> 00:09:17.840
How do they how does that work?
00:09:18.000 --> 00:09:20.399
Um, what is the outcome of that?
00:09:20.639 --> 00:09:25.519
And then how does that lead into like a successful something that they're gonna do with AI?
00:09:26.000 --> 00:09:27.200
Well, I I gotta do the Dr.
00:09:27.360 --> 00:09:27.679
Evil.
00:09:27.919 --> 00:09:29.759
It costs one million dollars.
00:09:31.039 --> 00:09:31.519
Yeah, right.
00:09:31.679 --> 00:09:32.480
Got gotta do that.
00:09:32.559 --> 00:09:34.720
That you you asked the the cost, right?
00:09:34.799 --> 00:09:38.320
So they're they're there right now, there are two flavors of the assessment.
00:09:38.480 --> 00:09:41.600
One is a light assessment, and we have two flavors of the light assessment.
00:09:41.679 --> 00:09:48.080
One is the company does all the questionnaires, does all that, submits it to us, we review it, and we give them their output.
00:09:48.159 --> 00:09:49.919
Now they're gonna get two scores.
00:09:50.000 --> 00:09:51.840
Um, both of those scores are designed.
00:09:51.919 --> 00:10:00.320
One is designed to um take a look at their AI operational space within the organization, the other is their technical space.
00:10:00.399 --> 00:10:01.840
Uh, what do they have, right?
00:10:01.919 --> 00:10:03.279
Uh what are the rules, right?
00:10:03.440 --> 00:10:11.360
If you're talking about a US federal government agency, let's say the IRS, we do not want them putting our tax returns into AI, right?
00:10:11.519 --> 00:10:19.679
Because the AI engines, not their fault, it's the way they're built, are going to use that to um, you know, to train on in the future.
00:10:19.919 --> 00:10:25.759
So a year from now, your tax returns are suddenly visible to everybody because it was data that was used to train, right?
00:10:25.919 --> 00:10:30.000
So we want to make sure that we understand what the the confines of that is.
00:10:30.080 --> 00:10:36.960
And so the light assessment, again, they can do most of the inbound or inputting of data, and then we do the reviews.
00:10:37.039 --> 00:10:40.240
That's normally between 40 and 80 hours.
00:10:40.320 --> 00:10:42.399
Um, so it's not a long process.
00:10:42.639 --> 00:10:53.120
Now, if they want to do the full assessment, that's gonna take a lot longer because we're gonna go through and we are going to get, we're gonna give you an absolute resolute score.
00:10:53.360 --> 00:10:56.480
This is what your score is, and it's on a scale.
00:10:56.559 --> 00:10:58.639
Uh, and the scale is clearly defined.
00:10:58.720 --> 00:11:03.600
That's actually defined in my book that's coming out, but it's also clearly defined in the output that you're gonna get.
00:11:03.759 --> 00:11:07.759
You'll be able to say, okay, resolutely, we can go forward.
00:11:08.000 --> 00:11:09.600
We are ready for AI.
00:11:09.759 --> 00:11:19.039
And then uh one of one of my partners uh in this group that I'm working with uh came up with a brilliant idea of doing a diagnostic, a quick kit diagnostic.
00:11:19.440 --> 00:11:21.440
So that's what we're working on right now.
00:11:21.519 --> 00:11:22.960
We're gonna release that very soon.
00:11:23.120 --> 00:11:33.279
That'll be available where you literally can go answer 20 questions, and the tool will come back and say, um, not only should you not do AI, but you probably should stop spelling it as well.
00:11:36.159 --> 00:11:37.440
There'll be a range of outcomes.
00:11:37.519 --> 00:11:40.000
Uh, hopefully, if you get that one, uh, and you know what?
00:11:40.080 --> 00:11:41.919
I think I'm gonna actually write that into the code.
00:11:42.080 --> 00:11:44.000
That's a great error response.
00:11:44.080 --> 00:11:49.360
If you you should not spell AI, in fact, you should probably put your computer away.
00:11:49.600 --> 00:12:01.360
Um, but yeah, so you'll get a range of this is this is what's gonna happen, and then we'll offer to to come in and do a light assessment, we'll offer to come in and do a full assessment, whatever's best for you.
00:12:01.519 --> 00:12:05.679
Um, and and look, it may just be that the diagnostic is enough, right?
00:12:06.799 --> 00:12:11.840
A lot of organizations, and I've worked with as you know, as you pointed out, I've worked with a lot of companies over the years.
00:12:12.080 --> 00:12:13.279
A lot of them are really smart.
00:12:13.440 --> 00:12:15.679
I mean, they're they're very, very, very smart, very skilled.
00:12:15.759 --> 00:12:19.519
They have very intelligent IT people, very intelligent leadership.
00:12:19.759 --> 00:12:25.600
Um they're gonna um they're gonna they're gonna help you uh be successful.
00:12:25.759 --> 00:12:41.519
Um, and then in the light assessment and the full assessment, uh, we're working right now to add in some new stuff uh from uh so there's a great uh podcast that I listened to um from MIT Sloan all about AI.
00:12:41.600 --> 00:12:44.960
Um and one of the things that they were talking about is a J curve.
00:12:45.120 --> 00:12:46.720
Um and they've Matt, right?
00:12:46.879 --> 00:12:48.000
The J curve is really cool.
00:12:48.080 --> 00:12:55.200
It it basically says, you know, anytime there's a new technology, there's a J curve, which means, you know, J goes like this and then J goes down.
00:12:55.360 --> 00:12:55.679
Yeah.
00:12:55.840 --> 00:12:56.080
Right.
00:12:56.159 --> 00:13:00.080
And and I think in the AI world right now, we're in that downside, right?
00:13:00.159 --> 00:13:01.759
We are in that going down.
00:13:02.080 --> 00:13:06.559
But the thing that people always forget is the J curve is really important because then it goes up.
00:13:06.720 --> 00:13:11.440
And when it goes up, it's ahead of where it was when it started.
00:13:11.519 --> 00:13:14.240
And I do believe personally, AI is going to do that.
00:13:14.320 --> 00:13:21.120
So we're building in a uh the J curve process for people in the light and full assessment, so you can go through and look at your organization.
00:13:21.440 --> 00:13:32.799
Um, because I mean, honestly, if I hear one more CEO say, Oh, we used uh we we adopted AI and we're we laid off 10,000 people.
00:13:33.039 --> 00:13:33.440
I know.
00:13:33.519 --> 00:13:34.320
Oh my goodness.
00:13:34.559 --> 00:13:38.159
Well, yeah, what happened when an automaker did that?
00:13:38.399 --> 00:13:42.240
They laid off all their QA people because AI is gonna take it over.
00:13:42.399 --> 00:13:43.519
And what happened a month later?
00:13:43.679 --> 00:13:49.200
They had to rehire all those Q QA people back at a 10 to 15% salary bomb.
00:13:49.519 --> 00:13:51.919
Because they weren't coming back if they didn't get more money.
00:13:52.159 --> 00:13:54.000
So we have to be really careful.
00:13:54.159 --> 00:13:56.559
AI is not gonna lay people off right now.
00:13:56.720 --> 00:13:58.960
AI is not gonna replace human beings.
00:13:59.120 --> 00:14:04.720
It's actually in a situation where if we use it intelligently, it augments what we're doing, makes us better.
00:14:04.879 --> 00:14:06.399
And who doesn't need to be better?
00:14:06.559 --> 00:14:09.039
I know I do a lot better.
00:14:10.000 --> 00:14:11.759
Yeah, 100%, Scott.
00:14:11.840 --> 00:14:12.720
That's so, so true.
00:14:12.799 --> 00:14:19.039
I love, you know, you you and I are very much in sync with with uh with our uh perspective on the marketplace and with what's out there.
00:14:19.200 --> 00:14:30.879
And um yeah, I want to I want to dive into more of the um uh the the implementation side uh that you see from a um uh perspective.
00:14:31.039 --> 00:14:38.720
So the assessment that you do, the organizational assessment with the light and the faux and the diagnostic, et cetera, that's mostly a readiness assessment, right?
00:14:38.799 --> 00:14:44.240
Like how ready are you to start leaning into AI and doing the adoption of AI and so forth.
00:14:44.399 --> 00:14:53.360
So when they're when they're doing it, and you mentioned like the IRS and federal agencies, and even uh most companies have proprietary data that they don't want to be sharing with the models, right?
00:14:53.440 --> 00:14:54.559
I mean, just the way it is.
00:14:54.799 --> 00:15:01.519
Certainly, if you have uh military government contracts, anything like that that you do, uh you know, you need to be very, very careful.
00:15:01.679 --> 00:15:05.120
Um, but um so two questions.
00:15:05.279 --> 00:15:10.799
Number one, uh the work that you do, uh, what size of company does it apply to?
00:15:10.960 --> 00:15:13.519
Is it uh mid-market, large enterprise only?
00:15:13.600 --> 00:15:15.200
Is it is it all of the above?
00:15:15.440 --> 00:15:21.440
And then um what because you're a lot of experience with data infrastructure.
00:15:21.600 --> 00:15:35.200
Um, what is your um, I guess if you were speaking to a business leader who maybe doesn't understand a lot about AI, how would you describe the appropriate way for them to protect their data and keep it safe?
00:15:35.360 --> 00:15:42.159
Is it to bring a model in-house that's uh an open open source model, or is it other things?
00:15:42.320 --> 00:15:49.759
So just describe that a little bit, sort of your um because you understand the architecture of it all, and 99.9% of people do not.
00:15:49.919 --> 00:15:54.399
So I'd love to hear if you're talking to a business leader, uh, what would be your recommendation?
00:15:54.480 --> 00:15:56.639
So, what size of companies do you engage with?
00:15:56.799 --> 00:16:03.120
And then uh what would be your general recommendation for the uh proprietary level of data that people want to protect?
00:16:03.360 --> 00:16:08.080
Well, so far we've been engaging with, you know, all the way down to the bottom and all the way up to the top.
00:16:08.159 --> 00:16:14.720
Um, although we're not doing massive huge companies yet, but we will uh the assessment would work for anybody.
00:16:14.879 --> 00:16:23.759
Um, it's you know, the it I would say if you're in a large company, you probably can't use a light assessment, you probably have to do the full, uh, just because it's it's not possible.
00:16:23.919 --> 00:16:28.240
So um, but the to to kind of get to the the gist of the question, right?
00:16:28.320 --> 00:16:30.399
What what's gonna cause you to trip, right?
00:16:30.559 --> 00:16:31.919
Because that's the big thing, right?
00:16:32.240 --> 00:16:34.000
What's gonna cause you to trip?
00:16:34.240 --> 00:16:34.879
Um, right.
00:16:34.960 --> 00:16:38.000
And and I so I've been doing international migrations for years.
00:16:38.159 --> 00:16:41.600
I mean, I've done some of the biggest migrations that Microsoft ever did.
00:16:41.759 --> 00:16:43.519
Um, they're painful.
00:16:43.679 --> 00:16:47.200
Um, and I've done big federal uh deployments as well.
00:16:47.360 --> 00:16:48.879
Again, they're painful.
00:16:49.039 --> 00:16:49.840
Um, right.
00:16:50.000 --> 00:17:02.000
The very first rule I always tell uh you know company and government leaders um if you have a partner that comes in and tells you that they're going to give you a pain-free migration, throw them out the nearest window.
00:17:02.080 --> 00:17:03.200
Uh, there's no such thing.
00:17:03.360 --> 00:17:05.039
I don't care what happens, right?
00:17:05.119 --> 00:17:17.680
There's all right, there's there's always that one PC that didn't update a month ago, and that update is critical, and you deploy the software and boom, shaka locker, you are now looking at a bricked PC, right?
00:17:17.839 --> 00:17:24.720
Or worse, it's an iPhone, or you know, back in the back in the day, uh BlackBerry.
00:17:24.880 --> 00:17:25.039
Right.
00:17:25.119 --> 00:17:37.039
So so the conversation uh really from from my perspective, the conversation starts with understanding your data first and understand the security confines of that data, right?
00:17:37.519 --> 00:17:40.240
Not all data needs to be secured.
00:17:41.359 --> 00:17:44.000
Not all data is proprietary, right?
00:17:44.160 --> 00:17:51.359
Many years ago, I I worked with a bunch of folks at Microsoft and we we developed um um uh a knowledge management system.
00:17:51.440 --> 00:17:54.720
Uh we called it ICE, Intellectual Capital Exchange.
00:17:54.960 --> 00:17:57.440
Uh awesome name, right?
00:17:57.599 --> 00:18:05.200
Uh and it and when you worked on the team, uh the the director, he was a hoot, used to call everybody a freaking ice hole.
00:18:05.359 --> 00:18:07.039
Uh that's what we were.
00:18:07.119 --> 00:18:08.319
We were ice holes.
00:18:08.480 --> 00:18:12.240
Um the concept is you know, understand your data.
00:18:12.400 --> 00:18:14.880
What it what it where is it, what is it, where does it live?
00:18:15.039 --> 00:18:16.240
How do you connect to it?
00:18:16.400 --> 00:18:18.160
What is the security around it?
00:18:18.319 --> 00:18:19.519
Okay, that's the first thing.
00:18:19.759 --> 00:18:29.519
The next thing is you you really have to have a very hard conversation with what are we willing to give up and what are we not willing to give up?
00:18:29.680 --> 00:18:34.960
Because if you're if you're not going to give something up, then you have to seriously look at the open source models.
00:18:35.119 --> 00:18:40.319
Now, I I have have to be honest, I've been playing a lot with the NVIDIA stuff lately.
00:18:40.559 --> 00:18:43.279
Um, it's pretty pretty good.
00:18:43.440 --> 00:18:45.440
Um, I I'm really impressed.
00:18:45.599 --> 00:18:46.960
It works really well.
00:18:47.119 --> 00:18:53.039
Uh, you can run it on a single GPU, shockingly, that NVIDIA would write something that works on a single GPU.
00:18:53.119 --> 00:18:56.880
Uh, because I do not have a 40 GPU cluster in my house.
00:18:57.440 --> 00:19:02.960
I only I only have one big uh massive GPU uh sitting in my desktop.
00:19:03.279 --> 00:19:06.400
So uh understanding, you know, where are you going to run the model?
00:19:06.480 --> 00:19:16.000
Now, the thing that I pushed for and I continue to push for um is the reality of um doing just in time AI.
00:19:16.160 --> 00:19:19.119
Um that's kind of my thing that I talk a little bit about.
00:19:19.200 --> 00:19:25.599
I've written a number of articles about it, um, but just in time AI is where I have a small language model on a device.
00:19:25.839 --> 00:19:34.160
Um, right, I can load a, in fact, I have uh four models right now sitting on my iPhone that I can run locally.
00:19:34.319 --> 00:19:41.039
Um now they're limited because they're in an environment that's controlled, but they're effectively I'm able to connect with them and use them.
00:19:41.200 --> 00:20:01.519
So imagine a sales rep goes and right, his calendar or her calendar is visible, and the AI system looks at it and goes, Oh, you're going to uh Nematode Motors, uh Bob the Builder, um, I don't know, uh trying to think of other kids' shows, uh Teletubby Land and uh Barneyville.
00:20:01.759 --> 00:20:02.240
Yeah, right.
00:20:02.400 --> 00:20:04.480
You're going to those four customers today.
00:20:04.799 --> 00:20:05.839
You know what I'm gonna do?
00:20:06.000 --> 00:20:08.240
I'm gonna preload their last 30 orders.
00:20:08.319 --> 00:20:15.279
I'm gonna preload any customer complaints that they've had, I'm gonna preload any problems that they've had onto your phone, right?
00:20:15.359 --> 00:20:19.279
So now the sales rep gets there, the CIO goes, What the heck happened?
00:20:19.440 --> 00:20:23.519
You guys shipped me a broken widget uh three weeks ago.
00:20:23.839 --> 00:20:26.319
I got 10,000 widgets, one was broken, right?
00:20:26.640 --> 00:20:30.400
Sales rep can go, oh wait, yeah, that's actually been shipped.
00:20:30.480 --> 00:20:33.519
The replacement's on its way, it's supposed to arrive tomorrow, right?
00:20:33.759 --> 00:20:36.400
You give them the concept of just in time information.
00:20:36.480 --> 00:20:38.720
So then you have to think about right, that's the other part.
00:20:38.880 --> 00:20:40.319
How do I secure the data?
00:20:40.559 --> 00:20:42.880
But also, when do I need the data?
00:20:43.440 --> 00:20:45.119
And where do I need the data?
00:20:45.359 --> 00:20:51.839
Um, my view, my vote, cell phone is the client for AI in the future.
00:20:52.319 --> 00:21:00.720
If you think about the power of your cell phone today, I mean, there's more umph in my iPhone than there was on the uh Apollo 11 computers.
00:21:01.759 --> 00:21:03.200
And those computers were big.
00:21:03.279 --> 00:21:06.240
I mean, think about those things weighed almost 800 pounds.
00:21:06.319 --> 00:21:09.839
Uh, NASA worked really hard to get them down to 800 pounds.
00:21:10.720 --> 00:21:16.400
But, you know, my cell, my cell phone does not weigh 800 pounds, although the otter case feels like it weighs 800 pounds.
00:21:16.559 --> 00:21:18.319
Um, but yeah, so now you've got that.
00:21:18.400 --> 00:21:20.160
So you've got, you know, where is the data?
00:21:20.240 --> 00:21:21.200
How do I get to it?
00:21:21.359 --> 00:21:23.680
What security confines do I have for it?
00:21:23.839 --> 00:21:25.119
And then what do I need to do?
00:21:25.279 --> 00:21:27.119
Can I run it as a local model?
00:21:27.200 --> 00:21:29.279
Uh, what can I do with the local models?
00:21:29.440 --> 00:21:33.519
And then, of course, as you brought out, right, there is the concept of open source.
00:21:33.680 --> 00:21:37.359
Now, I project the open source model is gonna change.
00:21:37.599 --> 00:21:46.079
I don't think China's gonna be allowing um the open source companies in China to continue to release nearly state of the art models in the next year.
00:21:46.240 --> 00:21:48.240
I think they're gonna start limiting it.
00:21:48.400 --> 00:21:53.200
Uh I I think they probably should have limited it already from an economic perspective.
00:21:53.279 --> 00:21:59.039
I don't think it's a smart idea to let them loose, but I understand uh why they did it to gain credibility.
00:21:59.440 --> 00:21:59.599
It
00:22:00.160 --> 00:22:05.359
Work, they're credible, but open source models are interesting in that sense.
00:22:05.599 --> 00:22:10.400
Um I see a lot of value in considering open source models.
00:22:10.480 --> 00:22:16.720
And again, as I mentioned, the NVIDIA models really I've been unbelievably impressed with them.
00:22:16.880 --> 00:22:20.880
Um, not as impressed with Facebook's meta product, the Llama.
00:22:21.440 --> 00:22:24.880
Um, but I they've discontinued Llama basically effectively.
00:22:25.119 --> 00:22:27.680
Um and they're moving to a a pay model.
00:22:27.759 --> 00:22:34.880
But um and I also have seen the open AI open source model, which was questionable.
00:22:35.440 --> 00:22:37.200
So yeah, interesting.
00:22:37.359 --> 00:22:38.480
That's uh fascinating.
00:22:38.640 --> 00:22:45.759
So um we may not be able to answer this this whole um next question, but it's I want to ask you one more good one before we before we wind it down.
00:22:46.000 --> 00:22:54.079
Is um um so when you're in the implementation phase of it, okay, they've done the assessment, they want to do more.
00:22:54.240 --> 00:23:06.880
We're talking about some of the implementation details about this data is like really it's PII data, it's very proprietary, it's whatever, it's special data that we don't want, we don't want to leave the four walls of our of our of our building, right?
00:23:07.119 --> 00:23:10.720
Um so do you uh partner with companies?
00:23:10.880 --> 00:23:16.160
Does your company do some of that um um uh architecture type work?
00:23:16.400 --> 00:23:19.519
Because clearly it seems like where you would recommend.
00:23:19.599 --> 00:23:33.599
So say it's a say it's a mid-sized company, say they're$500 million a year and they they have a thousand, um, a thousand employees and they're you know all knowledge workers and they they do a lot of things, whatever they're doing, production or uh or uh manufacturing.
00:23:33.759 --> 00:23:37.039
Um, but just say it's that type of size company.
00:23:37.759 --> 00:23:47.039
Um is it a multi um uh uh a multi um uh multi-part solution then that you'd recommend for them?
00:23:47.119 --> 00:23:50.720
Like, all right, we're gonna do this in internal for the proprietary stuff.
00:23:50.880 --> 00:23:53.039
That's where that data gets processed with AI.
00:23:53.119 --> 00:23:57.599
So we sort of like put some, like you said, like an internal uh AI engine in there.
00:23:57.680 --> 00:24:02.880
It's not going out to any of the uh frontier models, it's not going out to the cloud, it's just staying internal.
00:24:02.960 --> 00:24:06.960
But then we can do some cloud stuff too on the outside because that's just like the easy stuff.
00:24:07.119 --> 00:24:13.200
Maybe it's like uh field service stuff that just doesn't does it, it works with more public information, uh, etc.
00:24:13.519 --> 00:24:15.279
So I guess two parts.
00:24:15.440 --> 00:24:18.240
Do you get do you do you get involved with that?
00:24:18.400 --> 00:24:20.960
Um, that type of uh implementation.
00:24:21.119 --> 00:24:24.079
And then is it typical for that sort of mid-market company?
00:24:24.160 --> 00:24:29.440
Is it typical that they would have like a multi-component uh deployment of it?
00:24:29.519 --> 00:24:34.799
Or is it just like drop this AI solution in and it it fits it all across the org?
00:24:35.119 --> 00:24:39.359
Well, I so I I haven't seen anybody go as complex as what you just described.
00:24:39.519 --> 00:24:42.960
Um, I'm seeing people, you know, I it's I'm an advocate.
00:24:43.039 --> 00:24:44.960
I'm I've been an architect for 30 years.
00:24:45.039 --> 00:24:47.039
I'm I'm an advocate of crawl, walk, run.
00:24:47.440 --> 00:24:51.920
Um, so the answer to your question is most of the organizations are still in the crawl phase.
00:24:52.000 --> 00:24:54.000
I haven't seen a lot of walks yet.
00:24:54.160 --> 00:24:57.680
Um, and a walk would be where you start deploying that more complex solution.
00:24:57.839 --> 00:25:02.000
But yeah, right now I do see um a lot of organizations doing that split.
00:25:02.240 --> 00:25:07.599
Internal, private, you know, our private proprietary data is internal in an open source model.
00:25:07.759 --> 00:25:12.240
And then uh we use the public cloud for um, you know, the broader things.
00:25:12.400 --> 00:25:15.920
Now, there is the fatal flaw of image generation.
00:25:16.079 --> 00:25:21.039
The local models don't do as well by any stretch of the imagination.
00:25:22.000 --> 00:25:25.440
OpenAI and uh Google's nanobanana do pretty well.
00:25:25.519 --> 00:25:29.680
Uh and and Google gets like, you know, they get an A plus for the name.
00:25:29.920 --> 00:25:34.400
Um Image Gen is a nice name, but nanobanana rocks.
00:25:34.480 --> 00:25:41.519
I mean, I I I use Image Gen more than I use nanobanana, but I I snicker when I use nanobanana because it's a better name.
00:25:41.759 --> 00:25:46.240
Um, so yeah, it so you want to be careful, you want to have that that uh structure split.
00:25:46.400 --> 00:25:50.160
Most organizations, as I said, are still crawl, so they're looking at that initial phase.
00:25:50.480 --> 00:25:55.680
Uh so yeah, we're gonna help them, we're gonna architect the solution, we're gonna help them get to the point where they are.
00:25:55.920 --> 00:26:04.880
Um the painful reality for most companies um is data cleansing uh and getting the data to a position where it's effective.
00:26:05.200 --> 00:26:10.640
Because of that, it's it that that becomes the sticking point uh initially.
00:26:10.880 --> 00:26:14.640
After that, yeah, we start start helping them uh pick up more.
00:26:14.799 --> 00:26:17.680
I can't wait for the first customer I have that gets to walk.
00:26:17.839 --> 00:26:22.319
Um boy, when I find somebody that's running, that's that's that's gonna make me happy.
00:26:22.559 --> 00:26:23.759
Yeah, that's exciting.
00:26:23.920 --> 00:26:34.559
I'm looking forward to talking more with you about this uh outside the podcast, too, because there's a lot of um we we have similar perspectives in the way that we're engaging with with companies, with our company, your company, et cetera.
00:26:34.640 --> 00:26:36.079
So we can talk more into it.
00:26:36.160 --> 00:26:52.960
But I'm really I I appreciate um everything that you shared because I know a lot of what you just shared um are answers to the questions that a lot of these um you know top-level executives in these mid-market and even large enterprise companies are dealing with right now because we we hear these questions all the time.
00:26:53.119 --> 00:26:56.640
That's why I asked them of you uh because you have a unique perspective on it.
00:26:56.720 --> 00:27:01.519
So I really appreciate uh your time, attention, and all the insights that you shared, Scott.
00:27:01.599 --> 00:27:09.839
I'm looking forward to having you back on the podcast and maybe doing something else that we could do together to sort of collaborate and just um communicate with the marketplace.
00:27:10.079 --> 00:27:10.480
Awesome.
00:27:10.640 --> 00:27:11.359
Sounds good to me.
00:27:11.440 --> 00:27:12.160
Thank you so much.
00:27:12.240 --> 00:27:14.079
It was a blast talking to you.
