رونوشت 🔗
WEBVTT
00:00:01.112 --> 00:00:05.785
Hi everyone, welcome to figure out fabric where I ask dumb questions and you get smart answers.
00:00:05.785 --> 00:00:08.928
I've got our, I think where you are inaugural guests.
00:00:08.928 --> 00:00:10.368
I feel like it.
00:00:12.550 --> 00:00:13.009
Okay.
00:00:13.009 --> 00:00:18.053
Um, I've got one of our earliest guests back on for uh a reappraisal.
00:00:18.053 --> 00:00:22.536
Um, and we're going to talk about people's least favorite subject, which is AI.
00:00:22.536 --> 00:00:24.047
I know everyone's sick about it.
00:00:24.047 --> 00:00:26.239
So disclaimer now.
00:00:26.379 --> 00:00:26.940
Yeah.
00:00:26.940 --> 00:00:29.486
And this will be, this is going to be coming out.
00:00:29.486 --> 00:00:39.411
uh a few episodes after our episode with Kurt Bueller on AI, but we're going to be um a bit more dystopian in this one.
00:00:39.411 --> 00:00:45.354
So we're going to be talking about how you're more likely the product than you are the consumer with this stuff.
00:00:45.354 --> 00:00:56.908
um Kellan, um she's got decades of experience with Oracle, Microsoft SQL Server, but now she's an AI advocate at Redgate.
00:00:56.908 --> 00:01:00.572
Which I assume means that he's trying to usher in the apocalypse, right?
00:01:00.572 --> 00:01:06.326
So like when Mad Max happens, she's going to be the guy with guitar.
00:01:06.326 --> 00:01:06.787
Huh?
00:01:06.787 --> 00:01:13.954
oh! So this is more like Terminator where you're trying to, oh, I'm sorry.
00:01:13.954 --> 00:01:17.493
I thought you were like the rock and roll guy in Mad Max Period.
00:01:17.493 --> 00:01:19.774
No, I'm Linda Hamilton with the gun.
00:01:19.774 --> 00:01:20.293
Yeah.
00:01:20.293 --> 00:01:21.233
Bigger.
00:01:21.493 --> 00:01:22.353
Yeah.
00:01:22.367 --> 00:01:26.317
It's good to pick a side because the clankers aren't going to take prisoners.
00:01:26.317 --> 00:01:28.551
So you got to know what side you're on.
00:01:29.019 --> 00:01:31.109
I don't, am I Linda Hamilton though?
00:01:31.109 --> 00:01:33.070
Because I love AI.
00:01:33.070 --> 00:01:34.861
mean, let's be honest.
00:01:34.861 --> 00:01:37.384
We like jump past, past the introductions.
00:01:37.384 --> 00:01:38.524
I just went right into it.
00:01:38.524 --> 00:01:39.385
Didn't I?
00:01:39.385 --> 00:01:39.805
Yeah.
00:01:39.805 --> 00:01:44.828
m Who doesn't know me?
00:01:44.828 --> 00:01:45.227
Yeah.
00:01:45.227 --> 00:01:59.097
Because I had this weird epiphany as I was doing a keynote at an Oracle event that I'm now better known on the Microsoft side than the Oracle side, which was really surreal to realize that because I still do Oracle a lot.
00:01:59.097 --> 00:02:00.198
mean, I really do.
00:02:00.198 --> 00:02:07.840
um But that recognition and realizing that most of the Oracle community is just like, well, that's Callan.
00:02:08.181 --> 00:02:11.341
Because there's no women in the Oracle space.
00:02:11.341 --> 00:02:13.942
You'll never escape if you're a woman in Oracle.
00:02:13.942 --> 00:02:21.294
um But that I hang around mostly Microsoft and open source and DevOps events and stuff like that.
00:02:21.294 --> 00:02:25.175
And all my friends are kind of, well, I shouldn't say all my friends.
00:02:25.175 --> 00:02:26.543
I have a lot of friends.
00:02:26.543 --> 00:02:30.336
But my close friends, a lot of them are in the Microsoft community.
00:02:30.336 --> 00:02:32.157
They might peeps, you know.
00:02:32.337 --> 00:02:38.082
But, you know, I've been in the space for, I'm getting up on 30 years.
00:02:38.223 --> 00:02:49.112
I get to come into like customer engagements and go, yeah, I've been in tech longer than like Huxley's been alive and he's their solution engineer.
00:02:49.112 --> 00:02:51.354
So it's getting to that point in my life.
00:02:51.354 --> 00:02:52.856
That's where I'm at.
00:02:52.856 --> 00:02:53.856
I know it.
00:02:54.270 --> 00:02:59.194
We're all feeling our age, but I work with Steve Jones and Graeme Fritchie, know Pat Wright.
00:03:00.437 --> 00:03:01.618
They are lovely.
00:03:01.618 --> 00:03:02.820
They are wonderful.
00:03:02.820 --> 00:03:03.181
Yeah.
00:03:03.181 --> 00:03:04.092
balls.
00:03:04.094 --> 00:03:05.266
I love them both.
00:03:05.348 --> 00:03:08.689
You should, you absolutely should, because I never stop.
00:03:08.689 --> 00:03:17.854
Anybody who went to the past summit keynote should have figured out that Laura and my job up on stage was just to see if we could trip up Steve the whole way through.
00:03:17.854 --> 00:03:22.056
And we got Grant to just be a total goofball, which was awesome.
00:03:22.056 --> 00:03:23.316
Awesome, awesome.
00:03:23.336 --> 00:03:26.617
But yeah, I mean, I love my job at Red Gate.
00:03:26.617 --> 00:03:30.419
I am very, very involved with the engineers.
00:03:30.500 --> 00:03:34.531
I love that Grant tells people that I like to do the hard stuff.
00:03:34.768 --> 00:03:38.949
But I'm happiest when I'm deep into code and technology.
00:03:38.949 --> 00:03:39.818
I am.
00:03:39.818 --> 00:03:40.750
I need that.
00:03:40.750 --> 00:03:43.319
If I don't have that, I'm very unhappy.
00:03:43.319 --> 00:03:44.520
So I do a lot of coding.
00:03:44.520 --> 00:03:52.431
um I think I opened up two new GitHub repositories just as last week for my solution engineers and building out some scripts.
00:03:52.431 --> 00:03:54.223
And yeah, there was Python.
00:03:54.223 --> 00:03:55.723
There was T-SQL.
00:03:55.723 --> 00:03:58.704
There was uh PL SQL.
00:03:58.704 --> 00:04:01.063
There was Postgres SQL.
00:04:01.932 --> 00:04:02.782
Mm-hmm.
00:04:03.217 --> 00:04:04.819
There was stuff for everything in there.
00:04:04.819 --> 00:04:08.564
So whatever anybody needs, I just become a Swiss Army knife and do it.
00:04:09.355 --> 00:04:10.286
Yeah.
00:04:10.526 --> 00:04:10.817
Yeah.
00:04:10.817 --> 00:04:19.449
And I think it, you know, talking about AI and talking about your, your history and your role, um,
Load full transcript
I'm going to get his name wrong.
00:04:19.449 --> 00:04:22.100
Um, I think he's, I to say he's Polish.
00:04:22.100 --> 00:04:25.492
There's a guy, it's like gurgly ursas or something like that.
00:04:25.492 --> 00:04:30.095
He runs the pragmatic engineer, um, newsletter.
00:04:30.095 --> 00:04:30.954
It's a great newsletter.
00:04:30.954 --> 00:04:31.754
I pay for it.
00:04:31.754 --> 00:04:32.656
It's worth it.
00:04:32.656 --> 00:04:38.497
Um, but there's free versions and he's talking about, what does this new world potentially look like?
00:04:39.074 --> 00:04:49.040
And it's interesting how the skill sets that are valued are going to change because your skill set of multi platform used to be like hot.
00:04:49.040 --> 00:04:50.129
I guess it's my own podcast.
00:04:50.129 --> 00:04:52.021
I can swear it used to be hot shit.
00:04:52.141 --> 00:04:57.954
Used to be a hot commodity where like it was it was very rare to get people can be multi platform.
00:04:57.954 --> 00:05:04.507
But now with AI, you know, it's easier to learn and jump from one language to another.
00:05:04.968 --> 00:05:07.949
But the skills that are still going to matter.
00:05:08.242 --> 00:05:16.408
are all of the engineering and software engineering, understanding DevOps, understanding Git, understanding architectural stuff.
00:05:16.408 --> 00:05:19.850
And so it's interesting how you just gave your background.
00:05:20.610 --> 00:05:24.012
A fifth of that is now worthless.
00:05:24.012 --> 00:05:24.853
You know what I mean?
00:05:24.853 --> 00:05:39.238
Like I am working on a project to transpile power query from M to pandas to be able to move stuff from data flows to notebooks, right?
00:05:39.579 --> 00:05:45.483
But I could not do that at all if I didn't start learning about TDD DevOps, all this stuff.
00:05:45.483 --> 00:05:57.740
So it's interesting to hear how like the world is changing and the skills that you have, some are going to become way more valuable and some are going to become way less valuable.
00:05:58.889 --> 00:06:06.213
I'm gonna actually take that a little bit with a grain of salt because I don't know if they will lose their value completely.
00:06:06.213 --> 00:06:18.225
um My husband and I joke about that, you before Oracle got the Stargate project, before they became part of the AI bubble, okay?
00:06:18.225 --> 00:06:20.112
What are they doing?
00:06:20.336 --> 00:06:21.701
Go ahead, sorry.
00:06:21.701 --> 00:06:22.740
we can talk about that.
00:06:22.740 --> 00:06:23.560
Yeah.
00:06:23.560 --> 00:06:30.860
But before that happened, you know, I told Microsoft back in 2021, they're coming for the AI projects and like, no, they aren't.
00:06:30.860 --> 00:06:33.880
And then they won Stargate and they're like, Kellan, we need to talk.
00:06:33.880 --> 00:06:38.360
like, told you one second, let me go cash out on my Oracle stock.
00:06:38.360 --> 00:06:39.420
I'll be right back.
00:06:39.420 --> 00:06:40.661
You know, one of those.
00:06:41.081
--> 00:06:45.600
But I mean, when people would say this and they're like, oh, you're Oracle.
00:06:45.600 --> 00:06:46.800
And it's like, you know what?
00:06:46.800 --> 00:06:52.858
Somebody's going to bring me back to life in 3020 and my husband and go, We hear you know Oracle.
00:06:53.959 --> 00:06:59.723
It's just the Cobalt, it's the Pascal, the Fortran of its decade.
00:07:00.024 --> 00:07:03.788
Everybody who says, I have all these skills, they're worthless.
00:07:03.788 --> 00:07:16.000
They are not worthless because there is so much value in that data that it resides on and that data that is using some of that skill set that somewhere, somehow somebody's going to say, it ain't broke, I ain't going to fix it.
00:07:16.000 --> 00:07:17.666
I'm just going to keep using it.
00:07:17.666 --> 00:07:23.745
And then you are going to be just worth millions and millions of dollars because you're the only one who's going to be able to work on it.
00:07:24.848 --> 00:07:26.048
Let's be honest.
00:07:27.060 --> 00:07:28.201
Yeah, for sure.
00:07:28.321 --> 00:07:29.101
that's where I am.
00:07:29.101 --> 00:07:31.862
As much as people go, I got to learn this new AI thing.
00:07:31.862 --> 00:07:33.803
I got to learn that AI thing.
00:07:33.803 --> 00:07:38.716
I have a tendency to hold myself back and stay with some of the trusted stuff.
00:07:38.716 --> 00:07:40.547
Somebody says, we're going to do this.
00:07:40.547 --> 00:07:46.209
like right now, I'm working with ASACPRA collectors with Python.
00:07:46.209 --> 00:07:47.750
Yeah, I'll do that.
00:07:47.750 --> 00:07:48.930
I'll do that.
00:07:49.230 --> 00:07:51.151
Python, gotcha.
00:07:51.151 --> 00:07:56.391
But if somebody said, we've got this brand new language, and I'm not even going to try to guess one right now.
00:07:56.391 --> 00:07:58.322
I don't want to insult anyone because I will.
00:07:58.322 --> 00:07:59.874
is a new language I've heard of.
00:07:59.874 --> 00:08:02.577
think it's for like writing GPU kernels.
00:08:04.197 --> 00:08:07.119
somebody says that, I'm probably going to hold myself back.
00:08:07.119 --> 00:08:11.071
I'm probably going to say, I'm not going to waste brain cells there and time there.
00:08:11.071 --> 00:08:14.163
I'm very much against wasting my time.
00:08:14.163 --> 00:08:28.346
That might be cool, it may be the cool new thing, but I need to put my brain on something that I think I'm going to be able to retain and use for a long I've got a perfect example of that and where AI is helping me.
00:08:28.346 --> 00:08:29.687
let me find.
00:08:29.766 --> 00:08:31.887
So I have an Android phone.
00:08:31.908 --> 00:08:33.628
Java is a nightmare.
00:08:33.628 --> 00:08:35.956
Kotlin is a language built on Java.
00:08:35.956 --> 00:08:36.985
I'm sorry.
00:08:37.109 --> 00:08:38.230
Sorry.
00:08:38.466 --> 00:08:46.221
No offense, but, I've, I've always wanted all these personal apps and I've written some before AI.
00:08:46.221 --> 00:08:54.166
Like I, I'm going to write a blog post at some point about Eugene quest, which was this dumb Python flask and angular one app.
00:08:54.166 --> 00:09:04.630
I wrote to try and manage my life and it like worked for a while and then fell apart and then I kept using the exercise part like this tiny little seed out of this like rotten husk or whatever.
00:09:04.630 --> 00:09:09.712
But with AI, I can make an app for help me and my husband manage our time.
00:09:09.712 --> 00:09:14.534
And so we've got this YouTube list of channels that we watch and other stuff.
00:09:14.534 --> 00:09:19.496
And like, I gain no personal value from learning Kotlin.
00:09:19.496 --> 00:09:21.638
I'm never going to be a mobile dev.
00:09:21.950 --> 00:09:23.832
I never want to deal with Java.
00:09:23.832 --> 00:09:27.076
I AI helped me write a Minecraft mod in Java.
00:09:27.076 --> 00:09:34.876
And so I'm incentivized to learn just enough to help guide things, but it's never going to be my core competency.
00:09:34.876 --> 00:09:37.480
And so it's a waste of my time to learn Kotlin.
00:09:38.460 --> 00:09:39.629
I agree with you completely.
00:09:39.629 --> 00:09:48.618
I love Python, but did I know how to work with like some of this stuff that I've just been working with, the prawn stuff?
00:09:48.618 --> 00:09:53.072
I know I had never worked with it before, but I knew what to ask.
00:09:53.072 --> 00:10:07.769
I knew how to write all the main code and then I could go and ask questions of chat, GBT or Claude and get the answers that I needed to fill out all the missing pieces to complete a task that somebody was like, can we do this?
00:10:07.769 --> 00:10:08.840
Is this possible, Kellyn?
00:10:08.840 --> 00:10:10.052
Could you do this?
00:10:10.052 --> 00:10:11.192
Let me see.
00:10:11.774 --> 00:10:12.995
It makes a huge difference.
00:10:12.995 --> 00:10:14.778
It is a great collaborator.
00:10:14.778 --> 00:10:19.303
And I think for somebody like myself that's easily distracted, it collaborates.
00:10:19.303 --> 00:10:23.168
When I want to collaborate, I don't have to go look for somebody and wait for them to be free.
00:10:23.168 --> 00:10:25.352
And it's wonderful.
00:10:25.352 --> 00:10:26.263
So yeah.
00:10:27.206 --> 00:10:28.147
Yeah.
00:10:28.246 --> 00:10:33.941
So one of the things that you wanted to talk though about is some of the more sinister side of this.
00:10:33.941 --> 00:10:55.361
So like I said, you know, we're going to be talking a little bit about the dystopia and one of the, you know, notes here I have is about how a lot of times we are not the consumer of these things, especially if you're paying for free, but even if you're, even if you're paying money in chat, GPT Claude, I haven't looked into Gemini.
00:10:55.563 --> 00:11:02.916
If you're paying the $20, yeah, you have to go in and tick a thing to turn off help improve our models.
00:11:03.456 --> 00:11:05.157
Well, help improve our models.
00:11:05.157 --> 00:11:06.917
They never tell you what it
means.
00:11:06.917 --> 00:11:11.158
Like I have not been able to find conclusive and they know what it means.
00:11:11.177 --> 00:11:31.041
So theoretically they could train on your data and as data professionals, like it's very important to understand that as the models get bigger and more training, technically the risk of data exfiltration becomes smaller, but not zero.
00:11:31.100 --> 00:11:40.745
And during the chat GPT 3.5 days, researchers were able to exfiltrate verbatim copies of the New York Times articles.
00:11:41.065 --> 00:11:51.989
There, they found some exploit where if you just say the same word a thousand times or tell it to say the same word a thousand times, it exfiltrated someone's real phone number, personal phone number.
00:11:52.690 --> 00:11:57.409
And so you and open AI has been through multiple lawsuits.
00:11:57.409 --> 00:12:10.750
I think there was a judge judgment like this week, where they had to provide like 30 million logs of user entries, because of I don't know if it was a New York Times lawsuit or a different one.
00:12:10.750 --> 00:12:12.370
I forgot I just saw the headline.
00:12:12.750 --> 00:12:23.250
And so anything that you put into a consumer LLM, you should treat as if it's Same same way with Google searches, right?
00:12:23.250 --> 00:12:26.741
Like, They can get a warrant, they can look up your Google searches.
00:12:26.741 --> 00:12:34.277
So if you want to make a pipe bomb, get a local LLM and ask it or go on DuckDuckGo or something.
00:12:35.118 --> 00:12:40.321
No, no, no, theoretically, if you're hypothetically.
00:12:40.782 --> 00:12:49.408
And I don't think I don't think people get like, I think people are just taking queries from work data from work and just shoving it in there.
00:12:49.408 --> 00:12:51.328
And that it terrifies me.
00:12:52.169 --> 00:12:52.544
Yeah.
00:12:52.544 --> 00:13:10.711
you know, I've been talking about this for a couple of years now in different keynotes and that, you know, I, I started doing this like hall of AI shame, you know, where I would come up and go, and I had dealt with this firsthand with a customer engagement that isn't with my previous companies.
00:13:10.711 --> 00:13:16.163
I want to add that, um, where somebody had come to me and said, Kellan, can you advise our company on AI?
00:13:16.163 --> 00:13:17.115
And I said, no problem.
00:13:17.115 --> 00:13:18.596
It was one of my first gigs.
00:13:18.596 --> 00:13:20.197
Didn't think too much of it.
00:13:20.219 --> 00:13:26.283
The users wanted to help the IT group is what had happened.
00:13:26.283 --> 00:13:27.924
And they said, look what we did.
00:13:27.924 --> 00:13:34.250
We had free chat GPT and we uploaded our entire loan customer base.
00:13:34.250 --> 00:13:35.371
Loans.
00:13:35.552 --> 00:13:37.232
Think about it, loans.
00:13:38.274 --> 00:13:39.905
Into the free version of chat GPT.
00:13:39.905 --> 00:13:43.639
And I said, you need to notify your customers that you've had a breach.
00:13:43.639 --> 00:13:45.725
That you have created a breach.
00:13:45.725 --> 00:13:46.448
yeah.
00:13:46.448 --> 00:13:48.849
And they were like, no, no, what are you talking about?
00:13:48.849 --> 00:13:52.548
We built out all of these trends and reports out of chat GPT from this.
00:13:52.548 --> 00:13:56.208
This is really cool because they were saying they couldn't build it for for six months.
00:13:56.208 --> 00:13:58.928
We got all of that out of chat GPT in just a matter of a week.
00:13:58.928 --> 00:14:05.208
And I'm like, you uploaded all of loan data, all that PII into the free version of chat GPT.
00:14:05.208 --> 00:14:07.328
And it wasn't just them.
00:14:07.328 --> 00:14:13.469
They've already, there's a couple of different ones, different companies that have done this, these and.
00:14:14.075 --> 00:14:17.003
It'll come to me as we're talking what the actual companies are.
00:14:17.003 --> 00:14:21.947
with Samsung engineers lost their jobs because they had like trade secret stuff.
00:14:22.369 --> 00:14:23.330
Right.
00:14:23.373 --> 00:14:25.033
roadmap was uploaded.
00:14:25.033 --> 00:14:28.013
Yes, I use this as one of my examples too.
00:14:28.072 --> 00:14:33.552
And they were asking it, know, what do you see, what should we do with all these roadmap visions, everything else?
00:14:33.552 --> 00:14:35.773
And they didn't know it was publicly searchable.
00:14:35.773 --> 00:14:38.192
So it's constant.
00:14:38.432 --> 00:14:44.293
And they right now are saying, I haven't looked at the latest quarter, but last quarter it was 26%.
00:14:44.293 --> 00:14:51.964
A free version of chat GBT was PII, authorization codes, names, address, socials, security numbers, all this stuff.
00:14:51.964 --> 00:14:57.696
And now you have this huge challenge of a lot of Americans with the challenge that we have with healthcare.
00:14:57.696 --> 00:15:03.999
Going to chat GPT and asking it healthcare questions and uploading their actual medical history.
00:15:03.999 --> 00:15:06.480
Yes, and this is all available now.
00:15:06.480 --> 00:15:13.623
And it's like people are so worried about HIPAA and signing off and saying, you won't share my medical history, but it's right there.
00:15:13.623 --> 00:15:18.115
So the question is, is when AI will start being hacked by?
00:15:18.520 --> 00:15:25.124
instead of hackers using AI to hack medical institutions and healthcare organizations and stuff.
00:15:25.124 --> 00:15:29.785
So it's really interesting that people are not taught what AI is.
00:15:29.785 --> 00:15:37.580
It's just this magic box of AI instead of what generative AI is, you know, agentic AI is and all these different forms of it.
00:15:37.580 --> 00:15:44.836
And what's the difference between an AI application and a large language model and small language models and.
00:15:44.836 --> 00:15:48.740
What are ephemeral sessions and what is isolation?
00:15:48.899 --> 00:15:51.302
Nobody's explaining this to folks.
00:15:51.302 -->
00:15:56.626
And so folks are, it's just, it's a pension for disaster.
00:15:56.626 --> 00:16:00.669
That's what I'm seeing with AI instead of teaching people how to do this.
00:16:00.669 --> 00:16:11.437
And many of these big organizations, IT organizations that do have an investment in AI are perfectly happy with it because it's more data that they can use.
00:16:12.158 --> 00:16:18.173
And so as I was starting to search this, I'm like, So we're starting to look at that AI bubble.
00:16:18.173 --> 00:16:20.452
We're starting to wonder, is there really an AI bubble?
00:16:20.452 --> 00:16:32.633
And yes, there is revenue value that's already being gained from AI, even though OpenAI says they won't have any positive revenue until 2028, I think they're at, now 2030.
00:16:32.793 --> 00:16:33.206
Yeah.
00:16:33.206 --> 00:16:34.826
I will, I will bet you not.
00:16:34.826 --> 00:16:37.125
I don't think you, I think you'd take the same side as me.
00:16:37.125 --> 00:16:42.426
I would gladly bet $100 that by the end of this year, they're gonna have ads.
00:16:42.426 --> 00:16:44.365
They're just gonna.
00:16:45.225 --> 00:16:45.806
Yeah.
00:16:45.806 --> 00:16:49.745
Oh, you think they're gonna sell people's data like Verizon?
00:16:49.745 --> 00:16:51.225
Oh my God.
00:16:51.225 --> 00:16:53.029
I didn't even think about that.
00:16:53.101 --> 00:16:56.142
I was looking at the data as a whole, one, the cross investment.
00:16:56.142 --> 00:16:57.822
There's this like a hundred billion dollars.
00:16:57.822 --> 00:16:58.863
It just keeps going around.
00:16:58.863 --> 00:17:01.173
All these companies go and look, I just got a hundred billion dollars.
00:17:01.173 --> 00:17:03.024
You're like, no, that was their a hundred billion.
00:17:03.024 --> 00:17:03.764
I gave it to him.
00:17:03.764 --> 00:17:04.934
Now I've got a hundred billion.
00:17:04.934 --> 00:17:05.884
No, that was his.
00:17:05.884 --> 00:17:07.223
That's the same money.
00:17:07.223 --> 00:17:13.676
You people, know, and you're seeing this cross investment and then it's like, where is this weakest link?
00:17:13.676 --> 00:17:20.593
And I was starting to research into all this and I'm like, well, how much is AI really worth each individual?
00:17:20.593 --> 00:17:23.373
that they because you think about it, what is AI learning on?
00:17:23.373 --> 00:17:30.553
It's learning on data, your data that you put out in the web and Google's what Google is training on is a little different than what open AI is.
00:17:30.553 --> 00:17:36.093
There's overlap, but there's a little different to a Google has a little bit of a.
00:17:36.093 --> 00:17:41.313
An uptick on this and what they can train on versus what open AI is getting.
00:17:41.313 --> 00:17:42.633
Yeah, all of it.
00:17:42.633 --> 00:17:45.472
I mean it just Google searches in the Google web.
00:17:45.472 --> 00:17:47.093
You know everybody Google it.
00:17:47.093 --> 00:17:48.297
There you go.
00:17:48.337 --> 00:17:51.038
All of this stuff that's out there, Gmail, all of it.
00:17:51.038 --> 00:17:53.278
And so all of this stuff that you can train on.
00:17:53.278 --> 00:17:57.077
And so I said, what is each human being on the planet worth?
00:17:57.077 --> 00:18:01.438
And then you have to look at how much of the population is in different areas.
00:18:01.837 --> 00:18:07.137
China has in India have 48 % of the entire population of the entire world.
00:18:07.137 --> 00:18:08.458
I think it's actually larger than that.
00:18:08.458 --> 00:18:13.857
My numbers are coming up wrong here, but the North America is only 7.5.
00:18:13.978 --> 00:18:15.498
Europe is 7.8.
00:18:15.498 --> 00:18:17.765
So those are small numbers.
00:18:17.765 --> 00:18:20.445
that they're actually able to get profit from.
00:18:20.506 --> 00:18:43.760
Each individual as I started to go through it, and I'm doing this research using AI to dig into this, I had to correct the prompt as it went through and some of the results that came back, there were hallucinations in it, and realizing that it was undervaluing women, wasn't considering how much buying power women have, how much actual information women were providing.
00:18:43.760 --> 00:18:53.820
on children, on families, everything else that's all of a sudden contributing into AI, that it went anywhere from $200 to about $900 a day.
00:18:53.820 --> 00:18:56.941
That's what each or year that each person was worth.
00:18:56.941 --> 00:19:01.601
And that would end up being about $16,000 $19,000 in your lifetime.
00:19:01.601 --> 00:19:03.500
That's what each person was done.
00:19:03.500 --> 00:19:11.684
And that ended up coming up to $2.6 to $4.9 trillion annually in AI revenue.
00:19:11.684 --> 00:19:14.404
that could be spread across all of these companies.
00:19:14.404 --> 00:19:18.988
Well, if you look at how much they're in debt in their investment, there's a bubble.
00:19:18.988 --> 00:19:24.109
There's definitely a bubble because remember how much of that percentage is sitting in China and India.
00:19:24.230 --> 00:19:28.752
So I started looking who's the weakest link in this and OpenAI is a huge weak link.
00:19:28.752 --> 00:19:31.354
They're doing well, but there's a weak link there.
00:19:31.354 --> 00:19:33.795
But Oracle's a big weak link.
00:19:33.795 --> 00:19:38.017
And it's one of those links that's in there that everybody's like, why?
00:19:38.085 --> 00:19:38.929
Why?
00:19:38.929 --> 00:19:44.608
Yes, they had the infrastructure and they purchased a lot of the GPUs back in 2021.
00:19:44.729 --> 00:19:49.288
But what it really is about is the investment into the Fed.
00:19:49.949 --> 00:19:52.909
That their biggest customer is the DoD.
00:19:53.308 --> 00:20:02.518
And because it is their biggest customer, if Oracle was to go under, Guess who's going to have to go bail it out?
00:20:02.919 --> 00:20:03.979
The Fed.
00:20:04.819 --> 00:20:10.943
And so the Fed is probably going to do, we're going to have to have a federal bailout
of AI if the bubble bursts.
00:20:10.943 --> 00:20:14.967
That's what's going to happen because of all the cross investment and everything else.
00:20:14.987 --> 00:20:17.888
So you recognize why they were brought in and the value of this.
00:20:17.888 --> 00:20:20.221
And it's not that Larry Ellison can't do stuff.
00:20:20.221 --> 00:20:22.092
He can, absolutely can.
00:20:22.092 --> 00:20:23.363
Are they going to screw some stuff?
00:20:23.363 --> 00:20:23.952
Yes.
00:20:23.952 --> 00:20:29.553
Who will probably get Stargate if Oracle does end up in a bailout situation?
00:20:29.553 --> 00:20:31.134
It's probably gonna be Google.
00:20:31.714 --> 00:20:33.836
Google is where they think it will actually go.
00:20:33.836 --> 00:20:36.138
It's not gonna go to Microsoft, which surprised me.
00:20:36.138 --> 00:20:37.339
Who's the most insulated?
00:20:37.339 --> 00:20:38.461
Microsoft.
00:20:38.461 --> 00:20:42.723
Microsoft and AWS is the dark horse people aren't thinking about.
00:20:43.005 --> 00:20:48.910
And there's some of the stuff that's coming up from Alexa, which is surprising people right now.
00:20:48.910 --> 00:20:57.866
I mean, there's new Alexa products that look almost like a Google search and everything and Siri that is surprising folks right now.
00:20:57.881 --> 00:21:09.250
So there's a lot of problems with this whole AI bubble and you realize there is limited revenue that can come from it and there's going to be winners and losers and how this investment is going to go.
00:21:09.250 --> 00:21:17.156
And will AI even look like what it looks like today in 2030, which many data scientists don't think it's going to.
00:21:17.156 --> 00:21:22.441
What we use for large language models now may be completely out the door and we may be onto something completely different.
00:21:22.500 --> 00:21:28.605
that they will embed all of the data directly into the GPUs and they will bypass large language models completely.
00:21:28.765 --> 00:21:31.866
So there's, I mean, it's a really weird time.
00:21:31.948 --> 00:21:41.054
But as I was doing all this research and the keynote that I was coming up with all this calculations, I said, I'm about to go down this rabbit hole.
00:21:41.054 --> 00:21:43.796
You're coming with me and I'm having this talk with folks.
00:21:43.955 --> 00:21:51.351
All of a sudden, because I was doing this in cloud and chat GPT and I was working through all these numbers and I pages and pages of research as I was going through it.
00:21:51.471 --> 00:21:56.759
All of a sudden, and I don't remember which one, came back and said, and then you, Kellan Gorman.
00:21:58.470 --> 00:22:04.575
You know, and it comes back and it says, do you know how much you are worth because you're worth more?
00:22:05.155 --> 00:22:06.413
What do you mean I'm worth more?
00:22:06.413 --> 00:22:25.133
And it says, because of your contributions, because of everything that you've written, your books, all of this that can populate and be pulled in your blog posts, your speaking engagements, your video, all of this, this actually makes you six to 10 times more valuable than the average human being to AI.
00:22:25.133 --> 00:22:31.778
uh and it ended up valuing me instead of 16 to 19 thousand dollars in a lifetime.
00:22:31.778 --> 00:22:33.980
I'm actually worth millions of dollars.
00:22:34.122 --> 00:22:39.318
And it joked, it came back and joked and said, you may want to get somebody to invest in you.
00:22:40.781 --> 00:22:41.938
That's funny.
00:22:41.938 --> 00:22:46.601
AI is thinking about this, which means AI is telling AI investors this.
00:22:46.781 --> 00:22:47.632
You know what I'm saying?
00:22:47.632 --> 00:22:54.468
That this is all part of that, that they're looking at people and this is how it's investing.
00:22:54.468 --> 00:22:58.590
And of course, people are thinking, how can I rake in the money from this?
00:22:58.590 --> 00:22:59.741
Who is more valuable?
00:22:59.741 --> 00:23:06.027
And you're going to have classes of people that are more valuable inside AI that's going to be targeted.
00:23:07.586 --> 00:23:09.490
Yeah, and dystopian future.
00:23:09.490 --> 00:23:10.785
How do you like that?
00:23:11.263 --> 00:23:12.855
You know, it, checks out.
00:23:12.855 --> 00:23:22.241
mean, I think a thing of the audience right now, I a couple of questions that are probably on their mind is like, Eugene, I thought this was a fabric podcast, which I'll get to.
00:23:22.241 --> 00:23:26.464
Um, and two, like, you know, how does this affect me?
00:23:26.464 --> 00:23:27.414
And three, what can I do?
00:23:27.414 --> 00:23:35.090
And so some things off the top of my head, I don't know if you saw this came out like yesterday, the day before Microsoft acquired a company called Osmos.
00:23:35.090 --> 00:23:41.861
Um, and so And then there was a news article today that they're planning more layoffs, right?
00:23:41.861 --> 00:23:44.071
And so Microsoft is still full steam ahead.
00:23:44.071 --> 00:23:50.473
Now I will say I'm happy to see them doing some of the agentic stuff.
00:23:50.673 --> 00:24:03.897
Some reviews on Reddit did not give me a lot of uh optimism about Osmos, but I think one of the things that you see with Microsoft compared to like the AI labs, I thought this was really insightful.
00:24:04.342 --> 00:24:17.568
A lot of the AI labs are so bought into the idea that AI is going to keep getting better that to them productizing around limitations is a waste of time because it's just going to get better.
00:24:18.109 --> 00:24:26.147
And a lot of the co-pilot stuff I think was productized around the limitations of like chat GPT-4 and 4.0.
00:24:26.147 --> 00:24:29.766
It was basically like, what if we have chat GPT-4 forever?
00:24:29.766 --> 00:24:32.156
Well, how can we integrate that with everything?
00:24:32.458 --> 00:24:35.589
And now we're running into the situation.
00:24:35.589 --> 00:24:41.609
Like I've been high on Claude code for the past three weeks
and peep.
00:24:41.609 --> 00:24:48.009
I don't think people are going to want a copilot in every product because it's all isolated and can't talk to each other.
00:24:48.009 --> 00:24:52.009
Claude code's great because you don't have the software as a service silos.
00:24:52.009 --> 00:25:03.566
You have a file system and you have the bash command line and it knows from decades of stolen scripts and books that you wrote that it stole.
00:25:03.566 --> 00:25:06.425
that it knows how to use that stuff.
00:25:06.425 --> 00:25:12.685
And so I'm glad to see Microsoft wising up to the agentic piece.
00:25:13.145 --> 00:25:30.006
I'm, not excited that they're spending all this money on AI and not like fixing the thousand paper cuts, but you know, it's like you said, Microsoft has a future in this and the, the, we're playing a money game of musical chairs.
00:25:30.006 --> 00:25:32.705
And at some point everyone's going to have to sit down.
00:25:32.895 --> 00:25:39.265
And there's half as many chairs as there are AI companies and it's going to be real bad when it happens.
00:25:39.586 --> 00:25:50.220
Now to give you some good news on that, Satya has been talking a lot about 2026 being the year of justifying AI.
00:25:50.640 --> 00:26:01.653
Because there was such a push for years and this started way back in like 2019 where people were going, if you want your bonus this year, and this is back when I was at Microsoft, it made total sense.
00:26:01.653 --> 00:26:03.124
This is really cool.
00:26:03.124 --> 00:26:06.943
So everybody, if you want your bonus, you better be putting AI in your product.
00:26:07.585 --> 00:26:08.593
Think of...
00:26:08.593 --> 00:26:11.173
Early AI is technical debt.
00:26:11.752 --> 00:26:14.492
Now somebody has to go in and clean that all up.
00:26:15.212 --> 00:26:18.512
There are a lot of AI features inside Microsoft products.
00:26:18.512 --> 00:26:19.772
It's not just Microsoft.
00:26:19.772 --> 00:26:21.573
I see it in Apple.
00:26:21.573 --> 00:26:23.053
I see it in Google.
00:26:23.053 --> 00:26:25.173
I just brought up Adobe and went.
00:26:25.173 --> 00:26:28.512
Why is this AI pop up in the way of what I want to do?
00:26:28.512 --> 00:26:30.212
You know it's everywhere.
00:26:30.732 --> 00:26:35.752
I believe Satya already knows this and he's saying it's time to justify the AI.
00:26:35.752 --> 00:26:37.911
People are going to have to go in and clean that up.
00:26:37.911 --> 00:26:50.828
And I don't think that's something that Microsoft has always been good at, but I think he's going to make it an objective for most people's annual reviews that you need to go in and now clean it up and come up with a unified AI platform.
00:26:50.828 --> 00:26:57.472
I give, you know, I'm all team Satya, so I'm hoping that he's going to be able to pull this in cohesively.
00:26:57.772 --> 00:27:04.416
Yes, there are more layoffs, but most of those layoffs aren't AI replacing you.
00:27:04.664 --> 00:27:12.288
Most of the layoffs are so they can redistribute that financial savings into investment in AI.
00:27:12.288 --> 00:27:20.711
Because as you said, there's only so many chairs and when the music stops, you got to sit down and make sure you have that investment and you're done.
00:27:20.711 --> 00:27:34.356
One of the reasons that Oracle is on my list of being a weak beam in that, know, holding up that foundation is because they still had 65 billion that they had to come up with by the end of the year and they did not.
00:27:34.628 --> 00:27:36.133
They had people starting to pull out.
00:27:36.133 --> 00:27:37.987
You saw their stock go.
00:27:39.692 --> 00:27:42.494
It was going back down after that 400 % increase.
00:27:42.494 --> 00:27:52.059
Then it started going back down and it was because people were seeing that they could not come up with that investment even though they had laid off a ton of people.
00:27:52.180 --> 00:28:00.744
All of this was redistributing their assets, redistributing all of their financial in their direction to speed up and innovate faster.
00:28:00.905 --> 00:28:05.048
That's a huge challenge on any company for them to do that.
00:28:05.048 --> 00:28:07.348
Now that's where we get into like fabric.
00:28:07.653 --> 00:28:13.236
Fabric is a product that right now was, you know, just move as fast as you can and build this thing.
00:28:13.236 --> 00:28:17.699
And now the democratization of data is saying, how do you secure that data?
00:28:17.699 --> 00:28:19.109
And that's a big deal to me.
00:28:19.109 --> 00:28:22.361
Data protection, the age of IEI is a big deal.
00:28:22.561 --> 00:28:30.046
I am seeing some beautiful features coming out from Fabric where they are actually going in and saying, we are really focused on data protection.
00:28:30.046 --> 00:28:35.141
We are now focused on how do we protect this now that we are no longer just services.
00:28:35.141 --> 00:28:37.752
We are services and infrastructure and all this stuff.
00:28:37.752 --> 00:28:48.573
As you said, that multi-platform, that multi-skill set that is required because so often we see with fabrics that nobody understood the networking and this is where it's broke.
00:28:49.013 --> 00:28:53.037
Nobody understood the storage and we're over here in services and it's broke.
00:28:53.718 --> 00:29:00.284
It's open, you and you see some of the worst AI em holes of AI shame.
00:29:00.364 --> 00:29:01.297
Some of them was.
00:29:01.297 --> 00:29:06.416
They didn't do, they didn't secure their S3 storage underneath their AI tool.
00:29:06.416 --> 00:29:19.116
Things like this again, and I always thought it was really viewed as strange when I was at Microsoft and I took our team from being data and services, what we call it, because we were separated.
00:29:19.116 --> 00:29:20.977
Your database, your data and services.
00:29:20.977 --> 00:29:24.717
This is infrastructure networking, and I pulled this over to infrastructure networking.
00:29:24.717 --> 00:29:27.537
I'm like, we're an all infrastructure play.
00:29:27.537 --> 00:29:29.317
We have to understand the infrastructure.
00:29:29.317 --> 00:29:32.638
We need to understand VMs, we need to understand storage and we die.
00:29:32.638 --> 00:29:38.442
And when we did that, we became these weird data infrastructure specialists, but it was key to us winning.
00:29:38.482 --> 00:29:41.023
And now everything's kind of coming in there.
00:29:41.023 --> 00:29:55.492
And as I was talking to like Connor Cunningham and I just had a great time just, you know, taking them on at the end of a keynote at a summit event where he was talking about where they were going in the future of storage for fabric.
00:29:55.492 --> 00:30:00.248
I'm like, Dude, you're building an Exadata inside Fabric.
00:30:00.248 --> 00:30:01.229
I gotcha.
00:30:01.229 --> 00:30:02.529
I'm seeing ya.
00:30:02.529 --> 00:30:05.972
And he didn't understand why I was asking him all these questions.
00:30:05.972 --> 00:30:08.315
He's like, Kellyn, you made me earn my pay today.
00:30:08.315 --> 00:30:20.765
And I was like, you didn't understand what I was seeing because I know Exadata isn't everything you were talking about with the storage pools and offloading and being able to, you know, to pull this computer part and offload statements.
00:30:20.765 --> 00:30:23.486
just, you're building an Exadata.
00:30:23.866 --> 00:30:26.888
Some of the Oracle brain in me was just going about it.
00:30:27.757 --> 00:30:29.257
It is the way to go.
00:30:29.257 --> 00:30:31.288
It is something that will secure it.
00:30:31.288 --> 00:30:33.618
I mean, they are putting all of those features in there.
00:30:33.618 --> 00:30:50.474
And I think that's the most important thing that people need to remember with Fabric is that the focus on data protection in the age of AI is you're pulling all these AI services in will be essential to going forward and not being the product and being, but coming back to being the customer.
00:30:50.635 --> 00:30:54.269
Yeah, to wrap up since half an hour flew by very quickly.
00:30:54.329 --> 00:30:58.692
I agree, you know, going back to the whole, what are they doing with our data kind of thing?
00:30:59.314 --> 00:31:07.580
I, I don't necessarily trust Microsoft to have the best judgment about prioritizing, fixing the paper cuts.
00:31:07.701 --> 00:31:12.086
Or like you said, they need to start justifying the thousand co-pilots they made.
00:31:12.086 --> 00:31:18.786
Didn't need to put in paint, didn't need to put in notepad, but I'm right there with you on that one.
00:31:18.786 --> 00:31:21.487
I trust them 100 % with my data.
00:31:21.767 --> 00:31:26.847
Because they're in so many regulated industries that they're not going to mess around.
00:31:26.847 --> 00:31:27.788
Right?
00:31:28.048 --> 00:31:35.211
Sam Altman, he's probably taking my private logs and training his AI chat bot to chat with whatever.
00:31:35.211 --> 00:31:46.814
But Satya understands that they are in HIPAA industries, gov cloud industries, all the socks, everything.
00:31:46.814 --> 00:31:54.992
And so that's gonna be a distinguishing feature where like, Microsoft is doing a bunch of stuff I don't like.
00:31:55.113 --> 00:32:04.410
But if I had to pick a place to put my data and an AI service to use on my data, I trust Microsoft 100%.
00:32:04.410 --> 00:32:05.008
Bar none.
00:32:05.008 --> 00:32:06.269
built into the cloud?
00:32:06.269 --> 00:32:07.770
Always astounding me where people go.
00:32:07.770 --> 00:32:09.092
Well, I have my own data center.
00:32:09.092 --> 00:32:12.894
I trust my own data center and I'm like you can't build into your own data center.
00:32:12.894 --> 00:32:14.987
What is built into the Azure cloud?
00:32:14.987 --> 00:32:16.228
Plain and simple.
00:32:16.228 --> 00:32:17.628
The amount of layers and everything.
00:32:17.628 --> 00:32:27.978
And yes, that complexity also comes with vulnerabilities, but it is astounding what is built in there and I would trust that any day over somebody's know somebody's data center.
00:32:27.978 --> 00:32:29.519
There's no question about it.
00:32:29.519 --> 00:32:32.040
So yeah, you're right there, absolutely.
00:32:32.344 --> 00:32:34.434
So let's wrap up.
00:32:34.434 --> 00:32:42.923
um either AI or RedGate, is your favorite feature or tool that you have?
00:32:42.923 --> 00:32:47.547
um I know RedGate's been doing a bunch of stuff I keep hearing about from Louis, but I haven't looked into any of it.
00:32:47.547 --> 00:32:49.488
So anything you wanna talk about?
00:32:50.309 --> 00:32:50.700
Yeah.
00:32:50.700 --> 00:32:52.328
lot of AI features going in.
00:32:52.328 --> 00:32:56.550
The ones that I usually spend most of my time with is DevOps with Flyway.
00:32:56.590 --> 00:33:01.482
And what we're doing right now is doing assessments on your actual scripts to make recommendations.
00:33:01.482 --> 00:33:08.375
So before deployments go out, you will actually know what risks are in there, that you will get summary reports that let you know that.
00:33:08.375 --> 00:33:11.036
I am really enjoying as I'm working with those.
00:33:11.036 --> 00:33:14.656
The other thing is actually writing the scripts for you.
00:33:15.065 --> 00:33:20.627
building all those out so that you have a full summary of what's going to be deployed beforehand too.
00:33:20.627 --> 00:33:22.068
All of that is built into it.
00:33:22.068 --> 00:33:29.113
I think that's a real smart way to use AI in a simple way from a product that, you know, it's not in the way.
00:33:29.113 --> 00:33:30.393
It's not causing trouble.
00:33:30.393 --> 00:33:34.214
ah It also is using models for the right thing.
00:33:34.214 --> 00:33:42.619
A lot of times people go all in and go, I need to deploy all of Co-Pilot or I need to do all of this in OpenAI instead of using smaller language models.
00:33:42.619 --> 00:33:44.994
And I'm doing a lot of that with open source.
00:33:44.994 --> 00:33:57.396
and building out what
might not be in a product yet and using it as more of a self-service product than pulling in a small piece with a small language model and doing it with open source and doing it isolated so you have that data protection.
00:33:57.396 --> 00:34:02.401
Somebody does something stupid suddenly, remove the small language model, delete it, it's gone.
00:34:02.508 --> 00:34:03.337
Yeah.
00:34:03.397 --> 00:34:04.298
Yeah.
00:34:04.398 --> 00:34:04.798
Yeah.
00:34:04.798 --> 00:34:13.184
And you talked about flywheel on DevOps and we talked about it earlier, but like those skills are going to be incredibly important.
00:34:13.184 --> 00:34:18.286
Like I, I've, I've made some really cool stuff the past three weeks.
00:34:18.766 --> 00:34:31.293
I could not do it if I wasn't using get automated unit tests, automated integration tests, pre pre commit and pre push get hooks.
00:34:31.416 --> 00:34:45.974
to make sure that before I commit, before I push, there's a test for some of these things, doing runs uh with GitHub actions, like spin up a Docker thing, all this stuff, stuff that I couldn't care less about before.
00:34:45.974 --> 00:34:50.447
Like five years ago, three years ago, I read the Phoenix project.
00:34:50.447 --> 00:34:52.378
I was like, yeah, that makes sense.
00:34:53.099 --> 00:34:56.001
Couldn't care less beyond that, right?
00:34:56.001 --> 00:34:57.166
Like I...
00:34:57.166 --> 00:35:01.865
I had really hoped I'd make it through my career without having to learn about Docker or Kubernetes and I failed.
00:35:02.585 --> 00:35:09.306
So Anthony Nosentino is probably saying I told you so somewhere, but yeah, yeah, but this stuff's important.
00:35:09.306 --> 00:35:11.106
So no, very cool.
00:35:11.686 --> 00:35:13.585
Last question, where can people find you?
00:35:13.585 --> 00:35:16.326
Anything you want to pitch or promote or anything like that?
00:35:16.326 --> 00:35:18.666
Or are you just vibing on the river?
00:35:19.025 --> 00:35:20.605
easy to find.
00:35:20.925 --> 00:35:23.244
am gosh I'll be at SQL conference.
00:35:23.244 --> 00:35:26.724
I will be at SQL NL.
00:35:26.724 --> 00:35:29.344
I'm going to be at the Red Gate Summit events.
00:35:29.344 --> 00:35:30.385
All the different ones.
00:35:30.385 --> 00:35:31.804
Tour on the city.
00:35:31.804 --> 00:35:33.125
Is that right?
00:35:33.264 --> 00:35:33.965
Tour on the city.
00:35:33.965 --> 00:35:35.985
I think that's what we're calling right now.
00:35:35.985 --> 00:35:37.445
The Red Gate events as well.
00:35:37.445 --> 00:35:43.885
I will be the keynote speaker at the global PowerShell DevOps conference as well.
00:35:43.885 --> 00:35:45.585
Coming up in April.
00:35:45.585 --> 00:35:46.197
lady.
00:35:46.197 --> 00:35:47.197
Nice.
00:35:47.760 --> 00:35:48.219
Yeah.
00:35:48.219 --> 00:35:53.552
there will be a number of blog posts coming out about the new AI features that I'm writing right now.
00:35:53.552 --> 00:35:56.954
Going through the last final pieces, making sure I'm happy with them.
00:35:56.994 --> 00:36:00.226
And of course, DBA Kevlar, I'm always writing out there too.
00:36:00.695 --> 00:36:01.257
very cool.
00:36:01.257 --> 00:36:02.393
Well, thank you so much.
00:36:02.393 --> 00:36:05.034
We'll have to have you on in 2027.
00:36:05.797 --> 00:36:07.657
Definitely looking forward to it.
00:36:08.023 --> 00:36:08.855
Thanks.