WEBVTT
00:00:00.080 --> 00:00:02.720
Ben Pickett, lovely to meet you.
00:00:02.959 --> 00:00:11.199
We uh we hooked up on kind of LinkedIn and I was um fascinated by your AI newsletter.
00:00:11.519 --> 00:00:23.600
Even more fascinated when I learned your background and I won't steal your thunder, but you're you're now chief operating officer at a consultancy company called SWAR.
00:00:23.920 --> 00:00:29.280
And looks like you kind of were a group of ex-Nike people who went and set that up.
00:00:29.359 --> 00:00:30.559
So that's really interesting.
00:00:30.719 --> 00:00:33.679
But but yeah, your newsletter is absolutely brilliant.
00:00:33.759 --> 00:00:38.560
But Ben, do a little intro, please, and we'll get into the first topic today.
00:00:39.600 --> 00:00:47.920
You've got to stay on and listen to the first topic about KPMG's most recent report, shall we say, that is no longer available.
00:00:48.079 --> 00:00:49.600
But Ben, good to meet you, my friend.
00:00:49.759 --> 00:00:51.200
Thanks for coming on the podcast.
00:00:51.520 --> 00:00:52.560
Good to meet you as well.
00:00:52.719 --> 00:00:53.439
I'm Ben.
00:00:53.520 --> 00:00:57.679
I as you said, and uh I sell AI for a living.
00:00:58.000 --> 00:01:03.119
And uh as you said, uh um is uh an AI startup.
00:01:03.280 --> 00:01:12.560
And uh but for me, I write this newsletter about what most companies are bad at, which is buying it and implementing it.
00:01:12.879 --> 00:01:20.000
And the big thing there is this has become a really big passion product for me.
00:01:20.239 --> 00:01:25.439
I didn't start the newsletter as a way to reach customers.
00:01:25.599 --> 00:01:32.560
It was I wanted to write for a long time, wasn't able to just because of previous roles and things of that nature.
00:01:32.799 --> 00:01:40.000
And this was a way that I could get out there and share what I know and share what I've learned and then learn with the community as well.
00:01:40.159 --> 00:01:40.879
That's the big thing.
00:01:41.040 --> 00:01:52.799
The biggest thing that I've found is that while I write and I do the research and I dive into these different topics, the big part is learning from others, meeting others.
00:01:53.040 --> 00:02:19.199
And uh the end result of that is you end up with connections like you and others who come into your sphere, your network, and then we learn from each other and we help each other out, help companies adopt AI, help companies, you know, with their technology, just help people in general with their, you know, sometimes people are working on their careers or things of that nature, and they just want a little bit of advice and things of that nature.
00:02:19.360 --> 00:02:21.439
So it's been it's been a wild ride.
00:02:21.680 --> 00:02:26.319
I never ever expected it to take off like it has.
00:02:26.479 --> 00:02:30.159
I've been not doing this newsletter for that long.
00:02:30.319 --> 00:02:39.039
This is this last week was my 10th newsletter, and oddly enough, it just took off.
00:02:39.199 --> 00:02:50.719
And I say oddly enough because I expected that within my network that I would get, you know, a few hundred views here and there, and that friends and family and colleagues would read what I wrote.
00:02:50.960 --> 00:02:58.000
I never thought that it would get in front of 70 plus thousand people, but that just didn't that didn't register to me.
00:02:58.159 --> 00:03:04.800
So when that happened, I was like, oh, I wasn't quite ready for that, but that's okay.
00:03:04.960 --> 00:03:09.520
So I pivoted a little bit and I was like, okay, what should what do I need to do?
00:03:09.759 --> 00:03:18.000
Uh one to take advantage of that for all the things, help other people out, and then manage that traffic, if you will.
00:03:18.560 --> 00:03:29.280
And the biggest thing that it provided for me out of the gate is a as an outlet for me to have those conversations to get the ideas out there.
00:03:29.759 --> 00:03:38.719
And a lot of times I just had these thoughts about these Geneai articles that are coming through or technology articles that are coming through that we see every day.
00:03:38.800 --> 00:03:46.879
And I was like, man, you know, you know, if they would just, and not saying I have all the answers, but that there's little things that you can do.
00:03:47.360 --> 00:04:11.840
And uh my time at a previous Fortune Mounthagger company for a number of years that you all probably know, a large retailer out there, they I learned a lot, I guess you could say, about the all of the, you know, that global reach and how to connect with people and how to talk to people and uh all those things uh that went along with that.
00:04:12.000 --> 00:04:17.519
And so I've I've put all that passion and knowledge and things that I've learned, my background.
00:04:17.600 --> 00:04:23.279
I mean, I've done public service, I was a mayor, those type of things into the newsletter and into my thoughts.
00:04:23.439 --> 00:04:25.600
And so I've really enjoyed that process.
00:04:25.759 --> 00:04:29.360
It's been a really fulfilling project, I guess you could say.
00:04:29.519 --> 00:04:39.680
And this last week, the KPMG story that I wrote about, and allegedly KPMG forgot to check its homework.
00:04:40.079 --> 00:04:48.079
They allegedly had 40 of the 45 references not be accurate.
00:04:48.879 --> 00:04:57.759
And so I wrote about it because they used AI, allegedly, that to proof that and and give it out there, and then they had to retract it.
00:04:58.000 --> 00:05:00.560
And there's a number of stories about that.
00:05:00.959 --> 00:05:22.480
And I dove into that because it's not that KPMG itself had this is a common problem, I guess you could say, which is as we're adopting AI, we're learning how to live with AI, and we're also learning how to put it into our workflows in such a way that it makes sense.
00:05:22.800 --> 00:05:31.360
And AI comes across as very confident, and so and the work that it puts out is looks really good.
00:05:31.519 --> 00:05:36.720
So you get this wonderful looking document that comes through and it looks fantastic.
00:05:36.800 --> 00:05:47.600
And you might be the reviewer reviewing that, and everything looks great, and you're like, you're in a hurry because you know there's deadlines and you got to get it out.
00:05:47.759 --> 00:05:48.240
Looks great.
00:05:48.399 --> 00:05:50.160
Okay, great, off it goes.
00:05:51.040 --> 00:06:06.399
Not having those checks in place and the human part in the AI world to check things, check your links, make sure, check with another AI and use multiple AIs to check for your work as well.
00:06:06.639 --> 00:06:10.879
And all those things can help prevent something like that happening with the company.
00:06:11.120 --> 00:06:19.120
And as we adopt AI, it's not just in documents that we put out or or releases, but also with software engineers in code.
00:06:19.439 --> 00:06:26.160
The amount of code that software engineers are required to put out is immense now.
00:06:26.319 --> 00:06:33.120
Um, but there's an expectation that, oh, we have AI, so you must be able to put out a whole, you know, 10x, what you you had before.
00:06:33.360 --> 00:06:35.040
Same thing with the work that we're doing.
00:06:35.120 --> 00:06:40.639
We're expected that, you know, we we have a lot of productivity to go out to have that has to go out.
00:06:40.959 --> 00:06:50.639
And what ends up happening is you feel rushed, and also there's not that downtime to kind of think about what you're doing.
00:06:50.959 --> 00:06:59.680
Sit back and be like, okay, you know, that intuition that you have doesn't have time to kind of settle in and be like, you know, I probably should check that.
00:06:59.839 --> 00:07:00.720
Oh, you know what?
00:07:00.879 --> 00:07:02.160
I forgot to check those links.
00:07:02.240 --> 00:07:05.920
I need to go back and check those links because the time is so compact.
00:07:06.079 --> 00:07:10.319
And then you have nine of the things you have to go after and get out as well.
00:07:10.800 --> 00:07:15.680
So in this new world, we need to adapt our workflows.
00:07:15.839 --> 00:07:20.879
And there's a lot of discussion in the comment section in this week's newsletter, lots of great ideas.
00:07:21.120 --> 00:07:43.120
A lot of them are are very thoughtful and defined, and ideas I even think of is that we have the we have the ability to rework from beginning to end the workflows of which we get our workout using AI.
00:07:43.360 --> 00:07:48.240
But when we just bolt on AI, often you end up with these kind of results.
00:07:48.399 --> 00:07:55.920
That bolt-on with existing workflows without actually looking at your workflows, leaves these holes often that we don't think about.
00:07:56.480 --> 00:08:10.800
So, as a practitioner and an operator and somebody who is working with it every day, I use it all the time in my daily work, everything I do, there has to be those checks and balances along the way as you move things out.
00:08:11.040 --> 00:08:12.639
So yeah.
00:08:14.240 --> 00:08:23.680
And uh I was just kind of letting you speak there, Ben, not not interrupting you because everything that you're saying, I'm mentally going, yeah, tick tick.
00:08:23.920 --> 00:08:33.120
And the when I saw your newsletter, the most recent one about the KPMG retraction of that report that they put out.
00:08:33.279 --> 00:08:35.919
Um well, that's where I I learned that from you.
00:08:36.000 --> 00:08:48.559
And it's almost just like every time you have a conversation with another professional about AI, you tend to learn some come away from that, having just learned something that you did not know before.
00:08:48.879 --> 00:08:53.039
And uh and that that happens to me like multiple times a day.
00:08:53.279 --> 00:09:21.360
But yeah, I mean, the having 40 kind of sources, uh no, sorry, 45 sources at the bottom of your report, and 40 of them were effectively hallucinations at best, when you're a you know, an organization like KPMG just shows you that we need to be a little bit tread a little bit carefully here.
00:09:21.759 --> 00:09:30.159
And well, it I started this podcast for exactly the same reason as you're doing your newsletter, which I'd always always thought about.
00:09:30.320 --> 00:09:39.440
Uh I thought, well, I'm I need to learn, and so I I thought I'm I never even envisaged I'd have guests coming onto my podcast.
00:09:39.600 --> 00:09:48.080
I thought I'll just I'll just put out what I'm researching about AI and how AI was in the early days being adopted by enterprise.
00:09:48.320 --> 00:10:15.360
It I was using Chat GPT at the time, that was pretty much the only one we had, and that was giving me all these different stories of AI adoption, and it just made up so many of them to the point where I was like, I you couldn't even validate it, so I had to completely change my approach, but that was then, this is now.
00:10:15.519 --> 00:10:17.679
So, what happened uh around that?
00:10:17.840 --> 00:10:23.840
Let's just put a bow on the KPMG report from your the Upen studied it.
00:10:23.919 --> 00:10:28.559
I saw you had all of your sources cited on your newsletter.
00:10:28.799 --> 00:10:35.440
But what went, what happened and what went wrong with KPMG's um sorry, not newsletter there.
00:10:35.600 --> 00:10:36.000
I don't even know.
00:10:36.080 --> 00:10:38.159
Was it like a thought leadership piece?
00:10:38.559 --> 00:10:57.360
They they had a flagship report, and they also had uh allegedly four institutions in there that they made claims about, just your your usual piece that highlights the great things that they do and what they're doing with their other companies and and uh customers and things of that nature.
00:10:57.840 --> 00:11:07.440
And and allegedly, even some of the information that was in that report on their customers was also fiction as well.
00:11:07.679 --> 00:11:21.360
So not only did they get the the not only did they get the references wrong that they attributed to you, but they also had content within there that was also not um wrong, that was also wrong as well.
00:11:21.600 --> 00:11:37.200
And for a piece that you're using to send out to customers or prospective customers and also investors and things of that nature, you know, everybody sees type of pieces about, hey, this is what we're doing, and this is the great work we're doing, that that's painful.
00:11:37.360 --> 00:11:45.759
And KPG allegedly did say that, you know, uh they expect the people to check their work, and that's very valid.
00:11:45.919 --> 00:11:56.320
You expect people to check their work, but at the end of the day, that's not an ad problem, and it's not necessarily a beeble problem, but it comes back to being a process problem.
00:11:56.480 --> 00:12:01.200
What are the gates to get those articles out or those pieces out?
00:12:01.360 --> 00:12:08.879
What are the gates to get your code out to make sure that you got a human in the loop that is checking the work?
00:12:09.039 --> 00:12:16.000
And then you also, you know, yeah, it's just the checks are great, but you don't want to go with just one LOM, you want to actually have multiple.
00:12:16.320 --> 00:12:27.120
I mean, those are the things that I build into what I build every day, making sure that you have those cross-checks so you're not just relying on one source, also you get a better result.
00:12:27.360 --> 00:12:28.879
AI is is getting better every day.
00:12:28.960 --> 00:12:44.159
We have new models coming out all the time, and so this is improving, but AI is built to give you things confidently, and it also when you ask for something, it wants that very human, uh, we want to please you type thing.
00:12:44.320 --> 00:12:48.000
Yeah, it wants to tell you, wants to give you what you're asking for.
00:12:48.240 --> 00:12:51.279
So a lot of times that bias is what's coming back to you.
00:12:51.519 --> 00:13:06.559
It's much like in the advertising world where you as an individual consumer, you get all of this advertising coming at you, and it it's very, you know, myopic and and just to what you like and do.
00:13:06.639 --> 00:13:18.799
So your worldview gets, you think it's very broad, but it gets very constrained into just you know, this little pipeline of what you care about, and it keeps giving you more of those things, so then it reinforces that.
00:13:18.879 --> 00:13:31.360
So the same thing could happen with with uh the LLMs where it it reinforces your bias and things of that nature, it has context about what you like, so it'll include those into what you're doing.
00:13:31.600 --> 00:13:34.159
So you just got to be careful in that respect.
00:13:34.320 --> 00:13:40.080
Check your work, have somebody else check your work, and then also have those process checks, automated process checks.
00:13:40.240 --> 00:13:45.200
And then the human that takes a look and sits down and goes, you know, that doesn't seem quite right.
00:13:45.360 --> 00:13:46.480
And check all the links.
00:13:46.799 --> 00:13:47.440
Yeah, every link.
00:13:48.399 --> 00:13:49.279
Boom, boom, boom, boom.
00:13:49.519 --> 00:13:55.120
Yeah, and that but I that needs to happen early in the workflow, doesn't it?
00:13:55.279 --> 00:13:55.679
Oh, yes.
00:13:56.159 --> 00:14:00.559
Let's check, let's check this first before we go further.
00:14:00.639 --> 00:14:22.639
And you find that when you're code when I'm vibe coding, for example, it you soon learn when you just get dragged off down a path that you didn't validate, and then you hit roadblocks down the line, and you're like, why didn't you why and you're you're kind of blaming the AI tool, go, why didn't you know about this?
00:14:22.799 --> 00:14:31.679
And so I got to use when I use lovable, for example, I'm I'm on lovable, I'm on Clawed, I'm on Grok.
00:14:32.240 --> 00:14:42.720
And that's sometimes I'm validating all three sources of in information because I Grok's kind of helpful because I know lovable is based on Claude.
00:14:43.279 --> 00:14:48.960
I'm using Claude co-work, for example, but I know Groc's completely independent of those two.
00:14:49.039 --> 00:14:51.360
So it's a good, it's a good check and balance.
00:14:51.440 --> 00:15:08.639
And but just having that check early in the workflow, and Thomas Onenwold, who was on the podcast last week, he's been on a number of times, he talks about you know the importance of remembering that AI, you know, deterministic and non-deterministic.
00:15:08.799 --> 00:15:26.080
And so understanding how these models are uh work, which is what you're explaining, and their their kind of predilection to bias towards giving you what you want and saying that you're right, you've got to you've got to check that.
00:15:26.240 --> 00:15:37.840
And what when I did my um, this was in the early days of Chat GPT, I kind of did like a business, I was doing some business planning and Chat GPT was just being so agreeable.
00:15:38.000 --> 00:15:44.799
And I said to it, I said, look, imagine you're Elon Musk and he saw my business plan, what would he say?
00:15:44.960 --> 00:15:48.080
So and I know it's not passing an Elon test, right?
00:15:48.240 --> 00:15:52.559
And and Chat GPT came back and said some one line, very interesting.
00:15:52.639 --> 00:15:55.360
It said, and it was a very interesting line to me.
00:15:55.519 --> 00:15:59.519
And then one line was, oh, you want some tough love, do you?
00:15:59.679 --> 00:16:01.519
And I was like, Yes, I do.
00:16:01.679 --> 00:16:08.799
And then it got into giving me a non-agreeable answer.
00:16:10.000 --> 00:16:11.759
Yeah, yeah, absolutely.
00:16:12.000 --> 00:16:17.360
And and the thing about that is these are all great tools.
00:16:17.519 --> 00:16:20.879
There, there are methods though that you can adopt.
00:16:21.039 --> 00:16:30.000
One is think about creating personas, adversarial personas, so personas that help you check your work as who's your audience, okay.
00:16:30.320 --> 00:16:33.360
Have create personas that represent your audience.
00:16:33.519 --> 00:16:49.919
So if you're doing a newsletter like I am, I create personas to check my work where you know, and have a cross-section of the people who read my newsletter and be like, take a look at my newsletter that I spent all my time writing and give me some feedback, and then I go back and I adjust against that.
00:16:50.240 --> 00:16:54.960
And that helps one, to make sure that what I'm writing is clear.
00:16:55.120 --> 00:16:57.440
So there's a quality improvement over that.
00:16:57.759 --> 00:17:00.639
Two, I check all my links.
00:17:01.039 --> 00:17:03.759
I I definitely I check all my links manually, 100%.
00:17:04.640 --> 00:17:10.400
But at the same time, it can be like, you know, you can really dive into those those references.
00:17:10.559 --> 00:17:11.519
Are they good references?
00:17:11.599 --> 00:17:25.279
Because sometimes, even if you go and check the article, article looks good, the article that you're using the reference may not may actually be hallucinated as well because we're in this age where you really need to dig into those references.
00:17:25.440 --> 00:17:29.200
And for things that matter, you also got to wait things.
00:17:29.359 --> 00:17:41.519
You're not gonna necessarily be able to check everything all the time, but what are the things that matter if they get out that are gonna impact you and your company and your reputation?
00:17:41.680 --> 00:17:45.920
Those things should be hard stops and hard checks in that process.
00:17:46.319 --> 00:17:47.440
Doesn't have to be everything.
00:17:47.599 --> 00:17:54.960
You can really leverage the tools that are out there to check your work, but definitely, definitely have those hard checks for the things that matter.
00:17:55.119 --> 00:18:03.440
So, like when allegedly a KPMG has this happened, you're able to catch it before it gets out the door.
00:18:03.759 --> 00:18:13.839
Because you have the person signing off on it and says, I am staking my personal reputation and my my position on the fact that this thing is good.
00:18:14.000 --> 00:18:17.599
Just like editors do today in big publications.
00:18:17.759 --> 00:18:35.200
They check that work and make sure that it represents the company, the culture of their publication, and then also just the overall what they're trying to say out in the world and that it doesn't conflict with the things that they're you know, their mission, vision statement or what they're trying to do.
00:18:35.440 --> 00:18:57.920
Yeah, and and look, then I totally respect your your work in being diligent in doing that, because as you say, it's so easy to go see a store, a news art article that that is deliberately biased as well, because of the pub who the publication, their leanings, where their funding comes from.
00:18:58.160 --> 00:19:03.839
And um, and I know how much work it is to do what you're to do what you're doing.
00:19:04.079 --> 00:19:06.400
Um look, should we should we pivot back?
00:19:06.799 --> 00:19:10.079
I'm really interested in what you guys are doing today.
00:19:10.319 --> 00:19:30.000
You're you were I think I'm talking out of school when I say you're you're ex Nike, and um, and I'm guessing that's where without you, I think you were 10 years there, without sort of going into what they were doing around AI, what was your kind of like first experiences of generative AI?
00:19:30.160 --> 00:19:40.559
Because I'm guessing that must have been when you were you were working there, and then you've moved out of the sort of the big, huge giant corporate to to set up what you guys are doing.
00:19:40.640 --> 00:19:42.799
What was that journey like for you?
00:19:43.200 --> 00:20:00.880
Yeah, I started out in at Nike in 2013 as a contractor, and funny enough, I I it became a very successful opportunity that then led to a position and they asked me what I wanted to do.
00:20:01.119 --> 00:20:06.559
So, in that moment, of the things that were available, I was tackling some.
00:20:06.960 --> 00:20:11.440
I was a data guy, uh, you know, my roots are very data oriented.
00:20:11.839 --> 00:20:21.599
And uh so I I was brought in to help with performance and Nike, Nike.com and their back end databases and things of that nature.
00:20:22.000 --> 00:20:31.279
And so when I took the position, I was like, I want to tackle like test data management and you know, service socialization and the the CICD pipeline to production.
00:20:31.440 --> 00:20:33.680
So that's where I got my start.
00:20:33.839 --> 00:20:37.759
Uh, and you might see some things about in some of the things I'm doing.
00:20:38.000 --> 00:20:47.440
And then as I moved through my career at Nike, usually it was something I'd solve a problem, and then I didn't they'd be like, okay, hey.
00:20:47.680 --> 00:21:06.720
So I uh I helped with a very large project, and um that was a very successful project, and so I ended up with my next role, and that kept happening, and I got known as uh known for taking on large global projects and implementing them.
00:21:07.039 --> 00:21:18.240
And so then, long story short, I ended up as a global director of site reliability engineering at Nike, covering a number of different areas, and that was a really good fit for me.
00:21:18.480 --> 00:21:32.720
Had a a very um uh great experience working, you know, across retail, the dot-com, you know, infrastructures, global infrastructure, and then you know, supply chain and and also you know the wholesale piece.
00:21:33.200 --> 00:21:35.440
And that was a very subscribing position.
00:21:35.680 --> 00:21:45.519
And as I was in that position, uh, I think about three years or two and a half years back, JAI was the really everybody was like, okay, what do we do?
00:21:45.599 --> 00:21:47.440
This is this is a new tool.
00:21:47.920 --> 00:22:01.599
Organizations were seeing their their end users using ChatGDP to help them with their daily work, and there was no framework for managing that, and there was no idea on how to adopt a within the organization.
00:22:01.839 --> 00:22:04.880
So one day I was, and it's perfectly okay to share.
00:22:05.119 --> 00:22:10.720
One day I was talking to one of our C M VPs and I just gave him the whole like, here's what we need to do with Genny I.
00:22:10.880 --> 00:22:12.079
And he's like, okay, great.
00:22:12.240 --> 00:22:15.920
Uh why don't you build a strategy for us and for our division?
00:22:16.079 --> 00:22:17.680
And I was like, yeah, yeah, yeah.
00:22:20.400 --> 00:22:20.880
Yeah.
00:22:21.119 --> 00:22:22.960
So I proceeded to do that.
00:22:23.039 --> 00:22:35.039
Um I built out a strategy and learned a lot, did some pilots, got some early wins and things of that nature, also took on automation for the division.
00:22:35.359 --> 00:22:43.440
And that grew and grew and grew and came into this kind of uh cross-organizational strategy, if you will.
00:22:44.000 --> 00:22:56.079
And uh and then I really fell in love with the possibilities that uh AI brings to us and can give us all of us in in our daily work.
00:22:56.160 --> 00:23:10.720
And I'm not so much of a that there's very much the you know you hear oh nobody'll have to work anymore to things any paradigm shift that we have out there has never resulted in a long-term job loss.
00:23:10.960 --> 00:23:16.880
And I don't think and I know I'm probably one of the few I don't think Genai is any different.
00:23:17.039 --> 00:23:19.279
Yes, initially there's job losses.
00:23:19.440 --> 00:23:25.839
Yes, initially there is disruption but overall on the long term new jobs different jobs and more jobs.
00:23:26.000 --> 00:24:05.599
And if you look at what's actually happening with the developers who are using Gen AI, with the people who are using Gen AI in their daily work, they're busier never because and I think the NVIDIA don't quote me on this and I say that the Nvidia's CEO was talking about how they're seeing more code than ever being created by some of their engineers and that they have this expectation that some of their engineers should be spending hundreds of thousands of dollars on on spam just and they would expect that from their top engineers because they're doing more with AI and assisted in what they're doing.
00:24:06.559 --> 00:24:19.279
But at some point you know the the human factor comes in where you're going to need more of that because it's so easy to get the code because it's so easy to get the things through AI the the demand increases.
00:24:19.440 --> 00:24:30.319
So over time that demand because businesses now can get access to code at at a faster pace they don't have to wait months you know they there's more demand.
00:24:30.480 --> 00:24:53.680
So demand increases and then that's how that's how I see that you know boomeranging back around you see with more positions because as companies can more easily cheaper create the things that they need there's more demand which those positions come back in in a different way in a lot of ways but I do see I do see that this is will be a net increase in jobs over time.
00:24:56.319 --> 00:25:04.720
Yeah um so I'm an optimist on that front as well and it yes problem is is is the speed of this thing.
00:25:04.799 --> 00:26:09.519
Like there's lots of comparisons to industrial revolution and which saw a huge amount of automation and the rise of factories and you know that that was was going to be the huge impact on job and create lots of job losses did the exact opposite just grew everybody's economies massively and actually just massively increased the number of of jobs available but that took that took some time it's just like you know every every month it's just seems to be accelerating um so so I think that's the challenge and then the doomer talk kind of was started by the big players so the Darios and Sam's and Elons of this world have been talking the Doomer language and they kind of go first with the Doomer part and then they sort of then they started tagging on oh but don't worry because it's going to create a whole bunch of jobs that just don't even exist today that you'll be you'll be able to do.
00:26:09.920 --> 00:26:53.839
And then and now they've flipped on it and said no no no it's not going to take people's jobs it's all going to be great and that's kind of come a little bit late um given we're all animals very sensitive to any risks to us uh we're we're gonna look on the um on the possible downsides of new technologies aren't we what why why would we do anything else um so so I think you're right but what's your view on the how fast that change happens and and and and people go from oh I now realize that my job is not going to be is going to be done by the AI in the future but the AI's create all these new job openings for me.
00:26:54.079 --> 00:26:56.720
How does that how does that actually work in practice?
00:26:57.039 --> 00:27:21.519
Yeah and I think if you go back to and I wrote about this the companies who are doing AI adoption are there's not a lot of them out there that do it well you know if you look at that 95% number 95% of companies are aren't getting it right and 5% are and when you look at it from that lens you're like AI is everywhere and we're all using it whether we whether we realize it or not and just about everything that we do.
00:27:22.240 --> 00:27:45.920
But if companies are struggling to adopt even though AI is moving at this very fast pace how long is it going to take those companies to catch up and and maybe that's the slowdown piece that that reframes that gives people time to catch up and I do think people need to think about that in their own roles how can you use AI today?
00:27:46.079 --> 00:27:47.680
How can I do better prompts?
00:27:47.839 --> 00:28:02.000
What can I do from an individual to set myself apart in the roles that I'm doing or maybe even you know you realize that your role is you know that that role is probably not going to be needed.
00:28:02.079 --> 00:28:56.720
So what will what is that similar role or that passion thing that you really want to go after that you can leverage uh and dive into it assisted by AI I would say because to different you have to incorporate AI and and I know that uh out there in the world it's there there's some resistance to that and that's completely understandable it's a huge change like you said and it's happening fast and like you said as well that we have new models coming out new capabilities new announcements sometimes on a daily basis sometimes I think there was one weekend where we had nine different announcements over the weekend of new things that came out that were major and it's like you almost get news whiplash you're like that's a lot to absorb because you have full on capabilities of things that you can use and in your daily work potentially that are brand new that you have to learn.
00:28:57.039 --> 00:29:56.000
So I think the best approach is take small steps focus on the things that helped you the best learn you know be a sponge learn uh as things come out focus on the things that are relevant to you and try to get try to cut out the noise because there is a lot of noise as well and I think that'll help people a lot especially in as I'm helping organizations adopt AI I'm helping them uh you know on my daily and it's it it is a struggle right even at the organizational level because often you have the CEO the CO the CTO who have this expectation from potentially Wall Street or their investors that says you need to adopt AI because we want the sufficiency but then you have people on the other end of that who the end user who are who are saying you know what it takes me twice as long to use Chat GP to do X when I can just do it myself and be done with it.
00:29:56.319 --> 00:30:32.000
And that's a real thing that happens and a lot of that is not the organization and the individual as well not having the resources to get the training and sure you can prompt and self-train all you want but there are tips and tricks and methodologies that you can use that give you a lot better experience as you're using the and there are also tools out there that make it a lot easier too contact switching is one of those things that uh you have to do more of uh using AI.
00:30:32.079 --> 00:30:45.119
So sometimes you know like you said those different screens you have grok, touchdp, Gemini, Perplexy, and you're going through those and those are some of the things that I'm tackling is how do you get that in one place where I work today.
00:30:45.359 --> 00:31:40.160
And that's more of a passion thing because I don't want to switch between different screens I I want to be be able to work where I'm working and ask the questions and then cross check with all those LOMs at just one prompt because then that way there's less worry and then how do I incorporate that in my daily how can it help me with following up on leads help me with my sales funnel how can that help me grow those relationships that I have out there and uh make the best of this whole new world if you will yeah yeah I think again lots lots of good points lots of good advice there as well the point you made first resonated which is for the Doomer the kind of Doomer perspective on this and yeah the reality of how fast AI could actually automate out a human.
00:31:40.640 --> 00:33:07.119
I haven't seen it yet I've seen what I would class as AI washing um uh with companies who kind of were too big anyway especially the ones that got big during COVID and um needed to be right sized and they say well we're using AI we didn't have the wrong size company this is you know I I think there's a lot of AI washing going on so I think people should be really cognizant of that I am yet to see a and I'm looking I'm yet to see a job role completely eliminated by AI and I think your your your just kind of like approach is the right one which is if everybody just picks up one of these tools and goes right how can I actually use this in my day job to kind of take the mundane stuff away do some heavy lifting for me you know make me more effective and then I I keep the kind of the human touch and the domain experience but within the context of our business operations and you know how clunky these old companies are and siloed around their business operations you know that that requires people to really understand the business that they need to need to be there and and involved in any AI project as well.
00:33:07.279 --> 00:33:31.359
So so yeah I I'm I'm with you Ben I I just can't see this like oh we all wake up in January and everyone kind of says oh it's it's swept away every job that involved looking at a computer screen is now done by AI AI I just don't think it's that's not realistic and I don't actually think that is what is going to happen either.
00:33:31.680 --> 00:33:43.759
Yeah and I tend to agree the the a lot of what you talked about also in that the AI washing a lot of that is not necessarily the need to eliminate those roles that is part of it.
00:33:43.920 --> 00:34:00.960
But another part of that is companies are having to keep their margins and invest in AI so they're downsizing the people that they have so they can pay for compute and GPUs and things of that nature, build out data centers and power and all the things that go along with that.
00:34:01.200 --> 00:34:38.559
But as those prices go cheaper I think you'll see a little bit of boomerang from that too where those I mean that's a little predictive and I could be wrong but you know right now there's a race to you know there's a race on infrastructure and compute and resources that go into that because these big AI companies are throwing everything they can at it and it's also a resource thing too which is you know there there's a number of metrics out there we don't really have enough resources in the pipeline to meet the the the projected demands for AI.
00:34:39.280 --> 00:36:02.719
And unless we're good stewards with what we got which means you know you probably don't want to send your query to the most the strongest model out there to do a file move or to check your email or to even create an email you don't want to use for that you want to use grok fast or or one of the the cheap models open source if you can open source any of the LMs or models of open source real cheap models that do it well you know you're you're pennies versus dollars right and also on a compute side you're saving compute one of the one of the newsletters I did on Uber that uh that did pretty well as well which their Uber CEO allegedly came out and said hey uh we spent all of our I spend within like the first three or four months and they didn't really have any uplift in consumer facing spend or things that were material there that he can see for the spend that they made and a lot of that is because they're if you just throw resources at AI and you don't think about that implementation and the stewardship as you go, you're gonna end up spending tons of money and get not get the result necessarily that you want.
00:36:02.880 --> 00:36:32.719
Whereas when you take a proactive approach and you're like okay as I go through my day and as I'm building this out for my company how how can I make sure that the right model is the the best model for the the better price not necessarily the cheapest price but the better price gets me to my end result faster so that that spend is much easier for a company to um absorb if you will and also they get a lot more benefit from it.
00:36:32.960 --> 00:37:06.800
And the one I wrote about the the Walmart blueprint which actually did very well over I think 71,000 impressions on LinkedIn which just blew my mind was just about how Walmart took a thoughtful approach and they actually implemented AI in a in a thoughtful way using agents and then super agents and then also cord them off by different areas and actually increased it increased their correlationally increased their um their uh revenue in some areas I think by close to 30%.
00:37:07.599 --> 00:37:27.440
And so they ended up seeing that whole end-to-end result because they took a thoughtful approach and I think it took them like two years and that's yeah newsletter yeah can imagine yeah but uh yeah that that's kind of the kind of have in my newsletter I talk about those interesting topics of how people you know where people go wrong so we can learn from that.
00:37:27.679 --> 00:38:12.639
And then also I try to find things where hey you know what this company seems to get it right and maybe we can all learn from that as well because as we are trying to help our customers and we're trying to help our companies and organizations do uh do better with AI and and go faster because you know we all have to be competitive because all the other companies out there are adopting as well we have this opportunity to be thoughtful and also keep our costs down and also help ourselves and our customers and that whole pipeline from end to end be better overall.
00:38:12.880 --> 00:38:23.119
And I think that's something that is more human I guess you could say than just throwing an eye at it because there's people in that process they have to be there.
00:38:23.519 --> 00:38:26.880
And to get it right you still need the people to get it right.
00:38:27.039 --> 00:38:34.800
You can't just say hey yeah go out there solve everything because it's just you're gonna end up with chaos and a lot of money out the door.
00:38:35.119 --> 00:39:47.199
Yeah yeah yeah and we're we're already seeing the first headlines of companies saying absolutely this AI thing's a little bit expensive isn't it is it are we doing better with just the human resource uh in some of these processes so yeah those realities are I think we're gonna see more and more of those realities and I think we need to get chat uh sorry open AI and anthropic through their IPOs and out the other side and then and and this will be an another podcast episode where I then see that battleground is going to be you've got those giant companies that need revenue they need power they need compute and you've got a bunch of free of charge Chinese models that are really really great do a great job that don't require all of that expenditure but they're more difficult for an enterprise to use so I think we need to get those IPOs done and then I think some commercial realities are then going to set in for those companies themselves and also for enterprises who are consuming AI.
00:39:47.440 --> 00:40:02.480
But look that that was a that was a really fascinating conversation Ben I know we've hit time where do people find you where do they sign up for your newsletter please and whatever you mention I'll I'll drop the links down here in the show notes.
00:40:02.800 --> 00:41:00.880
Yeah um I'm on LinkedIn and I'll send you that as well and then I also have a website benpicket.com you can check me out and then uh the company I work for of mine at my daily is squaw-ai.com uh where we we help people and you know help people adopt AI uh in the places that they work and one more thing if you're interested in how you know what works things of that nature reach out uh you can DM me I'm happy to to uh respond I usually respond if I'm super busy it may take me a few days but I'll get back to you but happy to engage with people as well and then the biggest thing I would say and I think the last thing I'd leave you with whenever you're working with AI always ask AI what question am I not asking because that'll help you you surface things that you're not thinking about and that the AI isn't really focusing on.
00:41:01.039 --> 00:41:10.079
And that's been really helpful to me as well because that question tends to surface a lot of things that you wouldn't necessarily think about as you're using it.
00:41:10.400 --> 00:41:35.920
But overall this new this new age if you will isn't going away it's exciting it's in every part of our daily lives whether we realize it or not and uh it's a huge opportunity for all of us whether we're in sales or whether we're in leadership whether we're you know doing our daily job wherever we're at it can help us in in our roles and what we're doing.
00:41:36.159 --> 00:41:44.079
And um I think the future is bright overall sure there's some turmoil right now but I think that'll even out over the coming years.
00:41:44.400 --> 00:41:45.280
Yeah I agree.
00:41:45.840 --> 00:42:09.119
Really positive notes end on there Ben and um yeah I look forward to reading your newsletters as as they come out each week and I'll drop all of those links below and um yeah great having you on the podcast you've been a brilliant guest thank you very much I've learned learned a lot from you yeah thank you Ben thank you thank you for having me you all have a great day yeah you too you too thanks Ben bye bye bye bye bye
00:00:00.080 --> 00:00:02.720
Ben Pickett, lovely to meet you.
00:00:02.959 --> 00:00:11.199
We uh we hooked up on kind of LinkedIn and I was um fascinated by your AI newsletter.
00:00:11.519 --> 00:00:23.600
Even more fascinated when I learned your background and I won't steal your thunder, but you're you're now chief operating officer at a consultancy company called SWAR.
00:00:23.920 --> 00:00:29.280
And looks like you kind of were a group of ex-Nike people who went and set that up.
00:00:29.359 --> 00:00:30.559
So that's really interesting.
00:00:30.719 --> 00:00:33.679
But but yeah, your newsletter is absolutely brilliant.
00:00:33.759 --> 00:00:38.560
But Ben, do a little intro, please, and we'll get into the first topic today.
00:00:39.600 --> 00:00:47.920
You've got to stay on and listen to the first topic about KPMG's most recent report, shall we say, that is no longer available.
00:00:48.079 --> 00:00:49.600
But Ben, good to meet you, my friend.
00:00:49.759 --> 00:00:51.200
Thanks for coming on the podcast.
00:00:51.520 --> 00:00:52.560
Good to meet you as well.
00:00:52.719 --> 00:00:53.439
I'm Ben.
00:00:53.520 --> 00:00:57.679
I as you said, and uh I sell AI for a living.
00:00:58.000 --> 00:01:03.119
And uh as you said, uh um is uh an AI startup.
00:01:03.280 --> 00:01:12.560
And uh but for me, I write this newsletter about what most companies are bad at, which is buying it and implementing it.
00:01:12.879 --> 00:01:20.000
And the big thing there is this has become a really big passion product for me.
00:01:20.239 --> 00:01:25.439
I didn't start the newsletter as a way to reach customers.
00:01:25.599 --> 00:01:32.560
It was I wanted to write for a long time, wasn't able to just because of previous roles and things of that nature.
00:01:32.799 --> 00:01:40.000
And this was a way that I could get out there and share what I know and share what I've learned and then learn with the community as well.
00:01:40.159 --> 00:01:40.879
That's the big thing.
00:01:41.040 --> 00:01:52.799
The biggest thing that I've found is that while I write and I do the research and I dive into these different topics, the big part is learning from others, meeting others.
00:01:53.040 --> 00:02:19.199
And uh the end result of that is you end up with connections like you and others who come into your sphere, your network, and then we learn from each other and we help each other out, help companies adopt AI, help companies, you know, with their technology, just help people in general with their, you know, sometimes people are working on their careers or things of that nature, and they just want a little bit of advice and things of that nature.
00:02:19.360 --> 00:02:21.439
So it's been it's been a wild ride.
00:02:21.680 --> 00:02:26.319
I never ever expected it to take off like it has.
00:02:26.479 --> 00:02:30.159
I've been not doing this newsletter for that long.
00:02:30.319 --> 00:02:39.039
This is this last week was my 10th newsletter, and oddly enough, it just took off.
00:02:39.199 --> 00:02:50.719
And I say oddly enough because I expected that within my network that I would get, you know, a few hundred views here and there, and that friends and family and colleagues would read what I wrote.
00:02:50.960 --> 00:02:58.000
I never thought that it would get in front of 70 plus thousand people, but that just didn't that didn't register to me.
00:02:58.159 --> 00:03:04.800
So when that happened, I was like, oh, I wasn't quite ready for that, but that's okay.
00:03:04.960 --> 00:03:09.520
So I pivoted a little bit and I was like, okay, what should what do I need to do?
00:03:09.759 --> 00:03:18.000
Uh one to take advantage of that for all the things, help other people out, and then manage that traffic, if you will.
00:03:18.560 --> 00:03:29.280
And the biggest thing that it provided for me out of the gate is a as an outlet for me to have those conversations to get the ideas out there.
00:03:29.759 --> 00:03:38.719
And a lot of times I just had these thoughts about these Geneai articles that are coming through or technology articles that are coming through that we see every day.
00:03:38.800 --> 00:03:46.879
And I was like, man, you know, you know, if they would just, and not saying I have all the answers, but that there's little things that you can do.
00:03:47.360 --> 00:04:11.840
And uh my time at a previous Fortune Mounthagger company for a number of years that you all probably know, a large retailer out there, they I learned a lot, I guess you could say, about the all of the, you know, that global reach and how to connect with people and how to talk to people and uh all those things uh that went along with that.
00:04:12.000 --> 00:04:17.519
And so I've I've put all that passion and knowledge and things that I've learned, my background.
00:04:17.600 --> 00:04:23.279
I mean, I've done public service, I was a mayor, those type of things into the newsletter and into my thoughts.
00:04:23.439 --> 00:04:25.600
And so I've really enjoyed that process.
00:04:25.759 --> 00:04:29.360
It's been a really fulfilling project, I guess you could say.
00:04:29.519 --> 00:04:39.680
And this last week, the KPMG story that I wrote about, and allegedly KPMG forgot to check its homework.
00:04:40.079 --> 00:04:48.079
They allegedly had 40 of the 45 references not be accurate.
00:04:48.879 --> 00:04:57.759
And so I wrote about it because they used AI, allegedly, that to proof that and and give it out there, and then they had to retract it.
00:04:58.000 --> 00:05:00.560
And there's a number of stories about that.
00:05:00.959 --> 00:05:22.480
And I dove into that because it's not that KPMG itself had this is a common problem, I guess you could say, which is as we're adopting AI, we're learning how to live with AI, and we're also learning how to put it into our workflows in such a way that it makes sense.
00:05:22.800 --> 00:05:31.360
And AI comes across as very confident, and so and the work that it puts out is looks really good.
00:05:31.519 --> 00:05:36.720
So you get this wonderful looking document that comes through and it looks fantastic.
00:05:36.800 --> 00:05:47.600
And you might be the reviewer reviewing that, and everything looks great, and you're like, you're in a hurry because you know there's deadlines and you got to get it out.
00:05:47.759 --> 00:05:48.240
Looks great.
00:05:48.399 --> 00:05:50.160
Okay, great, off it goes.
00:05:51.040 --> 00:06:06.399
Not having those checks in place and the human part in the AI world to check things, check your links, make sure, check with another AI and use multiple AIs to check for your work as well.
00:06:06.639 --> 00:06:10.879
And all those things can help prevent something like that happening with the company.
00:06:11.120 --> 00:06:19.120
And as we adopt AI, it's not just in documents that we put out or or releases, but also with software engineers in code.
00:06:19.439 --> 00:06:26.160
The amount of code that software engineers are required to put out is immense now.
00:06:26.319 --> 00:06:33.120
Um, but there's an expectation that, oh, we have AI, so you must be able to put out a whole, you know, 10x, what you you had before.
00:06:33.360 --> 00:06:35.040
Same thing with the work that we're doing.
00:06:35.120 --> 00:06:40.639
We're expected that, you know, we we have a lot of productivity to go out to have that has to go out.
00:06:40.959 --> 00:06:50.639
And what ends up happening is you feel rushed, and also there's not that downtime to kind of think about what you're doing.
00:06:50.959 --> 00:06:59.680
Sit back and be like, okay, you know, that intuition that you have doesn't have time to kind of settle in and be like, you know, I probably should check that.
00:06:59.839 --> 00:07:00.720
Oh, you know what?
00:07:00.879 --> 00:07:02.160
I forgot to check those links.
00:07:02.240 --> 00:07:05.920
I need to go back and check those links because the time is so compact.
00:07:06.079 --> 00:07:10.319
And then you have nine of the things you have to go after and get out as well.
00:07:10.800 --> 00:07:15.680
So in this new world, we need to adapt our workflows.
00:07:15.839 --> 00:07:20.879
And there's a lot of discussion in the comment section in this week's newsletter, lots of great ideas.
00:07:21.120 --> 00:07:43.120
A lot of them are are very thoughtful and defined, and ideas I even think of is that we have the we have the ability to rework from beginning to end the workflows of which we get our workout using AI.
00:07:43.360 --> 00:07:48.240
But when we just bolt on AI, often you end up with these kind of results.
00:07:48.399 --> 00:07:55.920
That bolt-on with existing workflows without actually looking at your workflows, leaves these holes often that we don't think about.
00:07:56.480 --> 00:08:10.800
So, as a practitioner and an operator and somebody who is working with it every day, I use it all the time in my daily work, everything I do, there has to be those checks and balances along the way as you move things out.
00:08:11.040 --> 00:08:12.639
So yeah.
00:08:14.240 --> 00:08:23.680
And uh I was just kind of letting you speak there, Ben, not not interrupting you because everything that you're saying, I'm mentally going, yeah, tick tick.
00:08:23.920 --> 00:08:33.120
And the when I saw your newsletter, the most recent one about the KPMG retraction of that report that they put out.
00:08:33.279 --> 00:08:35.919
Um well, that's where I I learned that from you.
00:08:36.000 --> 00:08:48.559
And it's almost just like every time you have a conversation with another professional about AI, you tend to learn some come away from that, having just learned something that you did not know before.
00:08:48.879 --> 00:08:53.039
And uh and that that happens to me like multiple times a day.
00:08:53.279 --> 00:09:21.360
But yeah, I mean, the having 40 kind of sources, uh no, sorry, 45 sources at the bottom of your report, and 40 of them were effectively hallucinations at best, when you're a you know, an organization like KPMG just shows you that we need to be a little bit tread a little bit carefully here.
00:09:21.759 --> 00:09:30.159
And well, it I started this podcast for exactly the same reason as you're doing your newsletter, which I'd always always thought about.
00:09:30.320 --> 00:09:39.440
Uh I thought, well, I'm I need to learn, and so I I thought I'm I never even envisaged I'd have guests coming onto my podcast.
00:09:39.600 --> 00:09:48.080
I thought I'll just I'll just put out what I'm researching about AI and how AI was in the early days being adopted by enterprise.
00:09:48.320 --> 00:10:15.360
It I was using Chat GPT at the time, that was pretty much the only one we had, and that was giving me all these different stories of AI adoption, and it just made up so many of them to the point where I was like, I you couldn't even validate it, so I had to completely change my approach, but that was then, this is now.
00:10:15.519 --> 00:10:17.679
So, what happened uh around that?
00:10:17.840 --> 00:10:23.840
Let's just put a bow on the KPMG report from your the Upen studied it.
00:10:23.919 --> 00:10:28.559
I saw you had all of your sources cited on your newsletter.
00:10:28.799 --> 00:10:35.440
But what went, what happened and what went wrong with KPMG's um sorry, not newsletter there.
00:10:35.600 --> 00:10:36.000
I don't even know.
00:10:36.080 --> 00:10:38.159
Was it like a thought leadership piece?
00:10:38.559 --> 00:10:57.360
They they had a flagship report, and they also had uh allegedly four institutions in there that they made claims about, just your your usual piece that highlights the great things that they do and what they're doing with their other companies and and uh customers and things of that nature.
00:10:57.840 --> 00:11:07.440
And and allegedly, even some of the information that was in that report on their customers was also fiction as well.
00:11:07.679 --> 00:11:21.360
So not only did they get the the not only did they get the references wrong that they attributed to you, but they also had content within there that was also not um wrong, that was also wrong as well.
00:11:21.600 --> 00:11:37.200
And for a piece that you're using to send out to customers or prospective customers and also investors and things of that nature, you know, everybody sees type of pieces about, hey, this is what we're doing, and this is the great work we're doing, that that's painful.
00:11:37.360 --> 00:11:45.759
And KPG allegedly did say that, you know, uh they expect the people to check their work, and that's very valid.
00:11:45.919 --> 00:11:56.320
You expect people to check their work, but at the end of the day, that's not an ad problem, and it's not necessarily a beeble problem, but it comes back to being a process problem.
00:11:56.480 --> 00:12:01.200
What are the gates to get those articles out or those pieces out?
00:12:01.360 --> 00:12:08.879
What are the gates to get your code out to make sure that you got a human in the loop that is checking the work?
00:12:09.039 --> 00:12:16.000
And then you also, you know, yeah, it's just the checks are great, but you don't want to go with just one LOM, you want to actually have multiple.
00:12:16.320 --> 00:12:27.120
I mean, those are the things that I build into what I build every day, making sure that you have those cross-checks so you're not just relying on one source, also you get a better result.
00:12:27.360 --> 00:12:28.879
AI is is getting better every day.
00:12:28.960 --> 00:12:44.159
We have new models coming out all the time, and so this is improving, but AI is built to give you things confidently, and it also when you ask for something, it wants that very human, uh, we want to please you type thing.
00:12:44.320 --> 00:12:48.000
Yeah, it wants to tell you, wants to give you what you're asking for.
00:12:48.240 --> 00:12:51.279
So a lot of times that bias is what's coming back to you.
00:12:51.519 --> 00:13:06.559
It's much like in the advertising world where you as an individual consumer, you get all of this advertising coming at you, and it it's very, you know, myopic and and just to what you like and do.
00:13:06.639 --> 00:13:18.799
So your worldview gets, you think it's very broad, but it gets very constrained into just you know, this little pipeline of what you care about, and it keeps giving you more of those things, so then it reinforces that.
00:13:18.879 --> 00:13:31.360
So the same thing could happen with with uh the LLMs where it it reinforces your bias and things of that nature, it has context about what you like, so it'll include those into what you're doing.
00:13:31.600 --> 00:13:34.159
So you just got to be careful in that respect.
00:13:34.320 --> 00:13:40.080
Check your work, have somebody else check your work, and then also have those process checks, automated process checks.
00:13:40.240 --> 00:13:45.200
And then the human that takes a look and sits down and goes, you know, that doesn't seem quite right.
00:13:45.360 --> 00:13:46.480
And check all the links.
00:13:46.799 --> 00:13:47.440
Yeah, every link.
00:13:48.399 --> 00:13:49.279
Boom, boom, boom, boom.
00:13:49.519 --> 00:13:55.120
Yeah, and that but I that needs to happen early in the workflow, doesn't it?
00:13:55.279 --> 00:13:55.679
Oh, yes.
00:13:56.159 --> 00:14:00.559
Let's check, let's check this first before we go further.
00:14:00.639 --> 00:14:22.639
And you find that when you're code when I'm vibe coding, for example, it you soon learn when you just get dragged off down a path that you didn't validate, and then you hit roadblocks down the line, and you're like, why didn't you why and you're you're kind of blaming the AI tool, go, why didn't you know about this?
00:14:22.799 --> 00:14:31.679
And so I got to use when I use lovable, for example, I'm I'm on lovable, I'm on Clawed, I'm on Grok.
00:14:32.240 --> 00:14:42.720
And that's sometimes I'm validating all three sources of in information because I Grok's kind of helpful because I know lovable is based on Claude.
00:14:43.279 --> 00:14:48.960
I'm using Claude co-work, for example, but I know Groc's completely independent of those two.
00:14:49.039 --> 00:14:51.360
So it's a good, it's a good check and balance.
00:14:51.440 --> 00:15:08.639
And but just having that check early in the workflow, and Thomas Onenwold, who was on the podcast last week, he's been on a number of times, he talks about you know the importance of remembering that AI, you know, deterministic and non-deterministic.
00:15:08.799 --> 00:15:26.080
And so understanding how these models are uh work, which is what you're explaining, and their their kind of predilection to bias towards giving you what you want and saying that you're right, you've got to you've got to check that.
00:15:26.240 --> 00:15:37.840
And what when I did my um, this was in the early days of Chat GPT, I kind of did like a business, I was doing some business planning and Chat GPT was just being so agreeable.
00:15:38.000 --> 00:15:44.799
And I said to it, I said, look, imagine you're Elon Musk and he saw my business plan, what would he say?
00:15:44.960 --> 00:15:48.080
So and I know it's not passing an Elon test, right?
00:15:48.240 --> 00:15:52.559
And and Chat GPT came back and said some one line, very interesting.
00:15:52.639 --> 00:15:55.360
It said, and it was a very interesting line to me.
00:15:55.519 --> 00:15:59.519
And then one line was, oh, you want some tough love, do you?
00:15:59.679 --> 00:16:01.519
And I was like, Yes, I do.
00:16:01.679 --> 00:16:08.799
And then it got into giving me a non-agreeable answer.
00:16:10.000 --> 00:16:11.759
Yeah, yeah, absolutely.
00:16:12.000 --> 00:16:17.360
And and the thing about that is these are all great tools.
00:16:17.519 --> 00:16:20.879
There, there are methods though that you can adopt.
00:16:21.039 --> 00:16:30.000
One is think about creating personas, adversarial personas, so personas that help you check your work as who's your audience, okay.
00:16:30.320 --> 00:16:33.360
Have create personas that represent your audience.
00:16:33.519 --> 00:16:49.919
So if you're doing a newsletter like I am, I create personas to check my work where you know, and have a cross-section of the people who read my newsletter and be like, take a look at my newsletter that I spent all my time writing and give me some feedback, and then I go back and I adjust against that.
00:16:50.240 --> 00:16:54.960
And that helps one, to make sure that what I'm writing is clear.
00:16:55.120 --> 00:16:57.440
So there's a quality improvement over that.
00:16:57.759 --> 00:17:00.639
Two, I check all my links.
00:17:01.039 --> 00:17:03.759
I I definitely I check all my links manually, 100%.
00:17:04.640 --> 00:17:10.400
But at the same time, it can be like, you know, you can really dive into those those references.
00:17:10.559 --> 00:17:11.519
Are they good references?
00:17:11.599 --> 00:17:25.279
Because sometimes, even if you go and check the article, article looks good, the article that you're using the reference may not may actually be hallucinated as well because we're in this age where you really need to dig into those references.
00:17:25.440 --> 00:17:29.200
And for things that matter, you also got to wait things.
00:17:29.359 --> 00:17:41.519
You're not gonna necessarily be able to check everything all the time, but what are the things that matter if they get out that are gonna impact you and your company and your reputation?
00:17:41.680 --> 00:17:45.920
Those things should be hard stops and hard checks in that process.
00:17:46.319 --> 00:17:47.440
Doesn't have to be everything.
00:17:47.599 --> 00:17:54.960
You can really leverage the tools that are out there to check your work, but definitely, definitely have those hard checks for the things that matter.
00:17:55.119 --> 00:18:03.440
So, like when allegedly a KPMG has this happened, you're able to catch it before it gets out the door.
00:18:03.759 --> 00:18:13.839
Because you have the person signing off on it and says, I am staking my personal reputation and my my position on the fact that this thing is good.
00:18:14.000 --> 00:18:17.599
Just like editors do today in big publications.
00:18:17.759 --> 00:18:35.200
They check that work and make sure that it represents the company, the culture of their publication, and then also just the overall what they're trying to say out in the world and that it doesn't conflict with the things that they're you know, their mission, vision statement or what they're trying to do.
00:18:35.440 --> 00:18:57.920
Yeah, and and look, then I totally respect your your work in being diligent in doing that, because as you say, it's so easy to go see a store, a news art article that that is deliberately biased as well, because of the pub who the publication, their leanings, where their funding comes from.
00:18:58.160 --> 00:19:03.839
And um, and I know how much work it is to do what you're to do what you're doing.
00:19:04.079 --> 00:19:06.400
Um look, should we should we pivot back?
00:19:06.799 --> 00:19:10.079
I'm really interested in what you guys are doing today.
00:19:10.319 --> 00:19:30.000
You're you were I think I'm talking out of school when I say you're you're ex Nike, and um, and I'm guessing that's where without you, I think you were 10 years there, without sort of going into what they were doing around AI, what was your kind of like first experiences of generative AI?
00:19:30.160 --> 00:19:40.559
Because I'm guessing that must have been when you were you were working there, and then you've moved out of the sort of the big, huge giant corporate to to set up what you guys are doing.
00:19:40.640 --> 00:19:42.799
What was that journey like for you?
00:19:43.200 --> 00:20:00.880
Yeah, I started out in at Nike in 2013 as a contractor, and funny enough, I I it became a very successful opportunity that then led to a position and they asked me what I wanted to do.
00:20:01.119 --> 00:20:06.559
So, in that moment, of the things that were available, I was tackling some.
00:20:06.960 --> 00:20:11.440
I was a data guy, uh, you know, my roots are very data oriented.
00:20:11.839 --> 00:20:21.599
And uh so I I was brought in to help with performance and Nike, Nike.com and their back end databases and things of that nature.
00:20:22.000 --> 00:20:31.279
And so when I took the position, I was like, I want to tackle like test data management and you know, service socialization and the the CICD pipeline to production.
00:20:31.440 --> 00:20:33.680
So that's where I got my start.
00:20:33.839 --> 00:20:37.759
Uh, and you might see some things about in some of the things I'm doing.
00:20:38.000 --> 00:20:47.440
And then as I moved through my career at Nike, usually it was something I'd solve a problem, and then I didn't they'd be like, okay, hey.
00:20:47.680 --> 00:21:06.720
So I uh I helped with a very large project, and um that was a very successful project, and so I ended up with my next role, and that kept happening, and I got known as uh known for taking on large global projects and implementing them.
00:21:07.039 --> 00:21:18.240
And so then, long story short, I ended up as a global director of site reliability engineering at Nike, covering a number of different areas, and that was a really good fit for me.
00:21:18.480 --> 00:21:32.720
Had a a very um uh great experience working, you know, across retail, the dot-com, you know, infrastructures, global infrastructure, and then you know, supply chain and and also you know the wholesale piece.
00:21:33.200 --> 00:21:35.440
And that was a very subscribing position.
00:21:35.680 --> 00:21:45.519
And as I was in that position, uh, I think about three years or two and a half years back, JAI was the really everybody was like, okay, what do we do?
00:21:45.599 --> 00:21:47.440
This is this is a new tool.
00:21:47.920 --> 00:22:01.599
Organizations were seeing their their end users using ChatGDP to help them with their daily work, and there was no framework for managing that, and there was no idea on how to adopt a within the organization.
00:22:01.839 --> 00:22:04.880
So one day I was, and it's perfectly okay to share.
00:22:05.119 --> 00:22:10.720
One day I was talking to one of our C M VPs and I just gave him the whole like, here's what we need to do with Genny I.
00:22:10.880 --> 00:22:12.079
And he's like, okay, great.
00:22:12.240 --> 00:22:15.920
Uh why don't you build a strategy for us and for our division?
00:22:16.079 --> 00:22:17.680
And I was like, yeah, yeah, yeah.
00:22:20.400 --> 00:22:20.880
Yeah.
00:22:21.119 --> 00:22:22.960
So I proceeded to do that.
00:22:23.039 --> 00:22:35.039
Um I built out a strategy and learned a lot, did some pilots, got some early wins and things of that nature, also took on automation for the division.
00:22:35.359 --> 00:22:43.440
And that grew and grew and grew and came into this kind of uh cross-organizational strategy, if you will.
00:22:44.000 --> 00:22:56.079
And uh and then I really fell in love with the possibilities that uh AI brings to us and can give us all of us in in our daily work.
00:22:56.160 --> 00:23:10.720
And I'm not so much of a that there's very much the you know you hear oh nobody'll have to work anymore to things any paradigm shift that we have out there has never resulted in a long-term job loss.
00:23:10.960 --> 00:23:16.880
And I don't think and I know I'm probably one of the few I don't think Genai is any different.
00:23:17.039 --> 00:23:19.279
Yes, initially there's job losses.
00:23:19.440 --> 00:23:25.839
Yes, initially there is disruption but overall on the long term new jobs different jobs and more jobs.
00:23:26.000 --> 00:24:05.599
And if you look at what's actually happening with the developers who are using Gen AI, with the people who are using Gen AI in their daily work, they're busier never because and I think the NVIDIA don't quote me on this and I say that the Nvidia's CEO was talking about how they're seeing more code than ever being created by some of their engineers and that they have this expectation that some of their engineers should be spending hundreds of thousands of dollars on on spam just and they would expect that from their top engineers because they're doing more with AI and assisted in what they're doing.
00:24:06.559 --> 00:24:19.279
But at some point you know the the human factor comes in where you're going to need more of that because it's so easy to get the code because it's so easy to get the things through AI the the demand increases.
00:24:19.440 --> 00:24:30.319
So over time that demand because businesses now can get access to code at at a faster pace they don't have to wait months you know they there's more demand.
00:24:30.480 --> 00:24:53.680
So demand increases and then that's how that's how I see that you know boomeranging back around you see with more positions because as companies can more easily cheaper create the things that they need there's more demand which those positions come back in in a different way in a lot of ways but I do see I do see that this is will be a net increase in jobs over time.
00:24:56.319 --> 00:25:04.720
Yeah um so I'm an optimist on that front as well and it yes problem is is is the speed of this thing.
00:25:04.799 --> 00:26:09.519
Like there's lots of comparisons to industrial revolution and which saw a huge amount of automation and the rise of factories and you know that that was was going to be the huge impact on job and create lots of job losses did the exact opposite just grew everybody's economies massively and actually just massively increased the number of of jobs available but that took that took some time it's just like you know every every month it's just seems to be accelerating um so so I think that's the challenge and then the doomer talk kind of was started by the big players so the Darios and Sam's and Elons of this world have been talking the Doomer language and they kind of go first with the Doomer part and then they sort of then they started tagging on oh but don't worry because it's going to create a whole bunch of jobs that just don't even exist today that you'll be you'll be able to do.
00:26:09.920 --> 00:26:53.839
And then and now they've flipped on it and said no no no it's not going to take people's jobs it's all going to be great and that's kind of come a little bit late um given we're all animals very sensitive to any risks to us uh we're we're gonna look on the um on the possible downsides of new technologies aren't we what why why would we do anything else um so so I think you're right but what's your view on the how fast that change happens and and and and people go from oh I now realize that my job is not going to be is going to be done by the AI in the future but the AI's create all these new job openings for me.
00:26:54.079 --> 00:26:56.720
How does that how does that actually work in practice?
00:26:57.039 --> 00:27:21.519
Yeah and I think if you go back to and I wrote about this the companies who are doing AI adoption are there's not a lot of them out there that do it well you know if you look at that 95% number 95% of companies are aren't getting it right and 5% are and when you look at it from that lens you're like AI is everywhere and we're all using it whether we whether we realize it or not and just about everything that we do.
00:27:22.240 --> 00:27:45.920
But if companies are struggling to adopt even though AI is moving at this very fast pace how long is it going to take those companies to catch up and and maybe that's the slowdown piece that that reframes that gives people time to catch up and I do think people need to think about that in their own roles how can you use AI today?
00:27:46.079 --> 00:27:47.680
How can I do better prompts?
00:27:47.839 --> 00:28:02.000
What can I do from an individual to set myself apart in the roles that I'm doing or maybe even you know you realize that your role is you know that that role is probably not going to be needed.
00:28:02.079 --> 00:28:56.720
So what will what is that similar role or that passion thing that you really want to go after that you can leverage uh and dive into it assisted by AI I would say because to different you have to incorporate AI and and I know that uh out there in the world it's there there's some resistance to that and that's completely understandable it's a huge change like you said and it's happening fast and like you said as well that we have new models coming out new capabilities new announcements sometimes on a daily basis sometimes I think there was one weekend where we had nine different announcements over the weekend of new things that came out that were major and it's like you almost get news whiplash you're like that's a lot to absorb because you have full on capabilities of things that you can use and in your daily work potentially that are brand new that you have to learn.
00:28:57.039 --> 00:29:56.000
So I think the best approach is take small steps focus on the things that helped you the best learn you know be a sponge learn uh as things come out focus on the things that are relevant to you and try to get try to cut out the noise because there is a lot of noise as well and I think that'll help people a lot especially in as I'm helping organizations adopt AI I'm helping them uh you know on my daily and it's it it is a struggle right even at the organizational level because often you have the CEO the CO the CTO who have this expectation from potentially Wall Street or their investors that says you need to adopt AI because we want the sufficiency but then you have people on the other end of that who the end user who are who are saying you know what it takes me twice as long to use Chat GP to do X when I can just do it myself and be done with it.
00:29:56.319 --> 00:30:32.000
And that's a real thing that happens and a lot of that is not the organization and the individual as well not having the resources to get the training and sure you can prompt and self-train all you want but there are tips and tricks and methodologies that you can use that give you a lot better experience as you're using the and there are also tools out there that make it a lot easier too contact switching is one of those things that uh you have to do more of uh using AI.
00:30:32.079 --> 00:30:45.119
So sometimes you know like you said those different screens you have grok, touchdp, Gemini, Perplexy, and you're going through those and those are some of the things that I'm tackling is how do you get that in one place where I work today.
00:30:45.359 --> 00:31:40.160
And that's more of a passion thing because I don't want to switch between different screens I I want to be be able to work where I'm working and ask the questions and then cross check with all those LOMs at just one prompt because then that way there's less worry and then how do I incorporate that in my daily how can it help me with following up on leads help me with my sales funnel how can that help me grow those relationships that I have out there and uh make the best of this whole new world if you will yeah yeah I think again lots lots of good points lots of good advice there as well the point you made first resonated which is for the Doomer the kind of Doomer perspective on this and yeah the reality of how fast AI could actually automate out a human.
00:31:40.640 --> 00:33:07.119
I haven't seen it yet I've seen what I would class as AI washing um uh with companies who kind of were too big anyway especially the ones that got big during COVID and um needed to be right sized and they say well we're using AI we didn't have the wrong size company this is you know I I think there's a lot of AI washing going on so I think people should be really cognizant of that I am yet to see a and I'm looking I'm yet to see a job role completely eliminated by AI and I think your your your just kind of like approach is the right one which is if everybody just picks up one of these tools and goes right how can I actually use this in my day job to kind of take the mundane stuff away do some heavy lifting for me you know make me more effective and then I I keep the kind of the human touch and the domain experience but within the context of our business operations and you know how clunky these old companies are and siloed around their business operations you know that that requires people to really understand the business that they need to need to be there and and involved in any AI project as well.
00:33:07.279 --> 00:33:31.359
So so yeah I I'm I'm with you Ben I I just can't see this like oh we all wake up in January and everyone kind of says oh it's it's swept away every job that involved looking at a computer screen is now done by AI AI I just don't think it's that's not realistic and I don't actually think that is what is going to happen either.
00:33:31.680 --> 00:33:43.759
Yeah and I tend to agree the the a lot of what you talked about also in that the AI washing a lot of that is not necessarily the need to eliminate those roles that is part of it.
00:33:43.920 --> 00:34:00.960
But another part of that is companies are having to keep their margins and invest in AI so they're downsizing the people that they have so they can pay for compute and GPUs and things of that nature, build out data centers and power and all the things that go along with that.
00:34:01.200 --> 00:34:38.559
But as those prices go cheaper I think you'll see a little bit of boomerang from that too where those I mean that's a little predictive and I could be wrong but you know right now there's a race to you know there's a race on infrastructure and compute and resources that go into that because these big AI companies are throwing everything they can at it and it's also a resource thing too which is you know there there's a number of metrics out there we don't really have enough resources in the pipeline to meet the the the projected demands for AI.
00:34:39.280 --> 00:36:02.719
And unless we're good stewards with what we got which means you know you probably don't want to send your query to the most the strongest model out there to do a file move or to check your email or to even create an email you don't want to use for that you want to use grok fast or or one of the the cheap models open source if you can open source any of the LMs or models of open source real cheap models that do it well you know you're you're pennies versus dollars right and also on a compute side you're saving compute one of the one of the newsletters I did on Uber that uh that did pretty well as well which their Uber CEO allegedly came out and said hey uh we spent all of our I spend within like the first three or four months and they didn't really have any uplift in consumer facing spend or things that were material there that he can see for the spend that they made and a lot of that is because they're if you just throw resources at AI and you don't think about that implementation and the stewardship as you go, you're gonna end up spending tons of money and get not get the result necessarily that you want.
00:36:02.880 --> 00:36:32.719
Whereas when you take a proactive approach and you're like okay as I go through my day and as I'm building this out for my company how how can I make sure that the right model is the the best model for the the better price not necessarily the cheapest price but the better price gets me to my end result faster so that that spend is much easier for a company to um absorb if you will and also they get a lot more benefit from it.
00:36:32.960 --> 00:37:06.800
And the one I wrote about the the Walmart blueprint which actually did very well over I think 71,000 impressions on LinkedIn which just blew my mind was just about how Walmart took a thoughtful approach and they actually implemented AI in a in a thoughtful way using agents and then super agents and then also cord them off by different areas and actually increased it increased their correlationally increased their um their uh revenue in some areas I think by close to 30%.
00:37:07.599 --> 00:37:27.440
And so they ended up seeing that whole end-to-end result because they took a thoughtful approach and I think it took them like two years and that's yeah newsletter yeah can imagine yeah but uh yeah that that's kind of the kind of have in my newsletter I talk about those interesting topics of how people you know where people go wrong so we can learn from that.
00:37:27.679 --> 00:38:12.639
And then also I try to find things where hey you know what this company seems to get it right and maybe we can all learn from that as well because as we are trying to help our customers and we're trying to help our companies and organizations do uh do better with AI and and go faster because you know we all have to be competitive because all the other companies out there are adopting as well we have this opportunity to be thoughtful and also keep our costs down and also help ourselves and our customers and that whole pipeline from end to end be better overall.
00:38:12.880 --> 00:38:23.119
And I think that's something that is more human I guess you could say than just throwing an eye at it because there's people in that process they have to be there.
00:38:23.519 --> 00:38:26.880
And to get it right you still need the people to get it right.
00:38:27.039 --> 00:38:34.800
You can't just say hey yeah go out there solve everything because it's just you're gonna end up with chaos and a lot of money out the door.
00:38:35.119 --> 00:39:47.199
Yeah yeah yeah and we're we're already seeing the first headlines of companies saying absolutely this AI thing's a little bit expensive isn't it is it are we doing better with just the human resource uh in some of these processes so yeah those realities are I think we're gonna see more and more of those realities and I think we need to get chat uh sorry open AI and anthropic through their IPOs and out the other side and then and and this will be an another podcast episode where I then see that battleground is going to be you've got those giant companies that need revenue they need power they need compute and you've got a bunch of free of charge Chinese models that are really really great do a great job that don't require all of that expenditure but they're more difficult for an enterprise to use so I think we need to get those IPOs done and then I think some commercial realities are then going to set in for those companies themselves and also for enterprises who are consuming AI.
00:39:47.440 --> 00:40:02.480
But look that that was a that was a really fascinating conversation Ben I know we've hit time where do people find you where do they sign up for your newsletter please and whatever you mention I'll I'll drop the links down here in the show notes.
00:40:02.800 --> 00:41:00.880
Yeah um I'm on LinkedIn and I'll send you that as well and then I also have a website benpicket.com you can check me out and then uh the company I work for of mine at my daily is squaw-ai.com uh where we we help people and you know help people adopt AI uh in the places that they work and one more thing if you're interested in how you know what works things of that nature reach out uh you can DM me I'm happy to to uh respond I usually respond if I'm super busy it may take me a few days but I'll get back to you but happy to engage with people as well and then the biggest thing I would say and I think the last thing I'd leave you with whenever you're working with AI always ask AI what question am I not asking because that'll help you you surface things that you're not thinking about and that the AI isn't really focusing on.
00:41:01.039 --> 00:41:10.079
And that's been really helpful to me as well because that question tends to surface a lot of things that you wouldn't necessarily think about as you're using it.
00:41:10.400 --> 00:41:35.920
But overall this new this new age if you will isn't going away it's exciting it's in every part of our daily lives whether we realize it or not and uh it's a huge opportunity for all of us whether we're in sales or whether we're in leadership whether we're you know doing our daily job wherever we're at it can help us in in our roles and what we're doing.
00:41:36.159 --> 00:41:44.079
And um I think the future is bright overall sure there's some turmoil right now but I think that'll even out over the coming years.
00:41:44.400 --> 00:41:45.280
Yeah I agree.
00:41:45.840 --> 00:42:09.119
Really positive notes end on there Ben and um yeah I look forward to reading your newsletters as as they come out each week and I'll drop all of those links below and um yeah great having you on the podcast you've been a brilliant guest thank you very much I've learned learned a lot from you yeah thank you Ben thank you thank you for having me you all have a great day yeah you too you too thanks Ben bye bye bye bye bye
