Latest / The DEX Show: A Show for IT Change Makers / Navigating AI Models: Neural Networks and Digital Experience (w/ Pedro Bados)

Transcript

Navigating AI Models: Neural Networks and Digital Experience (w/ Pedro Bados) · The DEX Show: A Show for IT Change Makers

0:01
Welcome it, change Makers to the Deck Show with Tim Flower and Tom McGrath.
0:07
Let's get into it. Hello, change Makers, Welcome back to an
0:11
extremely special anniversary edition Deck Show. I'm Tom McGrath, joined by my main
0:18
man, Tim Flower. And Tim. One thing I always noticed about the
0:21
show lately is I stay in the same place and you kind of make your
0:25
way steadily around the world. You know, this is a theme I've noticed.
0:29
I'm just you're constantly on the move. Where are you, Totim?
0:34
Yeah, so I experiment with different backgrounds and keep everybody on their toes.
0:37
I am on, believe it or not, my ninth week on the road,
0:42
nine out of the last ten weeks on the road. Uh. Just
0:45
got back from a great decks Masters of Decks out in San Jose and made
0:50
my way here over to Dallas to see a really great American Airlines company and
0:55
then I'll be home by midnight tonight. Well, listen, this sounds like
0:58
a HR issue, Tim nine, it's actually you know, it's a testament
1:03
to the company. I can work from anywhere in the world. I can
1:07
drop in and see my parents in Arizona, I can see my son in
1:10
Boston. The flexibility is outstanding and we have Pedro to thank for that.
1:15
I was going to say, if you're wondering, listeners, why Tim is
1:17
being quite so great nine week stint, it is because we are joined by
1:23
Next Thing founder and CEO for the third time, Pedro Badas Pedro. Welcome
1:29
back. So nice to have you. Thank you very much, Tom and
1:33
Tim. I gotta say it's a it's a privilege to be here with you
1:37
both. I went to a gardener event a few few weeks ago and actually
1:42
everybody new team because of this podcast, not so much the New Next Thing
1:47
of myself, but definitely the dext podcast was was there. So congratulations,
1:53
massive success. Thank you so much. Shout out to everyone again listening,
1:57
We love you. Keep you in. Yeah, the secondary for you were
2:01
going to say, not so much you, Tom. I was prepared to
2:07
here's fine. I'm sure if you were there, Tom, if you were
2:12
there, they would have recognized you as well. I just happened to be
2:14
in the room. Don't bother me, up, It's okay. Is it
2:19
is, Pedro the twentieth anniversary of the wonderful company that you founded, and
2:24
I got to begin by asking, how does it feel to have founded a
2:29
company almost old enough to drink in the US. It feels great, and
2:36
honestly, I think right now we are in a We're in a position and
2:42
next thing in which everything works pretty well. I think we have a pipeline
2:46
of great technology. When I talk about AI right now, that is one
2:51
of my passions. We have a great team, We have more and more
2:53
customers, the analysts out there, they are recognizing the categ there are more
3:00
and more CIOs that they are understanding, and ploye experience is one of the
3:05
top priorities. And I think, honestly with more energy than ever before and
3:09
committed for another twenty years. Listen, I'm not just saying it. There
3:15
is a great vibe, but next think it's a great place to work,
3:17
and we've got incredible momentum this year. It's absolutely true, and we are,
3:22
as you say, Pedro going to be focusing and honing in a little
3:24
bit on AI and the future of decks, which I think is a fascinating
3:29
topic. But since it's a twentieth anniversary edition of the show, and since
3:36
since the company is rooted in an exploration of AI, I thought we could
3:40
begin there. Because not everyone will know Pedro the next think was effectively born
3:46
out of your own I believe AI postgraduate study at the EPFL in Switzerland now
3:53
over two decades ago. So I thought an interesting place to begin would be,
3:57
like, what was your initial interest in Why did you want to study
4:00
it? Because now everybody's talking about it not so much, and I guess
4:03
the very early naughties, right, absolutely, So, I think like any
4:13
other kid, when I was very young, I really fell in love with
4:17
technology, and in my particular case, artificial intelligence was kind of a way
4:24
of almost doing magic with technology, almost creating life, right, almost creating
4:30
you know, intelligence systems that they can really enhance. They can really today
4:35
we say co pilot certain tasks, and that was really I thought it was
4:41
really a lot of fun and a great challenge. And I spent a lot
4:46
of time when I was young, obviously coding you know computer games that they
4:49
were trying, you know, to emulate intelligence back in the days, very
4:55
very primitive. But I really fell in love with this domain and actually that
4:59
that was one of the reasons, or the main reasons why I moved to
5:01
Switzerland when I was twenty years old to really complete my stallies at the Artificial
5:06
Sell Intellience laboratory in the Swiss valis you to have technology, and there,
5:12
obviously I was exposed to a lot of different technologies, case based reasoning,
5:18
hidden Markov models, neural networks, you know, some of their techniques for
5:23
speech recognition, and I really loved it. And to what degree did you
5:29
have you continued to follow in the intervening decades the emergence the continued emergence of
5:33
AI. Has it been a passion over the last couple of decades. I
5:36
mean now it's increasingly at the forefront of what we do and what we make.
5:41
But is it something you're able to keep up with? From an almost
5:44
academic perspective, I have to say a little bit what is interesting in AI?
5:51
You know, before the transformers they came out, it was quite quite
5:58
an open I would say, Butterfield, there were a lot of techniques.
6:02
I think computer vision probably was the most promising technology in AI until the the
6:12
lllms they came out. You know, their possibility to look at X rays
6:16
and being able to diagnostic cancer and you know problems and things like that,
6:21
they were kind of the most promising application of AI. So computer vision definitely,
6:28
LM models, speech recognition, you know, automatic translations and things like
6:31
that using AI techniques were very promising, so I was kind of following that,
6:38
you know, when Google invested also, you know, buying a big
6:43
company in AI. Yet I was able to learn how to play games.
6:46
All these types of things were really exciting. But obviously in the last eighteen
6:49
months everything accelerated so much that I was even spending more time than before.
6:56
If you could go back, I mean, obviously AI was kind of baked
6:59
into the the prototype that became next thing you designed that apfl. If you
7:04
could go back and change something to almost future proof us to the way AI
7:11
has subsequently developed, what would it be. Is there something you could Is
7:15
there some change you would make in the past to better prepare us for the
7:17
future or as it worked out perfect. I think that's a great question.
7:24
I think everybody who was somehow working on AI was not very optimistic about neural
7:31
networks. So in my particular case, I really did a lot of research
7:36
in hidden mark of models, which is kind of a neural network in which
7:41
you have some sort of hidden states and then you basically the hidden states they
7:47
have some observable states that actually you can see and therefore you try to infer
7:54
you know, the hidden states and the weights behind which you know. It
8:00
was great, it was working up to a certain point in some particular problems.
8:05
But the fact that we were able to emulate the complexity of human beings
8:11
and actually in the entire world into a neural network like what happens today with
8:16
the transformers and the scalability there was kind of unexpected. Right even for those
8:22
who actually invented the transformer, the initial use case was was quite different was
8:28
in translation. So that's that is something that I guess everybody kind of missed.
8:35
And obviously, I mean from next in perspective, if we had known
8:37
that, probably would have invested a little bit more in neural networks and these
8:41
type of technologies. But right now we're you know, catching up. We're
8:46
doing all right, We're doing all right. I mean, maybe a good
8:48
thing to do actually for listeners who don't already know Pedro is give a give
8:52
give a picture of what AI is already with next thing making possible in digital
8:56
employee experience like as we speak today. Absolutely, and you know, sometimes
9:01
people they think of AI like gen AI and obviously because of you know,
9:07
CHTPD and all these models, but there is a lot of different techniques in
9:11
AI. So the first thing that for us is really interesting when you know
9:16
you are a data company, you actually need AI models that they're able to
9:20
understand data, They're able to give predictions, they are able to understand root
9:26
cause analysis when a metric or a certain value in the data is abnormal.
9:31
So we invested a lot in AI algorithms that they are able to capture,
9:37
you know, all these trends in the different customers and understanding when something breaks
9:45
in a particular customer being able to predict it in another customer. There is
9:48
a model called cloud intelligence, and we actually invested quite a lot in AI
9:54
algorithms and statistics to make sure that at the end of the day, not
9:58
only we say there is something not but we are able to explain it,
10:01
because that's one of the challenges sometimes with AI, it's not about giving diagnostic
10:07
but also explaining to a human being why. So this is one area.
10:11
The other area that I think is closer to LMS is what we call next
10:15
thing assist. So we think the paradigm between software and users of software is
10:22
going to change. People They're not going to spend a lot of time in
10:28
the UI unless you are really an expert. So you're going to basically tell
10:33
the system what you want. I want to understand which employees they have issues
10:37
with these applications, which employees are not adopting this particular digital service, How
10:41
can I improve this particular office in terms of digital experience, And you're going
10:46
to say it in a very natural language. So you need now to democratize
10:52
the world of tex and obviously llms are a perfect candidate for that because you
10:56
can basically take these dialogues, this voice now, especially with chachipt four oh,
11:03
and then translate it into actions or dashboards or anything that you can do
11:09
in a de ex solution like next thing. So, Pedro, I want
11:11
to explore AI from a couple of different angles. The first is some of
11:18
the misconceptions, right So, right now, AI models are seen as having
11:24
immense power, increasing power every day almost exponentially, and some people may believe
11:33
that AI can eventually just incorporate decks into themselves. Right. We talked to
11:39
in the past about that black hole theory where it's going to just consume everything
11:43
and leave a dex specific vendor like us behind. What assumptions does that overlook
11:48
and how does it influence our investments moving forward. Yes, so let me
11:56
maybe clarify this black hole concept. And it's true that you know, for
12:01
the last twelve months, many companies they try to build IP around LM models,
12:07
and then LM models they kept expanding, and therefore this IP is kind
12:11
of part of the M model, right. More and more, I think
12:18
what is going to happen with tex and any domain which is very specific,
12:24
is that you're going to need to take those models that they are pretty good
12:28
at generic information common sense information ninety nine percent of the case, and you
12:35
want to transform that into an expert which is quite right one hundred percent of
12:41
the time. So one of the things that is really important is fine tuning,
12:48
and fine tuning is very primitive today. So taking this foundational knowledge of
12:52
an LM and saying, oh, now this model is going to become an
12:58
expert in dex or is going to be an expert in this particular task.
13:03
That's something that is not clear today in the market. You need a specific
13:09
data and you need a specific technologies which probably are not lllms to really get
13:16
to that point of proficiency. I think what is really important to understand as
13:22
well is lms are extremely good at everybody knows right, predicting the next world,
13:28
so basically looking like they're saying the right thing. However, they are
13:33
not so good at complex reasoning. The human being is much more I would
13:41
say that's changing, but it's better an understanding these kind of concepts and then
13:48
making a plan based on you know, the evidence is in front of us.
13:54
LM models is more like an iterative way of saying what is common sense
13:58
right, So it is still unclear what will be the level of reasoning in
14:05
an LM. I think they are progressing a lot, but it's still unclear,
14:09
and some people they actually think LM technologies will not be very good at
14:13
reasoning and complex reasoning, and X will need complex reasoning because it's a complex
14:18
problem. On the other side, also you need data, and obviously a
14:24
company like next Thing has all the different data that you're required to train those
14:30
models. So you know, LM will be a very important part of the
14:33
decks of the future. But still I think there will be other set of
14:37
technologies and data that they are required. So with that misconception maybe handled a
14:45
little bit there, we can't also discount the big players. Right, there
14:52
are four or five big data and analytics you know, the Microsoft's, the
14:58
Googles, the the Metas of the world. How do you see those large
15:05
players encroaching into our market? Right? What are the implications for specialized vendors
15:11
like US and a category like dex as the bigger models evolve and potentially could
15:18
start to specialize as well. Our view on that is dex it's a fast
15:26
growing market. It's one of the most important markets for an IT department.
15:33
But I think right now the battle for Open AI, Meta, Microsoft AWS,
15:41
it's even more fundamental than that, right, It's really the race to
15:46
get a more intelligent, a more multitask type of system. So I don't
15:52
think they're really focused on specializing the models for specific domains. I think they're
15:58
respecting the ecosystem to take those models and to actually specialize them. And that's
16:04
at least what we're seeing. So I think we have a great opportunity to
16:08
actually continue leveraging all the research, all the great innovation from Microsoft, OpenAI,
16:18
WS and open source. I think we have to say something about Meta
16:22
and you know their commitment to open source to really have a super powerful desolution.
16:30
So it's not very often. I would think that folks have an opportunity
16:33
to get some advice from someone that's been in this space for so long.
16:38
When we go to conferences like you said, we had the Gartner last week
16:41
with service Now's knowledge, it seems that every vendor has tacked on a dot
16:47
AI onto their product or their company name. It feels a lot like bolt
16:52
on. It feels a lot like marketing and a little less reality. What
16:56
advice would you give to an enterprise IT team as they develop up their holistic
17:00
AI strategy maybe not specific to ducks, and how do they cut through the
17:04
noise and really figure out if capability is real, is it something that actually
17:10
fits in their strategy? And how does an AIDS product compliment that broader strategy.
17:15
What advice do you have to navigate this this funny new world. It's
17:21
a great question, team, absolutely. I you know, there is like
17:26
any other you know, emerging space, there is a lot of great things,
17:30
but also there is a lot of hype and sometimes you know it feels
17:36
for me then there is a lot of companies they are just taking chat GPD,
17:38
putting some technology around and selling it like you know, this ropative innovation.
17:45
I think it's very important to understand the full chain, to understand really
17:49
what llms are able to do and what they are not able to do,
17:55
because llms they do have some limitations, especially when it comes to reasoning,
18:00
as I said, or data. You know, they are not able to
18:03
reason with a lot of numerical data. Just maybe let me give you a
18:07
small anecdote. Chat GPT three point five is not able to to do the
18:12
addition of two large numbers, right, It was very hard. So because
18:17
they are optimized for text, right, and that some of these things they
18:21
are changing, but still, you know, they don't have the concepts that
18:25
we have as human beings. So understanding those limitations are really important because you
18:30
know, we need to separate. I would call it the killing use cases
18:36
where you know, LM models generative AI. It's really changing the game from
18:42
you know, sumerization. I think chatbots up to a certain point versus other
18:49
you know use cases in which probably is more a co pilot at best,
18:53
or actually it's not really a great solution because they're not really designed for that
18:57
problem, right. I don't foresee in the short term, for instance,
19:03
chat ChiPT being a great financial analyst things like that. Right, it might
19:10
happen down the road, but today specialize in other things. So you've been
19:15
in this world for over twenty years, but it seems like the last twelve
19:22
to eighteen months it's really just exploded into everybody's consciousness, right, Generative AI
19:27
in the last year or so is all anybody can talk about. What have
19:32
you seen, what's the influence over that last twelve months, and how has
19:37
it changed your vision of the future. What do you think the next three
19:41
to five years looks like for this industry in terms of benefit to customers,
19:45
product development? How has the last year changed your outlook for the future.
19:52
I think it's not only me, it's an entirely humanity. I was reading
19:56
yesterday. Actually the website of Penny Eye got more than ninety five million people
20:03
in one day, which is a new record for a website. I think
20:08
human beings we are very good at. You know, we get very excited
20:12
with something that looks like human. You know, it could be an alien,
20:18
it could be a chabod, it could be oh, it looks like
20:21
a human, and then we got really excited about it. So we have
20:25
some sort of because we really care about humans, right, And I think
20:29
chapt probably was this moment in which, oh, it looks like human,
20:33
it looks like a real intelligence, human intelligence. So obviously that created a
20:38
lot of emotions. But we don't have to forget that. And I'm part
20:42
of the people that think that this is just a great tool. It's a
20:48
great tool. I think it's very useful for many domains. We saw some
20:53
demos the other day about you know, automatic or real time translation. That's
20:57
amazing. You know, content sumlization will be able to for instance, every
21:03
every kid could have a tutor, you know, to learn how to do
21:07
math, how to do reading, writing. So it's a great tool,
21:11
and I think it will continue evolving. I think more and more capabilities.
21:19
I'm not sure about general what they call general intelligence, because at the heart
21:27
of an LLM model is a great simulation of a human being or a great
21:33
simulation of the world. But it's a simulation. We have to understand it's
21:37
a simulation. It doesn't work like a human brain. It works in another
21:41
way that looks like so, and that's more of a philosophical question. Is
21:48
a simulation. It's a simulation. I would say a perfect simulation is the
21:59
same as a human being, because you cannot make the distinction, right,
22:03
So at the end of the day, it doesn't matter if it's a simulation
22:06
or not a simulation. At the end of the day, if it's a
22:07
perfect simulations, the same as a general intelligence. So there's this philosophical debate
22:15
and part of the again and part of the people, I think that humans
22:19
are much more than intelligence. We have emotions. We actually do care about
22:23
humans. When you look at what's going on with the chess for instance,
22:26
right we have computers for a while that they were able to be the best
22:32
chess masters, and still we are very passionate about, you know, people
22:37
playing chess, and we almost put computers on the side. We don't care
22:40
about them. We care about the people. So we really care about people.
22:44
And I think that will continue and we'll have these tools that they will
22:47
be better and better. Like computers, they are better and better, but
22:51
always really really on the side, not on top of us. That's kind
22:56
of the future that I think back to Next Thing's twentieth anniversary. There was
23:03
a phrase actually Tim brought to the discussion when we were preparing to the show,
23:07
but it heard you use it was integrated human system, and you were
23:11
saying that, I think you were talking at some event, Tim was that
23:15
and saying that, you know, the usual cliche of a company being like
23:18
a family with something you put aside, and you preferred the notion of an
23:22
integrated human system, and we were interested in hearing what that meant to you
23:26
and how that is reflected in next thing. Yes, I think sometimes it
23:36
was really hard for people when they join companies to realize that a company is
23:45
very similar to a human body in the sense that you might work in the
23:51
marketing department, you might work on the sales department, you might work as
23:55
an engineer. But at the end of the day, for the company to
23:57
perform, every system has to perform up to a certain level, right,
24:04
And the outcome of one system is the input of another system. So let
24:08
me give an example. Right, the output of the engineering department probably is
24:15
taken by product management. The product management is given that to you pre sales
24:19
or the sales department, and they're taking that to the customer, back to
24:25
probably customer success, and that feedback is, you know, back to product
24:29
management and engineering. So it's actually a loop. And for a company to
24:33
perform, every system which is couple loosely coupled, but couple somehow needs to
24:40
perform well. And when there is one system that doesn't work well, the
24:45
entire company suffers, and the entire company feels that, right, And it's
24:51
a little bit the same. I try to illustrate that with the human body.
24:53
Right, if you muscles or your bones or your or whatever doesn't work
25:02
one hundred percent, the entire body suffers. And I think that's the kind
25:06
of the image that I try to explain to people when they join. So
25:11
therefore they put the company first, their own team second, and probably themselves
25:17
third when they make decisions. That's very cool. And what would you say
25:22
outside of that, what are you guiding principles as founder and CEO today twenty
25:26
years into your next thing journey? What continues to inspire and form your thinking
25:30
as the leader here? You know, I'm part of these people that I
25:37
think what we like is the journey in the life. In life, I
25:41
mean, at some point we're all gonna die, we know that, right,
25:44
We're gonna look backwards and we're gonna say was it worth? And one
25:49
of the things that I think it's really worthy for me is to build something
25:57
with a great team and to to help a lot of customers to make sure
26:03
that you know, we are really improving one part of their lives. And
26:08
get up every morning and have a great team with great ideas and great execution.
26:15
And I honestly feel that right now and next thing, we have it
26:18
more than ever before. So we have so many interesting opportunities ahead of us,
26:22
so many great people in the company that honestly, it doesn't feel like,
26:27
oh, it's twenty years therefore we are tired. It is the opposite.
26:30
I think we have more energy than twenty years ago. So that's it
26:36
might be surprising, but it's actually it's actually great. You know, there
26:41
is sometimes in startups a little bit of a transactional mindset. You know,
26:48
oh, you want to build a company, and then you want to sell
26:51
the company or you know, an IPO and then you know something like that.
26:56
That's not the way we want it run. Next thing, we really
27:00
wanted to build a company for one hundred years in which people they can thrive.
27:07
We call it a school of talent. People they can grow. You
27:10
look at management and many other people, they have been in the company ten
27:12
years and still their world class managers and executives. That's really what we want
27:18
to build, and I think we are actually succeeding, and people internally,
27:22
they really appreciate that we just had Greg Wis on the show, your first
27:29
ever employee, and still here incredibly and still enjoying themselves. So, you
27:33
know, as a testament to that, one area that's obviously evolved with Next
27:37
Thing, So own growth and evolution is our experience events, Pedro, I
27:41
believe preregistration is now live and people can check the show notes if they want
27:45
to get their tickets to Boston or London this year now. But for you,
27:53
what is your vision for experience going forward on what excites you about doing
27:59
the show. I'm not sure how how many isn't to experience that we know,
28:02
Petra, I'm going to take about ten, yeah, I would say
28:07
so, I think about ten experience events. I think there were like probably
28:11
two three during COVID that we call it the experience everywhere online and then now
28:17
we are back to you know, meeting customers and partners and everybody in person,
28:21
which is great and actually this year is special because the first year that
28:25
we're going to do more than Next Thing experience is actually the decks conference in
28:30
the world. I think there are a couple of things. One is,
28:33
we saw this gap in the market in which there is actually not a great
28:40
conference in which head of end user computing, VP of infrastructure, you know,
28:45
people that they really care about digital experience, they could meet and they
28:48
could learn, and that's what we are trying to do this year, like
28:52
the conference for end user computing right professionals. So that's one and the other
28:57
really is that we look at the market and we see their technologies like a
29:04
TSM and many others that actually the paradigm is changing. Right. Employees don't
29:10
want to open tickets companies. They want to get rid of incidents. They
29:15
want to be proactive. So actually we think dex is actually the X plus
29:18
AI probably would be the solution for that in the next three four years.
29:22
And to get there, we need the community. We need really people believing,
29:26
we need people showing the projects. There is tremendous success in many companies
29:32
around the world, you know, reducing incidents, improving experience, and I
29:37
think that's the forum in which they're going to show all that. I think
29:41
I just heard the IT firefighters drive by you there on fire and Buster.
29:48
That's a great point. And again, people, if you've never been to
29:51
an experience and then make sure this is your first visit this very year.
29:56
And also look, we went into some Next Thing laws day. If you
30:00
want to know more about the origins and growth of Next Think, which we
30:04
only touched upon tangentially in this conversation, you can check out I Think classic
30:08
episode with Pedro and co founders Vincent and Patrick Free Magic Kings. It's called
30:15
It is a really, really superb discussion and we'll include links to that in
30:19
the show notes. Also just remains for me to say thank you so much
30:23
Pedro for coming back on the show. It was lovely to have you in
30:26
a great discussion. Thanks Pedro, thank you very much. Is my honor
30:30
to be with you. Guys. To make sure that you never miss an
30:33
episode, subscribe to the show in Apple Podcasts, Spotify, or your favorite
30:38
podcast player, and if you're listening on Apple podcasts, make sure to leave
30:42
a rating of the show. Just have the number of stars you think the
30:45
podcast deserves. If you'd like to learn more about how Next Thing can help
30:48
me improve your digital employee experience, head over to Nextthink dot com. Thank
30:53
you so much for listening until next time