전사 🔗
WEBVTT
NOTE
This file was generated by Descript
00:00:00.037 --> 00:00:03.397
Alberto: I think there are some things
that haven't changed and they will
00:00:03.397 --> 00:00:07.627
never change, and some things that are
radically changing in the low level.
00:00:07.627 --> 00:00:09.967
Everything is changing in
the, in the, in the specifics.
00:00:09.967 --> 00:00:12.301
Everything is changing, in the skills.
00:00:12.721 --> 00:00:13.741
Nothing has changed.
00:00:19.242 --> 00:00:22.932
Chris: So, um, Alberto, welcome
to the, uh, the Neuro po.
00:00:22.957 --> 00:00:25.032
It's, uh, it's a pleasure
to have you on today.
00:00:25.062 --> 00:00:25.602
How are you doing?
00:00:25.602 --> 00:00:25.722
I.
00:00:25.956 --> 00:00:26.306
Alberto: Right.
00:00:26.306 --> 00:00:26.666
Thank you.
00:00:27.540 --> 00:00:28.140
Chris: You are welcome.
00:00:28.230 --> 00:00:33.180
Um, so yeah, we're going to, um,
discuss, uh, your background today.
00:00:33.600 --> 00:00:37.830
Um, some of the work you're doing
on agents and, um, but you've
00:00:37.830 --> 00:00:40.710
also been a career coach, so
something we've got in common there.
00:00:40.770 --> 00:00:45.150
Um, we can discuss that and, and the
market and as always, some future
00:00:45.150 --> 00:00:48.300
predictions and, and where you see
the current markets specifically in.
00:00:49.460 --> 00:00:50.300
Agen ai go in and know.
00:00:50.300 --> 00:00:54.290
It's kind of like how long is a piece
of string at the moment, because,
00:00:54.410 --> 00:00:57.050
um, the market's so, so fast moving.
00:00:57.230 --> 00:00:58.940
But, um, it'll be great
to get your thoughts.
00:00:58.940 --> 00:01:02.660
So, um, yeah, over to you Alberto.
00:01:02.660 --> 00:01:06.770
You're a, you're an engineer by
training, but now AI researcher
00:01:06.770 --> 00:01:10.460
and, and leading some products,
uh, projects over at Criteo.
00:01:10.910 --> 00:01:14.300
Uh, do you wanna just give it a quick,
high level overview of your career and,
00:01:14.480 --> 00:01:16.190
and how you got to where you are today?
00:01:16.605 --> 00:01:17.205
Alberto: Yeah, sure.
00:01:17.535 --> 00:01:22.005
So, so the thing is, uh, my case, I, I
have a kind of an, an orthodox career.
00:01:22.305 --> 00:01:25.090
So I started my career like 20
years ago as telecom and engineer.
00:01:25.455 --> 00:01:28.005
Telecom eng uh,
telecommunication engineer.
00:01:28.455 --> 00:01:31.305
And I started working in
Telefonica, in, in Spain, like
00:01:31.305 --> 00:01:32.895
as a, as a research engineer.
00:01:34.005 --> 00:01:37.575
so the thing is that after five,
six years, I wanted to do something
00:01:37.575 --> 00:01:40.935
else, something like beyond just
like, let's say plain engineering.
00:01:41.625 --> 00:01:44.340
then actually I got back
to university to study pH.
00:01:45.480 --> 00:01:49.080
Uh, but then I said, okay, it's, it's
kind of hard to mix philosophy with
00:01:49.080 --> 00:01:51.390
a technical, my, my engineer life.
00:01:51.750 --> 00:01:55.770
So then I started to find AI as
an answer to my, to my question.
00:01:55.770 --> 00:02:01.440
So it's the closest that I could find
in the technical world the closest
00:02:01.440 --> 00:02:03.960
that I could find to, uh, philosophy.
00:02:04.410 --> 00:02:07.350
So actually that's why I got
interested in, in, in ai.
00:02:08.580 --> 00:02:10.860
and then I enrolled,
uh, in a, in a master.
00:02:11.235 --> 00:02:15.735
And to, to, to, to follow my, then my, my
PhD And what happened at that time is was
00:02:15.825 --> 00:02:19.640
it was like 2020, uh, 20, 20 10, 20 11.
00:02:20.775 --> 00:02:28.305
So, you know, the 20, the, the 2008
crisis, uh, hit then we are in Spain
00:02:28.305 --> 00:02:31.335
in this post-crisis, uh, world.
00:02:31.875 --> 00:02:35.025
Um, people were starting to get angry.
00:02:35.505 --> 00:02:38.835
We had like super high levels of
unemployment rates in young people.
00:02:39.165 --> 00:02:43.215
Like, um, the crime crisis
also hit very hot in Spain.
00:02:43.485 --> 00:02:46.485
Like people were like evict,
evicted from their homes 'cause
00:02:46.485 --> 00:02:47.715
they couldn't pay the mortgage.
00:02:48.135 --> 00:02:49.695
And then also we have like many.
00:02:50.565 --> 00:02:51.555
corruption cases.
00:02:51.975 --> 00:02:55.335
Uh, and then people were like,
there was this feeling of this
00:02:55.335 --> 00:02:59.055
generation that the democratic
system was not good enough, right?
00:02:59.385 --> 00:03:04.970
So it started this the 50 may
movement, like 20, 20 11, and,
00:03:05.205 --> 00:03:06.405
and was involved in this, right?
00:03:06.405 --> 00:03:08.170
So I was like, study my master in ai.
00:03:09.420 --> 00:03:13.260
And, and at the same time, uh, getting
involved in this, in this movement.
00:03:13.260 --> 00:03:16.635
So what happened is that a group
of people, like more engineers and,
00:03:16.830 --> 00:03:20.940
and the like, we started to create a
group to analyze the social movements.
00:03:21.390 --> 00:03:22.200
Uh, online.
00:03:22.950 --> 00:03:26.700
we started to do like social network
analysis methods to analyze, to
00:03:26.700 --> 00:03:30.210
understand what what's going on, uh,
with this kind of social movements.
00:03:30.630 --> 00:03:34.260
And so the, this got me
interested in sociology and
00:03:34.260 --> 00:03:35.850
social network analysis and so on.
00:03:37.080 --> 00:03:40.200
I had the opportunity to
start a PhD HC exactly.
00:03:40.200 --> 00:03:41.820
On this, right on, on, on.
00:03:41.940 --> 00:03:42.300
Um.
00:03:43.200 --> 00:03:48.120
Analyzing, uh, uses behaviors, uh,
online with machine learning techniques.
00:03:48.390 --> 00:03:53.130
So that's how I just jumped in, into
the, into this machine learning world.
00:03:53.580 --> 00:03:57.900
Um, so I moved to France and I started
my PhD, and then I did a postdoc.
00:03:57.900 --> 00:04:02.220
And then finally I just, uh, joined
joint c some later as a, as a
00:04:02.220 -->
Load full transcript
00:04:04.590
research engineer, as a researcher,
00:04:05.046 --> 00:04:05.766
Chris: Nice.
00:04:06.210 --> 00:04:06.600
Alberto: sorry.
00:04:06.946 --> 00:04:07.486
Chris: Yes.
00:04:07.486 --> 00:04:08.566
Um, makes sense.
00:04:08.566 --> 00:04:13.066
And, uh, I guess that the recommended
systems connection to Criteo, it kind of,
00:04:13.126 --> 00:04:17.896
um, made sense I guess with the types of,
uh, clients and projects they work on.
00:04:18.424 --> 00:04:23.164
Um, all right, well, um, yeah,
let's talk in to, um, you
00:04:23.164 --> 00:04:24.964
know, your career now at, um.
00:04:25.759 --> 00:04:29.209
Criteo, and obviously you've got
a, a great scientific background
00:04:29.209 --> 00:04:31.129
with you, your experience so far.
00:04:31.639 --> 00:04:35.959
Um, your helps shape the scientific
agenda at Criteo, uh, specifically
00:04:35.959 --> 00:04:39.799
around agents, which is a, a huge
buzz in the market right now.
00:04:39.799 --> 00:04:44.209
A bit of a buzzword, um, but you are
very much chartering, uh, the unknown
00:04:44.209 --> 00:04:47.689
there with some of the cutting edge
research that you're doing, um, and
00:04:47.689 --> 00:04:50.149
developing, uh, real world products.
00:04:50.539 --> 00:04:51.139
Um.
00:04:52.774 --> 00:04:56.674
Do you want to just talk us through
kind of, um, some of the work around
00:04:56.674 --> 00:05:01.474
shopping assistance that you're doing in
agent interfaces and specifically, you
00:05:01.474 --> 00:05:06.514
know, everybody's kind of definitions of
agents may be different at the moment.
00:05:06.694 --> 00:05:09.184
How is it affecting
e-commerce specifically?
00:05:09.184 --> 00:05:11.944
And, um, you know, what type
of agents are you developing?
00:05:12.189 --> 00:05:17.349
Alberto: the agent systems are
gonna change the way people shop,
00:05:17.709 --> 00:05:21.939
people buy, and then therefore
it's gonna affect the Tels business
00:05:21.939 --> 00:05:23.694
or the ad tech, uh, business.
00:05:23.694 --> 00:05:26.079
And why?
00:05:26.559 --> 00:05:29.379
Well, because and more
you will have people.
00:05:30.159 --> 00:05:32.889
going directly to the website
to buy products, but they will
00:05:32.889 --> 00:05:38.589
ask assistance, their agents to
do it by, uh, for them, right?
00:05:40.299 --> 00:05:45.669
then it means that it doesn't make sense
for curtail to show any ad, uh, to the
00:05:45.669 --> 00:05:47.499
agent because the agent is, is a bot.
00:05:47.499 --> 00:05:48.879
So he, he doesn't care.
00:05:49.089 --> 00:05:50.829
He's not impressed by your display.
00:05:51.669 --> 00:05:54.219
you need to maximize now,
you need to convince.
00:05:55.179 --> 00:05:56.649
The This middleman.
00:05:56.679 --> 00:06:01.239
This middleman, who is the, the shopping
assistant you have the best product,
00:06:01.239 --> 00:06:02.589
that you have the best recommendation.
00:06:03.069 --> 00:06:06.999
So you have to find, you have
to be sure that you are giving
00:06:06.999 --> 00:06:08.319
us super good recommendation.
00:06:08.349 --> 00:06:09.204
And you have also to explain.
00:06:09.824 --> 00:06:14.024
Why you think like all the, all
the, you have to give all the things
00:06:14.384 --> 00:06:18.344
that you didn't give in traditional
recommended systems, such as why
00:06:18.674 --> 00:06:20.084
you think this product is the best.
00:06:20.474 --> 00:06:23.804
Comparing products, comparing
alternatives, like all, all doing all
00:06:23.804 --> 00:06:25.424
these kind of product search that we do.
00:06:25.424 --> 00:06:27.944
When you buy a computer, you are
comparing three alternatives.
00:06:27.944 --> 00:06:29.594
You got super type of doing this.
00:06:29.594 --> 00:06:33.344
So we need to, uh, give all
the information to the agent.
00:06:33.764 --> 00:06:36.554
To, to, to make, to, to, to, yeah.
00:06:36.739 --> 00:06:39.589
Put everything together for
the, for the user to make this,
00:06:39.589 --> 00:06:40.789
to take this final decision.
00:06:41.869 --> 00:06:45.919
Um, that's so in, in, in shopping, that's
one thing that's, that's will happen.
00:06:45.919 --> 00:06:52.159
So in curtail, what aim for is
to be the best, recommend the
00:06:52.159 --> 00:06:55.099
system for this kind of agents.
00:06:56.754 --> 00:06:57.834
to, to call.
00:06:57.864 --> 00:06:58.104
Right.
00:06:58.914 --> 00:07:01.674
and these recommended systems,
these agents will, and we are seeing
00:07:01.674 --> 00:07:04.614
already in open ai, we're seeing,
we are seeing already the first
00:07:04.614 --> 00:07:06.294
movements towards these directions.
00:07:06.294 --> 00:07:09.804
You will have this, like the main
open ai, Google and whatever.
00:07:10.044 --> 00:07:12.744
So they will need these
catalogs and they will.
00:07:13.479 --> 00:07:17.139
Not just contact retail, but they
will contact like many providers.
00:07:17.439 --> 00:07:21.069
So that's why you really need to be
better than the, than than the others.
00:07:21.069 --> 00:07:21.279
Right.
00:07:22.719 --> 00:07:24.609
also we can also provide our own agents.
00:07:24.669 --> 00:07:27.039
There is also why, like, why
not, like we're also exploring
00:07:27.039 --> 00:07:29.439
these alternative on developing
some products in this direction.
00:07:29.919 --> 00:07:32.169
Like, uh, doing this
end-to-end solution where we.
00:07:33.129 --> 00:07:35.559
Give this agent to the
user in some context.
00:07:35.949 --> 00:07:41.229
And we also give the the recommendations
and, and then also we are
00:07:41.229 --> 00:07:43.209
developing agents for our clients.
00:07:43.509 --> 00:07:47.649
So our clients now they have to deal
with, for instance, when you onboard
00:07:47.829 --> 00:07:50.229
in any at the company, at you tell.
00:07:50.874 --> 00:07:53.214
Well, have to create a campaign.
00:07:53.544 --> 00:07:55.284
You have to upload your catalog.
00:07:55.314 --> 00:08:00.444
You have to like do a lot of
bureaucracy to get things started.
00:08:00.924 --> 00:08:02.844
And this is painful.
00:08:03.054 --> 00:08:05.904
Like this is like, yeah, you
have to curate your catalog.
00:08:05.904 --> 00:08:10.254
You have to decide your audience
to give all the, yeah, a lot
00:08:10.254 --> 00:08:11.664
of information, metadata.
00:08:11.904 --> 00:08:16.044
So we want to automate, we are
automatizing this so that we.
00:08:17.124 --> 00:08:21.834
Let's say we shorten the cycles
from days to hours or minutes.
00:08:23.372 --> 00:08:25.547
Chris: And, you know, just
take it back a second.
00:08:25.547 --> 00:08:27.707
What, what do you think's
actually gonna add?
00:08:28.112 --> 00:08:35.522
Happened to the ad tech industry now,
um, you know, AgTech, um, interfaces
00:08:35.522 --> 00:08:37.502
and agents are being introduced.
00:08:37.532 --> 00:08:39.962
Um, how's it gonna shape the future?
00:08:39.962 --> 00:08:45.122
Will it just be a case of agents
will now be interacting, um, with
00:08:45.122 --> 00:08:48.752
huge catalogs of data and, um, yeah.
00:08:48.752 --> 00:08:49.857
What, what are your thoughts there?
00:08:50.163 --> 00:08:53.924
Alberto: What do you I don't, I dunno
if I, can you, can you repeat the
00:08:54.095 --> 00:08:57.455
Chris: In, in terms of the, the,
the ad tech industry, obviously
00:08:57.455 --> 00:08:59.255
it's a, it is a huge industry.
00:08:59.675 --> 00:09:02.765
Um, you know, if people aren't
looking at ads anymore, how do you
00:09:02.765 --> 00:09:06.815
actually think it's, it's going to
be affected moving forward in terms
00:09:06.815 --> 00:09:10.145
of the size of the industry and um,
00:09:10.619 --> 00:09:10.909
Alberto: Yeah.
00:09:11.585 --> 00:09:11.915
Chris: yeah.
00:09:13.079 --> 00:09:15.989
Alberto: We have like different, different
markets, different realities, right?
00:09:15.989 --> 00:09:20.309
So the traditional ads, they won't
disappear because you will still,
00:09:20.309 --> 00:09:25.199
like with your own eyes, uh, read
a website or visit a retailer
00:09:25.199 --> 00:09:26.699
website or things like that.
00:09:26.699 --> 00:09:27.209
So it will exist.
00:09:28.954 --> 00:09:33.689
and you will have this traditional
displays, um, even better because this
00:09:33.689 --> 00:09:37.829
traditional now with lms, we're able
to understand more about the context.
00:09:37.829 --> 00:09:40.769
For instance, like before it was
kind of hard to understand what
00:09:40.769 --> 00:09:42.389
was the website talking about?
00:09:42.674 --> 00:09:46.259
And, and now you can say, okay,
this website is talking about this
00:09:46.259 --> 00:09:51.419
topic so I can put a more relevant
ad, um, in this, in this website.
00:09:51.419 --> 00:09:51.659
Right?
00:09:52.079 --> 00:09:56.639
Um, but then also you
have this space where.
00:09:57.464 --> 00:10:00.764
The user is not interacting with the
website, but with the agent itself.
00:10:01.094 --> 00:10:05.024
But yet still the, the, these
agents need some way to monetize
00:10:05.024 --> 00:10:07.454
their, their, their platforms.
00:10:07.454 --> 00:10:07.664
Right?
00:10:07.664 --> 00:10:10.844
It's, it's expensive nowadays to
call an lm, even if it's, well,
00:10:10.844 --> 00:10:13.724
the price is getting, uh, lower.
00:10:13.784 --> 00:10:18.344
Uh, but, uh, but still, I mean, you, you
need a business model for this, right?
00:10:18.344 --> 00:10:23.534
So, we think that many, many agents.
00:10:23.924 --> 00:10:31.964
Chatbots in retailers, in WhatsApp, in
OpenAI, whatever, will need this the way,
00:10:32.024 --> 00:10:39.314
this way to monetize, to make their space
available for, uh, for, for brands, right?
00:10:39.644 --> 00:10:41.924
To put their, their, their
recommendations there.
00:10:42.674 --> 00:10:46.879
So yeah, I think that the, the two walls
will, will, will, will quite co coexist.
00:10:47.377 --> 00:10:47.827
Chris: Okay.
00:10:48.067 --> 00:10:48.727
Interesting.
00:10:48.817 --> 00:10:53.557
Um, and yeah, just, just to pick up on
where you left off, uh, a second ago,
00:10:53.977 --> 00:10:57.757
uh, designing agents for commerce,
um, you've kind of touched on some
00:10:57.757 --> 00:10:58.682
of the challenges there around it.
00:10:59.032 --> 00:11:00.472
Expense, et cetera.
00:11:00.922 --> 00:11:04.852
But what, what are the, um,
real challenges in developing
00:11:04.942 --> 00:11:09.112
agents for e-commerce over,
uh, say traditional search and
00:11:09.112 --> 00:11:11.092
recommendation engines and systems?
00:11:11.613 --> 00:11:13.658
Alberto: You can, you can even
see there's an opportunity rather
00:11:13.958 --> 00:11:15.218
than as a challenge if you want.
00:11:15.608 --> 00:11:17.438
'cause the thing is that you can do more.
00:11:17.528 --> 00:11:17.888
Yeah.
00:11:17.888 --> 00:11:20.678
It's really, it's like you can do
like traditional recommendation if
00:11:20.678 --> 00:11:22.868
you want, but you can also do more.
00:11:23.288 --> 00:11:26.228
So you can use context that
you couldn't use before.
00:11:26.228 --> 00:11:31.238
You can planning, you can do,
like now you can, I mean, you
00:11:31.238 --> 00:11:32.228
can interact with the user.
00:11:32.228 --> 00:11:34.058
That's super, uh, one,
one important thing.
00:11:34.178 --> 00:11:39.278
Now the user now in many,
in many, in many, in many.
00:11:40.028 --> 00:11:43.148
Use cases, the user just, uh,
for instance, in a retailer,
00:11:43.148 --> 00:11:47.738
the user enters a keyword like,
I'm looking for milk for a bike.
00:11:48.038 --> 00:11:50.078
And then you will have this
conversation, and now you will
00:11:50.078 --> 00:11:51.728
have this conversational interface.
00:11:52.178 --> 00:11:53.078
We'll say, okay, bike.
00:11:53.078 --> 00:11:54.128
What kind of bike?
00:11:54.338 --> 00:11:56.618
What it's for my daughter.
00:11:56.648 --> 00:11:57.068
Okay.
00:11:57.073 --> 00:11:57.203
What's.
00:11:58.043 --> 00:11:59.093
How old is your daughter?
00:11:59.513 --> 00:12:02.513
Like, give me all these clarifying
questions that will have the recommended
00:12:02.513 --> 00:12:05.273
system to find a more relevant product.
00:12:06.023 --> 00:12:08.753
And it will also be able to
remember your preferences.
00:12:09.113 --> 00:12:15.443
It will remember if it's your personal
or you have a, you are locked in a, in
00:12:15.443 --> 00:12:16.853
a, you have an accounting disservice.
00:12:17.033 --> 00:12:19.253
It will remember your,
your, yeah, your preference.
00:12:19.253 --> 00:12:20.303
What you said before.
00:12:20.663 --> 00:12:24.353
If remember you have a family that you
live in London, that you don't like this
00:12:24.353 --> 00:12:26.603
brand, that you like the other brand.
00:12:27.518 --> 00:12:31.478
these things so that the user
doesn't, doesn't have to repeat
00:12:31.658 --> 00:12:34.028
things to the agent, uh, every time.
00:12:34.028 --> 00:12:34.268
Right.
00:12:34.688 --> 00:12:38.678
So the experience will be much more
enjoyable for the, for the user.
00:12:38.978 --> 00:12:41.498
And then on our side, on the
recommended side, we'll have to
00:12:41.498 --> 00:12:44.888
be able to, yeah, to deal with all
these contents, with all the memory.
00:12:44.888 --> 00:12:48.758
For instance, there is a lot of open
research in terms of how do you manage
00:12:48.758 --> 00:12:51.158
the memory, what do you have to remember?
00:12:51.848 --> 00:12:54.158
Memory recommended
systems was didn't exist.
00:12:54.473 --> 00:12:57.323
You just have a classic machine
learning system where you have a
00:12:57.323 --> 00:13:00.713
lot of examples, input, output, and
you train your system With this, now
00:13:00.713 --> 00:13:02.393
you have an agent for every user.
00:13:02.843 --> 00:13:03.568
So this agent.
00:13:05.003 --> 00:13:08.783
We'll have to remember to have
a specific memory for that user.
00:13:09.083 --> 00:13:11.813
And that means many
strategies of memory, right?
00:13:11.813 --> 00:13:16.553
So you will have like episodic memory, uh,
contextual memory, like short-term memory.
00:13:16.553 --> 00:13:19.523
Long-term memory, depending on
what, like the user didn't like
00:13:19.523 --> 00:13:21.173
this product and he told me why.
00:13:21.383 --> 00:13:25.613
So I should remember, it's somehow in
some kind of memory the user clicked here.
00:13:25.613 --> 00:13:28.523
I should remember the user
said he has a dog, so I should
00:13:28.523 --> 00:13:29.423
remember it somewhere, right?
00:13:29.723 --> 00:13:30.323
And then.
00:13:31.223 --> 00:13:32.033
Memory is not enough.
00:13:32.063 --> 00:13:38.363
Then in every context, you need to be able
to retrieve the proper, uh, memory piece
00:13:38.573 --> 00:13:40.973
that's gonna be relevant for this context.
00:13:41.183 --> 00:13:44.543
So how do you retrieve this so that
the context, uh, the working context
00:13:44.543 --> 00:13:48.623
of the, of recommendation is as
clear and as useful as possible?
00:13:48.623 --> 00:13:48.773
Right.
00:13:49.478 --> 00:13:51.758
And then also you have the adaptability.
00:13:51.758 --> 00:13:56.378
That's another, another big,
big stream of research that's,
00:13:56.438 --> 00:13:57.968
uh, that's starting these days.
00:13:58.328 --> 00:14:00.278
So this agent needs to be adaptable.
00:14:00.338 --> 00:14:03.848
Uh, you may change, you users
will change the preferences.
00:14:03.848 --> 00:14:04.568
It's not the same.
00:14:04.568 --> 00:14:10.718
For instance, uh, you shopping, uh,
patterns in Christmas, then in summer,
00:14:11.018 --> 00:14:15.848
uh, maybe if it's your agent, you are,
you have been with your agent for years.
00:14:16.263 --> 00:14:17.043
It'll have to adapt.
00:14:17.043 --> 00:14:19.203
Your, your, I mean your, yeah,
your preference and so on.
00:14:19.353 --> 00:14:20.013
They adapt.
00:14:20.343 --> 00:14:22.023
You buy objects they have to remember.
00:14:22.023 --> 00:14:28.473
So this kind of adaptability for
life, um, phases and even for,
00:14:28.713 --> 00:14:30.393
for many different circumstances.
00:14:30.753 --> 00:14:32.818
That's another thing that
agents need to, to deal with.
00:14:33.655 --> 00:14:37.585
Chris: And, um, yeah, you've
already touched on how it's gonna
00:14:37.585 --> 00:14:42.175
personalize and create a more enjoyable
shopping experience for people.
00:14:42.775 --> 00:14:47.845
Um, and I guess nobody really knows,
uh, the answer to this question, but
00:14:48.115 --> 00:14:51.235
how do you think it's gonna evolve over
the next couple of years in terms of
00:14:51.235 --> 00:14:54.445
shaping people's experiences and adoption?
00:14:54.925 --> 00:14:58.405
Um, you know, where do
you actually see it going?
00:14:59.988 --> 00:15:01.188
Alberto: The agent shopping.
00:15:01.968 --> 00:15:02.568
Um,
00:15:02.950 --> 00:15:03.490
Chris: Yeah.
00:15:04.188 --> 00:15:07.803
Alberto: I think what we are seeing now
that younger people, they're embracing it.
00:15:09.018 --> 00:15:11.418
Uh, I mean, it's a very
typical pattern, right?
00:15:11.418 --> 00:15:15.948
Like all the people, they are more
reluctant they, and, and the younger
00:15:15.948 --> 00:15:19.818
people, they are already, like most
of them, they're already used to chat
00:15:19.818 --> 00:15:24.678
with these systems every day and they
trust more in these systems and so on.
00:15:24.948 --> 00:15:27.708
So I think what will happen is
that now we are seeing a phase.
00:15:28.278 --> 00:15:30.318
We're seeing, we, we, we are, we, we have.
00:15:30.723 --> 00:15:35.073
Been through a phase where people were
more, what if the system hallucinates?
00:15:35.463 --> 00:15:37.473
Uh, what if like, yeah.
00:15:37.473 --> 00:15:39.543
A lot of fears and a lot of reluctancy.
00:15:40.293 --> 00:15:47.253
We will see that these systems were a
lot get better with time, much better.
00:15:48.093 --> 00:15:54.963
Um, and then actually there was this
study that, You know, the, the capacity
00:15:55.023 --> 00:16:00.903
of the systems, it's like being
multiplied by a factor of two every,
00:16:01.713 --> 00:16:05.943
what was it, every, every seven months
I think it was the, the pattern, right?
00:16:06.303 --> 00:16:13.653
So, um, then we will learn also to
better engineer the systems to increase
00:16:13.713 --> 00:16:15.213
the, more the efficiency and so on.
00:16:15.483 --> 00:16:18.693
So system, so people will, will
also trust more in the systems.
00:16:18.693 --> 00:16:21.363
And I think, I mean, I, I think like, um.
00:16:22.488 --> 00:16:25.518
Let's say in, in some years,
everyone will use an agent for
00:16:25.518 --> 00:16:27.798
shopping their daily lives.
00:16:29.178 --> 00:16:31.008
That would be my, my prediction.
00:16:31.945 --> 00:16:32.575
Chris: Yeah.
00:16:32.965 --> 00:16:37.675
And, um, I suppose touching on your
philosophy background here, you know,
00:16:37.765 --> 00:16:43.825
if people are adopting agents more,
and this is completely unrelated to
00:16:43.825 --> 00:16:48.385
machine learning, but do you think it
will affect the retail High Street and
00:16:48.385 --> 00:16:50.875
people actually doing in-person shopping?
00:16:50.875 --> 00:16:54.715
If you've got an, you know, an
agent doing all the legwork, you
00:16:54.715 --> 00:16:58.975
know, you're just clicking buy at
the checkout, et cetera, um, yeah.
00:16:58.975 --> 00:17:00.505
Do you think it'll affect the High Street?
00:17:00.633 --> 00:17:01.083
Alberto: yeah, yeah.
00:17:01.083 --> 00:17:07.143
I think it will affect, we already
seeing some movement from like, um,
00:17:07.353 --> 00:17:14.523
plexity or Open ai, um, trying to like
opening their, and accessing to, to, to,
00:17:14.523 --> 00:17:16.028
to, to catalogs, to product catalogs.
00:17:17.838 --> 00:17:24.798
And, um, and then also doing the checkout
themselves without you leaving their
00:17:24.798 --> 00:17:28.908
website, you are being able to buy a
product here in this catalog, another
00:17:28.908 --> 00:17:32.838
product from other catalog, and then
you do everything, uh, with, with,
00:17:32.898 --> 00:17:34.458
without leaving their, their website.
00:17:34.788 --> 00:17:37.818
And that's challenge for retailers
because they, they want the opposite.
00:17:37.818 --> 00:17:41.508
They want people go to
the website they also
00:17:41.860 --> 00:17:41.980
Chris: Hmm.
00:17:42.258 --> 00:17:43.698
Alberto: to control the user experience.
00:17:43.893 --> 00:17:49.473
This is also interesting data for the
retailers, that they can learn from.
00:17:49.833 --> 00:17:54.603
So there is gonna be a T of war,
and I don't think retailers are
00:17:54.603 --> 00:18:00.723
gonna give up easily, uh, this,
this, uh, uh, this thing to, to
00:18:00.723 --> 00:18:03.423
the, to the big, uh, tech companies.
00:18:04.083 --> 00:18:05.403
Um, yeah, but they will.
00:18:05.463 --> 00:18:09.853
But the thing is, I think, so it
will be an opportunity there are.
00:18:10.953 --> 00:18:14.643
Is moving in this direction is the
position of s is helping retailers
00:18:14.643 --> 00:18:19.203
actually to, to position themselves
very well in this new scenario.
00:18:19.533 --> 00:18:22.653
So the challenge is how do you.
00:18:23.793 --> 00:18:24.723
Monetize.
00:18:24.723 --> 00:18:27.573
How and how do you prepare
your catalog to be accessible
00:18:27.573 --> 00:18:29.913
for agents, but in a fair way.
00:18:30.183 --> 00:18:33.333
So you don't want your catalog to
be stolen, consumed by agents, like
00:18:33.333 --> 00:18:37.983
tracked by agents and doing the
checkout on any any other place.
00:18:38.403 --> 00:18:42.483
And out of your, out of your
website, you, so how do you.
00:18:43.288 --> 00:18:44.458
find a good, right.
00:18:44.458 --> 00:18:48.658
How do you, so, and that means that
they need to create their catalogs to be
00:18:48.778 --> 00:18:53.518
accessible for like, give some value for
agents to consume and then gain getting
00:18:53.608 --> 00:18:55.498
paid for getting, uh, paid for it, right?
00:18:55.858 --> 00:18:58.888
And that also implies the
idea of data marketplaces.
00:18:59.248 --> 00:19:02.788
And that also some thing that we're
exploring cur retail, this idea of
00:19:02.788 --> 00:19:10.503
data marketplace where, um, like our
brands, our partners, our clients will.
00:19:11.853 --> 00:19:17.978
able to control the way they share
their products with third parties.
00:19:18.368 --> 00:19:18.998
Chris: Interesting.
00:19:19.478 --> 00:19:25.238
And, um, you know, there's some retailers
out there with, uh, amazing, uh, data
00:19:25.238 --> 00:19:29.978
and AI teams, but do you think that in
general, imagine Criteo's got a lot of
00:19:29.978 --> 00:19:35.888
different types of customers that, um,
you know, the average retailer is ready
00:19:35.888 --> 00:19:41.348
for the agent wave of a user having its
own personal agent shopping assistant?
00:19:41.678 --> 00:19:42.278
Um.
00:19:42.740 --> 00:19:44.630
And people are ready for what's to come
00:19:44.995 --> 00:19:48.295
Alberto: You mean if client, if Q tells
clients they're ready for what's gone?
00:19:49.231 --> 00:19:52.591
Chris: it ju just in general, from
what you see in the marketplace, is
00:19:52.591 --> 00:19:57.301
the average retailer ready for, um,
you know, the agent wave, do you think
00:19:57.713 --> 00:19:57.893
Alberto: I
00:19:58.119 --> 00:19:58.239
Chris: I.
00:19:58.313 --> 00:20:03.278
Alberto: the market is ready yet,
but I think that we have no choice.
00:20:04.913 --> 00:20:07.283
It's not a, it's, it's a,
it's like a force of nature.
00:20:07.283 --> 00:20:11.873
It's gonna come, but you have
to adapt, and if possible, you
00:20:11.873 --> 00:20:15.413
have to be first and you have to
be the leader to shape the way.
00:20:15.848 --> 00:20:23.018
The world and the market is changing,
but, uh, but that's gonna change for sure.
00:20:23.438 --> 00:20:24.908
And everyone, everyone is gonna be adapt.
00:20:24.908 --> 00:20:26.903
I mean, if you and those
who don't adapt will.
00:20:28.133 --> 00:20:31.253
As, as usual, but
yeah, but well, that's a
00:20:31.399 --> 00:20:31.689
Chris: Okay.
00:20:32.363 --> 00:20:35.843
Alberto: companies as the role is to
help these companies, these retailers,
00:20:35.843 --> 00:20:38.663
to, to, to adapt to this new scenario.
00:20:39.323 --> 00:20:41.933
And I think there will be another,
like more opportunities for new
00:20:41.933 --> 00:20:45.233
companies, for new, there will be
new business opportunities that
00:20:45.233 --> 00:20:47.183
will appear in the, in the next
00:20:47.354 --> 00:20:47.684
Chris: Hmm.
00:20:47.753 --> 00:20:52.313
Alberto: idea of data curation, of how
do you, how do you create your data?
00:20:52.313 --> 00:20:53.693
How do you create like good.
00:20:54.386 --> 00:20:59.486
Understandable catalogs that
are easily searchable by agents,
00:21:00.296 --> 00:21:02.041
something that needs still to be done.
00:21:02.836 --> 00:21:03.136
Chris: Yeah.
00:21:03.136 --> 00:21:06.826
Well you've, that segues nicely onto
the next section of the podcast.
00:21:06.826 --> 00:21:11.356
You mentioned to me data is a competitive
moat and you know, a lot of people
00:21:11.356 --> 00:21:13.246
want to implement the latest models.
00:21:13.246 --> 00:21:17.656
And you said before models
are increasingly replicable.
00:21:17.986 --> 00:21:21.196
Um, but unique data is a real edge.
00:21:21.256 --> 00:21:23.326
Um, you know, what, what
do you mean by that?
00:21:23.469 --> 00:21:23.769
Alberto: Yeah.
00:21:24.369 --> 00:21:25.089
Yeah, exactly.
00:21:25.089 --> 00:21:31.149
So what I mean is that a model in
ai, I mean, I don't know of any
00:21:31.539 --> 00:21:37.514
company whose advantage is having the
best neural net, let's say, right?
00:21:37.989 --> 00:21:41.019
So maybe there is, in some niche
market, you have a specialized
00:21:41.019 --> 00:21:44.114
model for some specific use cases.
00:21:44.904 --> 00:21:49.884
They will probably exist, it's
not the main strategy, let's say.
00:21:49.884 --> 00:21:50.094
Right.
00:21:51.834 --> 00:21:54.774
in companies like retail,
that's, that's gonna be the case.
00:21:54.774 --> 00:21:58.914
So the model is not your mode, the model.
00:21:59.694 --> 00:22:02.394
at some point we are seeing this
with agents with l lms, right?
00:22:02.394 --> 00:22:08.184
So you are, you are having like, open
source models are today powerful.
00:22:08.544 --> 00:22:11.454
And the closed source
models were one year ago.
00:22:11.814 --> 00:22:14.334
So there is a, there
is a delay if you want.
00:22:14.754 --> 00:22:17.454
But sooner or later we get there, right?
00:22:17.459 --> 00:22:19.164
The, the open source
committee is gonna get there.
00:22:20.184 --> 00:22:23.124
And that happens with, all AI models.
00:22:23.334 --> 00:22:28.044
There are like, uh, common, common
knowledge there is no secret sauce.
00:22:28.464 --> 00:22:30.024
There is some secret sauce at some point.
00:22:30.024 --> 00:22:32.334
Now, open ai, for instance,
they train the models.
00:22:32.634 --> 00:22:37.134
They post train, they create
the data, they create like super
00:22:39.069 --> 00:22:44.829
Synthetic data, data sets, that
create very powerful models.
00:22:45.189 --> 00:22:50.889
But you have behind like open
source companies that are like
00:22:50.889 --> 00:22:54.099
catching up and saying, okay, you're
doing this, you're doing that.
00:22:54.099 --> 00:22:56.019
Now we are going, we're
gonna replicate it.
00:22:56.139 --> 00:22:59.139
That's something, for instance,
we saw with, with deep seek.
00:22:59.139 --> 00:22:59.379
Right?
00:23:00.219 --> 00:23:01.299
point, deep seek was okay.
00:23:02.124 --> 00:23:06.924
Found the method to do the same thing
that you guys do with less resources,
00:23:07.374 --> 00:23:08.784
and that's gonna happen every time.
00:23:09.204 --> 00:23:10.704
So the, so the mode is the data.
00:23:10.704 --> 00:23:18.804
So the thing is that the majority of
the, the most part of the data is, uh.
00:23:19.389 --> 00:23:24.999
Has been already used by, I mean, most
part of the public data, all the public
00:23:25.239 --> 00:23:29.979
data has been used by lms, has been
seen and used to train these, these
00:23:29.979 --> 00:23:36.339
all lms, they all have seen already,
posts, all websites, all the newspapers.
00:23:36.429 --> 00:23:39.519
That's already, that's,
that's done, that's used.
00:23:40.089 --> 00:23:42.339
So what remains is two things.
00:23:44.049 --> 00:23:45.609
Um, synthetic data.
00:23:46.344 --> 00:23:46.794
one thing.
00:23:46.914 --> 00:23:51.264
And then we are seeing like, um,
in research and like companies,
00:23:51.264 --> 00:23:59.094
they are figuring out how to
create synthetic data is realistic
00:23:59.094 --> 00:24:01.044
enough for your model to learn.
00:24:02.784 --> 00:24:08.484
You bootstrap your data from some real
and then with some loops, with some,
00:24:08.544 --> 00:24:12.984
uh, small techniques, you create some
realistic synthetic dataset, which
00:24:12.984 --> 00:24:14.754
is way larger than the original one.
00:24:15.114 --> 00:24:17.694
And then you can use your model,
you can train your model there.
00:24:17.874 --> 00:24:19.134
So that's one main technique.
00:24:20.124 --> 00:24:22.044
And the other one is private data.
00:24:23.004 --> 00:24:25.119
So there is a lot of
private data that has.
00:24:25.929 --> 00:24:29.769
A lot of value, that doesn't
have the public data.
00:24:29.829 --> 00:24:31.539
For instance, commerce data, right?
00:24:31.539 --> 00:24:33.669
The shopping patterns of users.
00:24:33.849 --> 00:24:35.679
This is something that
you don't find on Reddit.
00:24:36.399 --> 00:24:40.269
So this is something that Tel
has, Google has and and so on.
00:24:40.269 --> 00:24:44.199
So having this data is gold because
then you have the same model or
00:24:44.199 --> 00:24:47.859
more or less similar model to the
one that everyone has, but you have
00:24:47.859 --> 00:24:51.699
fed this model with unique data,
and that's what creates your
mode.
00:24:53.501 --> 00:24:53.831
Chris: Hmm.
00:24:54.011 --> 00:24:54.581
Interesting.
00:24:55.001 --> 00:25:00.011
And, um, you know, what, what, what
are some of the most overlooked data
00:25:00.011 --> 00:25:03.311
in, in ad tech, um, that you think.
00:25:04.719 --> 00:25:06.394
Alberto: Sorry, can you
repeat the question, Chris?
00:25:07.471 --> 00:25:07.861
Chris: Sure.
00:25:07.921 --> 00:25:08.161
Yeah.
00:25:08.161 --> 00:25:08.341
Yeah.
00:25:08.341 --> 00:25:11.491
So what, what do you think some of
the most overlooked data a assets in
00:25:11.491 --> 00:25:17.131
ad tech are, um, that are, you know,
the most important for the ENT era?
00:25:18.095 --> 00:25:19.960
Alberto: The assets,
assets of data assets.
00:25:21.470 --> 00:25:21.830
Data
00:25:22.172 --> 00:25:22.442
Chris: assets.
00:25:22.677 --> 00:25:22.947
Yeah.
00:25:23.277 --> 00:25:23.607
Yeah.
00:25:23.607 --> 00:25:23.637
I.
00:25:24.604 --> 00:25:31.354
Alberto: I would say, well, one thing
is, but I dunno how it will play out,
00:25:32.344 --> 00:25:39.964
but the with all those services, I think
that will be, that might be important.
00:25:39.964 --> 00:25:43.384
So if your agent is isolated,
00:25:45.514 --> 00:25:46.774
it can only do so much.
00:25:47.239 --> 00:25:53.389
But if it can access to, uh, yeah, to
all services, like getting integrated,
00:25:53.389 --> 00:25:58.909
sharing data all those services like,
you know, banks, your, if you are in
00:25:58.909 --> 00:26:04.189
sports like your Strava, uh, data,
your, you know, uh, calendars, like
00:26:04.189 --> 00:26:11.149
this kind of integration, I think that
will be, that will be key to give, uh,
00:26:11.149 --> 00:26:15.544
your agent or any shopping assistant
or any personal agent, uh, value.
00:26:16.474 --> 00:26:17.134
That's one thing.
00:26:17.554 --> 00:26:19.714
The other thing is pro, pro descriptions.
00:26:19.954 --> 00:26:21.754
That's, for sure.
00:26:21.754 --> 00:26:24.364
And that's something we're working
very hard in, in cri retail.
00:26:25.144 --> 00:26:32.134
Um, we, the catalogs today, they are not,
they were not ready for the Argent era.
00:26:32.614 --> 00:26:36.874
So brands, they have to
explore the catalogs to many,
00:26:37.174 --> 00:26:38.494
let's say a tech companies
00:26:41.149 --> 00:26:45.769
The thing is that the traditional
display as a title, an image, a price,
00:26:46.549 --> 00:26:48.049
it, but that's not enough for the agent.
00:26:48.499 --> 00:26:50.089
The agent needs more information.
00:26:50.089 --> 00:26:54.229
The product description, like clean,
clear, pro description today, the pro
00:26:54.229 --> 00:26:57.859
descriptions that we have in our catalogs
are like robotics, like four words.
00:26:58.309 --> 00:27:04.639
Um, like often they repeat the information
that is already in the title or in the
00:27:04.639 --> 00:27:06.319
other metadata, so it's very useful.
00:27:06.994 --> 00:27:10.714
Companies will have to provide more
data for, for agents to consume.
00:27:11.674 --> 00:27:14.584
And Google also have to learn
to, to create this data.
00:27:15.064 --> 00:27:17.524
Um, or for trouble, for instance.
00:27:17.524 --> 00:27:22.114
For trouble is a specific, a
very challenging, uh, vertical
00:27:22.444 --> 00:27:23.884
because the data is dynamic.
00:27:24.319 --> 00:27:29.719
It's not like if you, when you go to buy
a lab, if you want to explore, catalog for
00:27:29.719 --> 00:27:32.359
electronics, everything is nothing moves.
00:27:32.359 --> 00:27:36.139
It's like you're looking for a computer,
just do some semantic, search, some
00:27:36.139 --> 00:27:37.609
filtering, find your computers.
00:27:37.969 --> 00:27:41.719
Then the, your agent is gonna think,
okay, is it relevant for the user or not?
00:27:42.019 --> 00:27:43.759
But for travel, it's okay.
00:27:43.759 --> 00:27:46.759
Go to go for, I go to
go on holidays to Italy.
00:27:47.119 --> 00:27:48.409
Okay, let me take some flights.
00:27:49.454 --> 00:27:52.064
No, there are no available
fights for flights for that date.
00:27:52.124 --> 00:27:56.564
What about this, the next weekend,
what about a train instead of a flight?
00:27:56.654 --> 00:27:58.634
What about, so this hotel is booked also.
00:27:58.634 --> 00:28:00.164
So what about this alternative?
00:28:00.164 --> 00:28:05.414
So you have to do this combination
of, of your, of, of multi-product
00:28:05.414 --> 00:28:06.344
recommendation, right?
00:28:06.764 --> 00:28:09.374
And, and you don't know
all the information.
00:28:09.404 --> 00:28:13.184
You have to query APIs to see whether
the service is still available and so on.
00:28:13.634 --> 00:28:14.174
So that's.
00:28:15.139 --> 00:28:15.859
That's the other thing.
00:28:15.949 --> 00:28:17.299
That's the other intent of data.
00:28:17.299 --> 00:28:18.529
That's, that's the other challenge.
00:28:19.159 --> 00:28:25.099
And, and then also the, the other data
asset will be, so before I talk about
00:28:25.099 --> 00:28:27.109
this, the memory, uh, challenges.
00:28:27.109 --> 00:28:29.329
So that's, that's the thing, right?
00:28:29.329 --> 00:28:33.169
So the, the long-term use of
history, uh, past interactions,
00:28:33.199 --> 00:28:37.429
um, personal information, family
data, um, all these kind of things.
00:28:37.519 --> 00:28:40.784
And, and I think here, maybe
we'll start in a new era.
00:28:41.599 --> 00:28:46.039
Where users that will be more willing
to give more information in exchange,
00:28:46.519 --> 00:28:48.619
uh, of good quality services.
00:28:50.045 --> 00:28:51.305
Chris: Yeah, that's an interesting point.
00:28:51.305 --> 00:28:55.685
And um, I think historically people
have been quite hesitant about that.
00:28:55.685 --> 00:29:00.485
But I guess if it was yours actually
to see a tangible increase in a better
00:29:00.485 --> 00:29:03.995
shopping experience, you might be
more willing to, um, to pass over.
00:29:04.445 --> 00:29:05.045
Interesting.
00:29:05.435 --> 00:29:05.735
Cool.
00:29:05.765 --> 00:29:06.095
Okay.
00:29:06.095 --> 00:29:07.235
Thank you for sharing that.
00:29:07.655 --> 00:29:12.725
Um, moving on to the next part, research,
engineering and, and the talent crossover.
00:29:13.325 --> 00:29:14.015
Um.
00:29:14.885 --> 00:29:18.545
Well, I think you've been fortunate
to work on, um, a, a couple
00:29:18.545 --> 00:29:20.525
different sides of, of defense.
00:29:20.975 --> 00:29:25.235
Um, but in particular your, your
research background is really strong
00:29:25.235 --> 00:29:29.645
and, um, yeah, really keen for
you to share some insights there.
00:29:30.095 --> 00:29:33.485
Um, you know, I've worked with.
00:29:34.295 --> 00:29:39.185
Maybe hundreds of companies at this
point, and a lot of them define, um,
00:29:39.275 --> 00:29:43.625
you know, the engineering capabilities
of a researcher in different ways.
00:29:43.745 --> 00:29:48.635
You know, some of them almost
want, um, ML engineering, um,
00:29:49.085 --> 00:29:51.395
levels of, of capabilities.
00:29:51.845 --> 00:29:57.095
In your opinion, you know, what, what
should every good ML researcher be able
00:29:57.095 --> 00:29:59.465
to do from a engineering standpoint?
00:30:00.755 --> 00:30:04.115
Alberto: I think there are some things
that haven't changed and they will
00:30:04.115 --> 00:30:08.345
never change, and some things that are
radically changing in the low level.
00:30:08.345 --> 00:30:10.685
Everything is changing in
the, in the, in the specifics.
00:30:10.685 --> 00:30:14.435
Everything is changing, uh, in the skills.
00:30:14.855 --> 00:30:15.875
Nothing has changed.
00:30:16.295 --> 00:30:20.195
So in the skill set, I think it's,
that's true for any engineer,
00:30:20.195 --> 00:30:22.235
for any profession, probably.
00:30:22.595 --> 00:30:26.675
So curiosity productiveness,
so the idea of.
00:30:28.160 --> 00:30:31.350
I mean, I don't know how
many times I have changed.
00:30:31.850 --> 00:30:36.980
My topic in terms of research product,
I started doing social network
00:30:36.980 --> 00:30:41.240
analysis, then I did patient inference,
recommended systems, multitask
00:30:41.240 --> 00:30:43.640
learning, then NLP, LLMs agents.
00:30:43.760 --> 00:30:48.110
So every two years is kind of a new
cycle, hopefully building, uh, on
00:30:48.110 --> 00:30:49.460
top of the, of the last one, right?
00:30:49.460 --> 00:30:52.340
So there is some connection
between cycles, but, you have
00:30:52.340 --> 00:30:56.990
to recycle a lot every time and,
and things are moving very fast.
00:30:57.230 --> 00:30:59.600
So you have to be, you have
to have this curiosity to.
00:31:00.960 --> 00:31:05.220
To be up to date with,
uh, with what's going on.
00:31:07.020 --> 00:31:10.290
that will mean that that's the difference
for a, between a good engineer, a good
00:31:10.290 --> 00:31:12.390
researcher, and a not as good one.
00:31:12.390 --> 00:31:12.600
Right?
00:31:12.600 --> 00:31:14.190
So the, the good ones
are the ones that are
00:31:16.410 --> 00:31:21.300
updated with the last, um, trends,
the last techniques, and, and so on.
00:31:21.390 --> 00:31:26.490
That's, and that's always true,
um, in the low level, in the more.
00:31:27.545 --> 00:31:32.945
Technical level, everything is changing
because now, um, you are not priming.
00:31:32.945 --> 00:31:38.645
I mean, in some, in some areas,
people are not perming anymore.
00:31:39.515 --> 00:31:45.545
using, so this year especially has
been the boom of, uh, coding agents.
00:31:45.545 --> 00:31:45.665
Right.
00:31:45.665 --> 00:31:51.095
So now, before, like one year ago, well
two years ago, you were like coding
00:31:51.095 --> 00:31:56.040
classically, then Chatt PT appeared and
you were like, okay, asking CT PT to.
00:31:56.870 --> 00:32:00.890
Or any, or code or whatever to give you
some function to do whatever you wanted
00:32:00.890 --> 00:32:02.330
and you copy paste it into your code.
00:32:03.080 --> 00:32:07.460
you have this integrated into your, into
your coding interface, into your IDE.
00:32:08.180 --> 00:32:11.660
now you are, you are in, and the, the,
the assistant is looking at your code
00:32:12.110 --> 00:32:15.890
and it's creating the test, and it's
doing the interface and it's doing,
00:32:16.070 --> 00:32:22.940
so it's, it's kind of augmenting your
capabilities, so being able to play with
00:32:22.940 --> 00:32:25.370
it, to build with it, that's critical.
00:32:25.745 --> 00:32:27.965
And still today we see some engineers.
00:32:28.085 --> 00:32:32.675
I see some engineers that they are
not using these tools and they are way
00:32:32.675 --> 00:32:34.715
slower than the ones that are using them.
00:32:34.715 --> 00:32:38.585
So now we have a cap, a gap the ones
that are really using these tools and
00:32:38.585 --> 00:32:40.715
the ones that are still not using them.
00:32:41.075 --> 00:32:42.305
So that's super important.
00:32:42.802 --> 00:32:43.072
Yeah.
00:32:43.238 --> 00:32:43.798
Chris: Yeah, in.
00:32:43.888 --> 00:32:44.518
Interesting.
00:32:44.578 --> 00:32:46.138
Um, okay.
00:32:46.198 --> 00:32:51.628
And, um, you know, in terms of
advising someone how to work with AI
00:32:51.628 --> 00:32:56.908
right now, um, you know, in terms of
co-generation tools, for example, um,
00:33:00.133 --> 00:33:00.473
you know.
00:33:01.372 --> 00:33:01.792
Alberto: Hello,
00:33:01.978 --> 00:33:07.708
Chris: often, how soon should DB implement
them into their, uh, daily workflows?
00:33:07.708 --> 00:33:11.848
For example, um, would you advise
people to use 'em straight into
00:33:11.848 --> 00:33:16.523
university or, um, you know, when,
when's the best time to, to start?
00:33:16.762 --> 00:33:18.172
Alberto: To start using these systems?
00:33:19.307 --> 00:33:19.527
Yes.
00:33:19.553 --> 00:33:19.843
Chris: Yeah.
00:33:20.773 --> 00:33:22.543
Alberto: I dunno, I don't
have the answer for this.
00:33:22.543 --> 00:33:25.483
So the advantage of my generation,
let's say, is that we have
00:33:25.483 --> 00:33:28.033
already built without agents.
00:33:28.093 --> 00:33:31.543
So we have this practice and
now we know how to validate.
00:33:31.543 --> 00:33:35.293
When the agent gives some code or some
solution, you can say, okay, this is
00:33:35.293 --> 00:33:37.213
good, or this is not good, or change this.
00:33:37.453 --> 00:33:38.833
You play the role as an architect.
00:33:39.748 --> 00:33:42.088
Rather than the role
of a developer, right.
00:33:44.998 --> 00:33:46.438
For people who start now?
00:33:47.908 --> 00:33:48.178
I dunno.
00:33:48.208 --> 00:33:50.488
I think yeah, for sure.
00:33:50.488 --> 00:33:57.448
Like doing things themselves from
scratch, it's, it will always needed.
00:33:58.048 --> 00:34:03.598
Um, but I mean, you only have a finite
amount of energy and enough time,
00:34:03.838 --> 00:34:07.228
so you will have to sacrifice some
part of it and dedicate some time.
00:34:07.438 --> 00:34:14.608
Some of this time you will have to move it
to learn and to use these, these AI tools.
00:34:15.578 --> 00:34:19.658
Chris: Yeah, the, the rea the reason you
ask, I, I found a friend of mine, um, he
00:34:19.658 --> 00:34:26.588
recently asked me to hire him, someone,
um, who'd done their PhD pre-chat, GPT.
00:34:26.708 --> 00:34:33.008
Um, and I just wondered, um, is he quite
unique or is, is that potentially gonna be
00:34:33.008 --> 00:34:37.688
a thing with, you know, someone's problem
solving ability or, or coding ability,
00:34:38.198 --> 00:34:43.688
uh, where the people start asking for, uh,
candidates, you know, pre, pre, pre, um.
00:34:43.773 --> 00:34:47.253
It's pre, um, open ai,
so, we'll, we'll see.
00:34:47.731 --> 00:34:47.971
Alberto: yeah.
00:34:47.976 --> 00:34:48.176
Yeah.
00:34:48.873 --> 00:34:50.973
Chris: Um, cool.
00:34:50.973 --> 00:34:53.643
And, you know, as a career coach,
what, what advice would you
00:34:53.643 --> 00:34:58.563
give someone in general entering
into the age of AI right now?
00:34:58.593 --> 00:35:03.153
Because it, the, like you say, some
things have changed, uh, some things
00:35:03.153 --> 00:35:05.103
haven't, but you know, a lot has changed.
00:35:05.103 --> 00:35:07.833
What, what advice would you give
someone coming into the market?
00:35:07.833 --> 00:35:08.373
I guess I.
00:35:08.961 --> 00:35:09.711
Alberto: Build things.
00:35:10.161 --> 00:35:12.051
Get your hands 30.
00:35:12.801 --> 00:35:19.701
Uh, so now you have the opportunity
of building things from end to end.
00:35:19.941 --> 00:35:26.721
Like if you are a researcher or a, like an
engineer, you don't have any training on
00:35:26.721 --> 00:35:29.001
front end, for instance, or on database.
00:35:30.291 --> 00:35:31.851
now with the systems.
00:35:32.601 --> 00:35:36.141
can use them to build all the other
things that you don't know how to
00:35:36.141 --> 00:35:39.471
build or that would take you MOS learn.
00:35:40.671 --> 00:35:43.191
So that's empowering, right?
00:35:43.491 --> 00:35:48.951
So now you can create like apps, uh,
anything you want with the system.
00:35:48.951 --> 00:35:48.961
So.
00:35:49.776 --> 00:35:52.956
You can, you can focus on the
central part, that it's more
00:35:52.956 --> 00:35:55.326
interesting to you of your profiles.
00:35:55.326 --> 00:35:57.006
You are a researcher or ML engineer.
00:35:57.006 --> 00:35:59.706
You can focus on your models,
your data, we have whatever.
00:36:00.156 --> 00:36:02.886
And, and then you can have all
the frontend and everything
00:36:03.246 --> 00:36:05.106
delegated to the, to the agents.
00:36:05.106 --> 00:36:10.536
So, and the advantage of this, so it
changes everything because now, for
00:36:10.536 --> 00:36:12.006
instance, in my case, it gives me.
00:36:13.386 --> 00:36:17.106
I had never called it a
complex frontend before.
00:36:17.406 --> 00:36:22.956
I had done like websites or CSS
played with some basic stuff, but
00:36:22.956 --> 00:36:27.336
then now I, I am able to, yeah, like
to do complex databases, complex
00:36:27.366 --> 00:36:30.216
frontends with react, uh, apps.
00:36:30.936 --> 00:36:37.326
Um, a lot of API so now I can talk to,
to the engines in my company, uh, like
00:36:37.331 --> 00:36:41.556
at the, at the, if at the same level,
but really it grounded things, right?
00:36:41.556 --> 00:36:46.176
So now I understand much more what
they have in mind because I have
00:36:46.176 --> 00:36:48.666
also built them on my, on my side.
00:36:49.446 --> 00:36:52.656
And yeah, so it's gonna create,
I think it should create like
00:36:53.316 --> 00:36:58.056
engineers that are more transverse
but know that are maybe specialized
00:36:58.056 --> 00:36:59.616
in one thing, but it's still know.
00:37:00.561 --> 00:37:03.321
How to do the other
things more than before.
00:37:03.723 --> 00:37:04.838
Chris: Hmm, interesting.
00:37:04.898 --> 00:37:05.678
Um, okay.
00:37:06.038 --> 00:37:10.538
And do you, do you think potentially,
you know, with an engineer or researcher
00:37:10.538 --> 00:37:15.188
able to do more with, um, you know,
these tools out there that will lead
00:37:15.188 --> 00:37:17.708
to companies wanting bigger teams?
00:37:17.708 --> 00:37:22.298
Because now people can do more or
actually, um, um, you know, companies will
00:37:22.298 --> 00:37:26.048
want their researchers to do more, uh, do.
00:37:26.103 --> 00:37:29.523
Do more with less people, basically
with the tools that are out there,
00:37:29.523 --> 00:37:31.038
what, what are your thoughts on that?
00:37:31.681 --> 00:37:33.091
Alberto: I think it would
depend the market, right?
00:37:33.871 --> 00:37:37.921
me it's like, uh, it's
an economics problem.
00:37:38.011 --> 00:37:43.006
Um, if I think if your
market is saturated.
00:37:44.176 --> 00:37:47.206
Then you can do the same
thing with less people.
00:37:47.236 --> 00:37:51.496
Because if you don't expect to grow your
market, then the only thing you can do
00:37:51.496 --> 00:37:54.496
to grow your profit is to reduce cost.
00:37:55.276 --> 00:37:58.126
So that means productivity
with less people.
00:37:59.146 --> 00:38:02.836
If, there is a still market for you, then.
00:38:04.471 --> 00:38:08.371
Will be room for you to, yeah.
00:38:08.371 --> 00:38:10.351
To, to, to increase your team even.
00:38:10.411 --> 00:38:10.651
Right.
00:38:11.581 --> 00:38:13.666
um, yeah, so it will depend on the market.
00:38:14.735 --> 00:38:15.275
Chris: Interesting.
00:38:15.395 --> 00:38:16.025
Okay, cool.
00:38:16.085 --> 00:38:19.055
Well, um, yeah, thanks, thanks
for sharing that as well.
00:38:19.115 --> 00:38:22.295
Let's have, um, a little bit of
fun around future predictions.
00:38:22.295 --> 00:38:24.215
Just a couple of
questions we ask everyone.
00:38:24.215 --> 00:38:28.985
And you know, I think it'd be pretty cool
to revisit it in the next, uh, six to
00:38:28.985 --> 00:38:31.505
12 months and, and see where we're at.
00:38:31.505 --> 00:38:35.105
So, um, you know, what, what
trend in development, it doesn't
00:38:35.105 --> 00:38:36.575
have to be in e-commerce.
00:38:36.980 --> 00:38:42.680
Um, Orgen AI did mention Agen AI
to you before the show, but what,
00:38:42.740 --> 00:38:46.640
um, yeah, what development in AI
are you looking forward to the most
00:38:46.700 --> 00:38:49.130
over the next, um, coming months?
00:38:50.449 --> 00:38:54.139
Alberto: Well, one thing that I
think would, would happen is glasses.
00:38:54.889 --> 00:38:57.469
that's something I don't personally love.
00:38:58.129 --> 00:39:00.859
don't see people like wearing
glasses rather than meta glasses.
00:39:02.209 --> 00:39:02.509
But,
00:39:02.621 --> 00:39:03.586
Chris: Yeah, even.
00:39:04.336 --> 00:39:04.626
Yeah,
00:39:05.629 --> 00:39:10.339
Alberto: but I think we will start seeing
this, uh, sooner or later, and that
00:39:10.339 --> 00:39:14.959
will also affect the attack industry
also, because that's another space which
00:39:14.959 --> 00:39:16.999
you might think of, ads or whatever.
00:39:16.999 --> 00:39:18.679
So it's, it's another
00:39:19.001 --> 00:39:20.546
Chris: I think I'd throw them in the bin.
00:39:21.041 --> 00:39:23.711
If I got an ad in my glasses,
it'd definitely go in the bin.
00:39:24.529 --> 00:39:26.779
Alberto: Either you have these or
you pay more expensive glasses.
00:39:26.839 --> 00:39:28.489
So it's depending on, depending on
00:39:28.511 --> 00:39:28.871
Chris: Yeah.
00:39:28.871 --> 00:39:29.291
Fair?
00:39:29.471 --> 00:39:29.801
Yeah.
00:39:29.801 --> 00:39:29.811
Yeah.
00:39:30.386 --> 00:39:31.971
Yeah, no, that's very good point.
00:39:32.119 --> 00:39:36.229
Alberto: uh, but um, yeah, I
mean that, that's one thing.
00:39:36.229 --> 00:39:39.769
And you know, I mean, we, we all know
that met times already trying to, to push
00:39:39.769 --> 00:39:41.689
for these and other companies and Yeah.
00:39:42.379 --> 00:39:45.079
the other thing that I would like to see.
00:39:45.964 --> 00:39:52.234
It's about open source, open source
projects, and funded projects.
00:39:53.014 --> 00:40:01.444
think it is very important now that
the open source community, uh, is a
00:40:01.444 --> 00:40:10.174
strong as support, uh, from society
and governments and so on to avoid.
00:40:12.004 --> 00:40:15.454
Increasing the gap between the
big tech companies with the
00:40:15.484 --> 00:40:17.224
big private models, with big,
00:40:19.594 --> 00:40:23.074
quantities of private data and so
on, and the rest of the world, right?
00:40:23.434 --> 00:40:28.414
So AI should become a commodity,
like electricity, right?
00:40:28.564 --> 00:40:29.524
So AI should be.
00:40:30.679 --> 00:40:33.979
Available for everyone, for any
company to use the, to use it
00:40:33.979 --> 00:40:38.539
to increase their productivity,
to build services and and so on.
00:40:38.539 --> 00:40:42.439
So as much as possible you will have
then of course, like some, like the
00:40:42.649 --> 00:40:44.749
very state of the art models, whatever.
00:40:44.749 --> 00:40:47.414
So things that more
especially that will be.
00:40:48.139 --> 00:40:52.139
Not community will be really the key,
the core business for some companies.
00:40:52.159 --> 00:40:52.519
Sure.
00:40:52.849 --> 00:40:58.759
But there are things that should be
like computing, like having, having,
00:40:58.819 --> 00:41:05.924
uh, sort of a Chat GPT for everyone an
accessible price should be a comodity
00:41:05.944 --> 00:41:07.294
And I think it'll become a commodity.
00:41:07.654 --> 00:41:12.154
But I think and also like multilingual
that doesn't discriminate, that it's not
00:41:12.154 --> 00:41:16.864
like super English focused and also deals
with other minority languages and so on.
00:41:17.164 --> 00:41:20.229
So this is, we need also governments
to get involved here to guarantee that.
00:41:21.709 --> 00:41:26.389
We, I mean, the data is not
biased, is not, has any, you
00:41:26.389 --> 00:41:28.279
know, like it's ethical and so on.
00:41:28.279 --> 00:41:32.149
And it's, uh, and that it's, uh,
and on top of this we train models
00:41:32.149 --> 00:41:34.189
that are accessible for, for anyone.
00:41:34.779 --> 00:41:34.839
Yeah.
00:41:34.839 --> 00:41:35.104
No, that's one thing.
00:41:35.124 --> 00:41:39.249
And the, the other thing is, so
the other thing is small models.
00:41:39.369 --> 00:41:42.159
Uh, there is this other trend about, uh.
00:41:43.179 --> 00:41:45.729
Yeah, like creating, now we
are like every, every time like
00:41:45.759 --> 00:41:52.779
open AI or, or um, These, these
companies, they release a new model.
00:41:52.779 --> 00:41:55.299
It's like even larger than the one before.
00:41:55.299 --> 00:41:59.199
We know, we don't even know how large
these models are, the private ones.
00:42:00.309 --> 00:42:03.429
so, but there is also the other
trend of trying to do the same with
00:42:03.819 --> 00:42:07.989
smaller models that fit into your
mobile, into your computer and so on.
00:42:08.139 --> 00:42:08.559
And that.
00:42:09.909 --> 00:42:10.929
Use less energy.
00:42:10.989 --> 00:42:11.949
That's super important.
00:42:12.399 --> 00:42:15.279
'cause otherwise you are seeing
like some alman and these companies,
00:42:15.279 --> 00:42:19.419
they're trying to, they are investing
in nuclear power because they say
00:42:19.419 --> 00:42:24.699
we won't have enough energy to back,
uh, their, our, our AI systems.
00:42:25.599 --> 00:42:29.679
So we need like a smarter
and smaller AI systems.
00:42:30.159 --> 00:42:30.759
Um.
00:42:31.914 --> 00:42:34.044
And the other one, the, the third
one I would say is autonomy.
00:42:34.044 --> 00:42:39.204
Something similar to what I, talked before
about, uh, self adaptation and so on.
00:42:39.564 --> 00:42:44.184
But the other trend will be like
having agents that will be, now you
00:42:44.184 --> 00:42:48.054
are having agents that are able to do
some tasks that takes, you know, some
00:42:48.054 --> 00:42:52.674
hours or, I dunno, let's like kind
of, this level of complexity, right?
00:42:52.674 --> 00:42:53.964
Some kind of level of complexity.
00:42:54.294 --> 00:42:58.194
This level of complexity and
composition of tasks will increase.
00:42:59.199 --> 00:43:02.259
at some point you will see agents
that are like independent autonomous
00:43:02.259 --> 00:43:07.269
for some days, and then they go back
to you say, okay, that is the answer.
00:43:07.269 --> 00:43:10.239
That is my research the properly
formed for you and so on.
00:43:11.109 --> 00:43:11.919
um, yeah.
00:43:12.001 --> 00:43:12.451
Chris: Yeah.
00:43:13.231 --> 00:43:13.831
Nice.
00:43:13.981 --> 00:43:18.516
And out of interest, are you using
any agent tools in your personal life?
00:43:19.136 --> 00:43:20.276
Um, not in work.
00:43:21.204 --> 00:43:25.554
Alberto: In my personal life, I have
descriptions to every model to cloud.
00:43:25.614 --> 00:43:26.364
I'm continuously
00:43:26.461 --> 00:43:27.121
Chris: Okay.
00:43:27.121 --> 00:43:27.421
Fair?
00:43:27.624 --> 00:43:28.344
Alberto: I subscription to
00:43:28.471 --> 00:43:28.801
Chris: Yeah.
00:43:28.876 --> 00:43:29.166
Yeah,
00:43:29.244 --> 00:43:29.604
Alberto: cloud.
00:43:29.634 --> 00:43:30.024
I use
00:43:30.146 --> 00:43:30.486
Chris: of course.
00:43:30.744 --> 00:43:31.974
Alberto: code a lot for coding.
00:43:32.514 --> 00:43:36.799
I, I also, also use for some, uh, cursor.
00:43:38.184 --> 00:43:38.724
Um.
00:43:39.564 --> 00:43:41.784
Well, basically these, basically
the programming ones, the
00:43:41.784 --> 00:43:42.894
co-pilots for programming.
00:43:43.074 --> 00:43:45.444
That's the main, these are
the main ones that I'm using.
00:43:46.044 --> 00:43:47.754
And then I also, I try to,
00:43:47.911 --> 00:43:48.001
Chris: Fair.
00:43:48.144 --> 00:43:53.574
Alberto: to use cps uh, connect
to my Gmail, uh, through agents
00:43:53.574 --> 00:43:56.244
to send emails, uh, automatically.
00:43:56.784 --> 00:44:00.414
Um, share tickets in the company.
00:44:00.414 --> 00:44:01.404
That's also useful.
00:44:01.434 --> 00:44:01.614
Right?
00:44:01.644 --> 00:44:04.799
We can now start dealing with your geox.
00:44:06.264 --> 00:44:08.064
thanks to, thanks to agents.
00:44:08.994 --> 00:44:09.719
And Yes, but,
00:44:09.781 --> 00:44:10.381
Chris: Nice.
00:44:10.381 --> 00:44:10.951
Cool.
00:44:11.004 --> 00:44:13.584
Alberto: MCP actually, because the
thing is that we still don't have
00:44:13.584 --> 00:44:17.304
enough tools for our agents to exploit.
00:44:18.264 --> 00:44:23.724
once we will have tools to access to
your bank account, to your gm a to your
00:44:24.924 --> 00:44:28.254
cleaning service, whatever, so it will
be your agents, your agents will be,
00:44:28.314 --> 00:44:29.604
will become more and more interesting.
00:44:31.651 --> 00:44:33.601
Chris: Yeah, it is gonna be
really interesting to see
00:44:33.661 --> 00:44:35.131
how it evolves next year.
00:44:35.131 --> 00:44:41.401
I think, um, there's some, some
predictions for a GI in 2027 and we'll,
00:44:41.461 --> 00:44:44.221
we'll see if we actually get there
and what that actually looks like.
00:44:44.221 --> 00:44:47.971
But it's certainly gonna be,
um, an exciting 18 months.
00:44:48.451 --> 00:44:49.261
Um, okay.
00:44:49.261 --> 00:44:55.291
And in terms of, you know, your research
background, um, and recent papers, um.
00:44:55.981 --> 00:45:00.031
Is there anything that shaped your
thinking or any research direction
00:45:00.031 --> 00:45:04.171
that you found especially interesting
or underrated in your career?
00:45:04.852 --> 00:45:09.772
Alberto: I think there are, there
is, for instance, recent paper that
00:45:09.772 --> 00:45:14.962
appeared last week from one single
author, uh, that is called Less is
00:45:14.962 --> 00:45:20.842
More Recursive Reasoning Withstanding
Networks, and it's rethinking the way.
00:45:21.277 --> 00:45:25.867
We are creating the, but the
basis to, to agents, right?
00:45:26.137 --> 00:45:30.247
So the idea is as I, this, this
goes into the direction of doing
00:45:30.547 --> 00:45:32.677
smaller models to the same thing.
00:45:33.157 --> 00:45:37.867
So this is a paper that is, uh,
proposing like a new architecture.
00:45:39.082 --> 00:45:45.082
Um, do the same thing with
much less, uh, um, resources.
00:45:45.082 --> 00:45:45.322
Right.
00:45:45.742 --> 00:45:47.692
So that's, that's one interesting thing.
00:45:48.052 --> 00:45:51.202
And also the thing is, it's, it's
amazing that it's one single author
00:45:51.532 --> 00:45:52.972
and I think that will also change.
00:45:53.032 --> 00:45:56.872
So AI will the way we do research now.
00:45:57.757 --> 00:46:02.707
Like you have to, research means
reading papers, uh, understanding
00:46:02.707 --> 00:46:06.997
the state of the art, doing research,
uh, doing experiments and, and so on.
00:46:06.997 --> 00:46:07.867
Especially in experiments.
00:46:07.867 --> 00:46:08.737
These are heavy.
00:46:08.737 --> 00:46:10.177
These are like painful.
00:46:10.177 --> 00:46:15.457
Sometimes you
have to spend days finding
a stupid bag that you, don't find, right?
00:46:15.757 --> 00:46:20.617
So thanks to these kind of systems, I
hope will, will become, research will
00:46:20.617 --> 00:46:25.297
become more enjoyable, faster, and
researchers will be able to focus on.
00:46:25.702 --> 00:46:26.512
What matters, right?
00:46:26.512 --> 00:46:30.652
Really the idea, how do you mix this
with this, this idea, with this idea.
00:46:30.652 --> 00:46:31.912
Like more the innovation part.
00:46:34.402 --> 00:46:35.422
and yeah.
00:46:35.422 --> 00:46:35.782
What else?
00:46:35.812 --> 00:46:40.192
That is the other trend about which,
which is called, which is called Regen,
00:46:40.192 --> 00:46:42.562
generative information retrieval, and.
00:46:44.347 --> 00:46:49.747
This is something that instead of,
because now when you are looking for
00:46:49.747 --> 00:46:54.607
a product, what happens is that you,
you input your query in the system.
00:46:54.997 --> 00:46:57.337
Then the system creates a
mathematical representation of this
00:46:57.337 --> 00:46:58.957
query, what we call an embedding.
00:46:59.617 --> 00:47:03.847
then you have done the same with any
candidate product, and it finds then
00:47:04.057 --> 00:47:05.767
the closest products to your query.
00:47:06.157 --> 00:47:09.757
So you embed your, your, your query.
00:47:10.507 --> 00:47:12.037
And then you find the closest ones.
00:47:12.667 --> 00:47:15.667
So generative information with
through systems, they're based
00:47:15.667 --> 00:47:17.647
on LMS models and transformers.
00:47:18.037 --> 00:47:23.017
What they do is they generate not
the embedding of the product to then
00:47:23.077 --> 00:47:25.987
look for not the embedding of the
query to then look for the closest
00:47:25.987 --> 00:47:30.787
products, but directly they generate
the ID product you're looking for.
00:47:31.957 --> 00:47:35.437
And that's a paradigm,
shift if you want, because.
00:47:36.277 --> 00:47:39.337
Then if you, you would be
able to generate the, so it's,
00:47:39.337 --> 00:47:41.077
it's one step instead of two.
00:47:42.802 --> 00:47:44.332
The system gives you the final answer.
00:47:45.022 --> 00:47:49.792
Um, and then also you can say, okay,
now if I can generate the ID with
00:47:49.792 --> 00:47:53.602
a generative system, like as NLM, I
could also generate the explanation
00:47:53.602 --> 00:47:55.252
of why I recommended this product.
00:47:55.702 --> 00:47:58.852
And I could also use reasoning,
uh, techniques, right?
00:47:58.852 --> 00:48:03.382
To, to do, to increase my, the
quality of my recommendation given the
00:48:03.382 --> 00:48:05.842
context or that I know about the user.
00:48:06.322 --> 00:48:08.002
So, yeah, that's so generative.
00:48:08.092 --> 00:48:10.762
Uh, AI is something,
interesting to follow.
00:48:11.488 --> 00:48:11.908
Yeah, I would say,
00:48:12.215 --> 00:48:12.695
Chris: Interesting.
00:48:12.695 --> 00:48:16.385
I'll, um, yeah, if you can share the
link to
that paper, I'll, I'll put
00:48:16.385 --> 00:48:17.915
it in the show notes as well after.
00:48:18.185 --> 00:48:18.845
That'd be great.
00:48:18.935 --> 00:48:24.665
Um, and final question, just what, just
one book, any book, um, or products, um,
00:48:24.695 --> 00:48:26.945
that's changed the way you think about ai.
00:48:27.043 --> 00:48:27.463
Alberto: Mm-hmm.
00:48:27.535 --> 00:48:30.900
Chris: Or, or, or, um, you know,
really challenged the way you think.
00:48:31.738 --> 00:48:34.058
Alberto: there are, yeah,
maybe there are three books.
00:48:34.058 --> 00:48:35.479
Let's say two categories.
00:48:36.384 --> 00:48:41.754
of them is about, uh, climate crisis
and what's the connection between
00:48:41.754 --> 00:48:46.884
AI and data and the internet with,
uh, with climate that thing that's
00:48:46.884 --> 00:48:50.364
important that everyone in this, in,
in the industry is aware of, of this.
00:48:50.784 --> 00:48:55.104
It's a French journalist,
so the book is in French.
00:48:55.764 --> 00:48:59.004
don't think there is a translation
to other language for now,
00:48:59.004 --> 00:49:00.324
or maybe to Czech, I dunno.
00:49:00.804 --> 00:49:05.604
But the book, in any case called
the in French, it's the digital.
00:49:07.344 --> 00:49:09.324
Um, so it's about, yeah.
00:49:09.354 --> 00:49:18.084
How are um, computer centers,
data centers, um, like data, um,
00:49:18.834 --> 00:49:22.674
all the internet, I mean, all the
data that we are processing with,
00:49:22.674 --> 00:49:25.014
uh, in the, in the digital world.
00:49:25.164 --> 00:49:31.314
And now even more with, with generative
ai, uh, what's the effect on the, on the.
00:49:32.334 --> 00:49:33.564
On the climate, right?
00:49:33.984 --> 00:49:37.614
What do we need to be the systems,
what's behind A GPO, right?
00:49:37.764 --> 00:49:42.864
In terms of minerals, in terms
of the scarcity of, uh, of yeah.
00:49:42.864 --> 00:49:43.914
Of some minerals and so on.
00:49:43.914 --> 00:49:46.524
So what's, so that's very,
very interesting book.
00:49:47.694 --> 00:49:52.974
And the other one also is about,
um, what they call it, the, it's
00:49:52.974 --> 00:49:54.654
about the tive accumulation.
00:49:55.074 --> 00:49:59.694
So I think what's happening now, so
the books are Caliban and the witch.
00:50:01.254 --> 00:50:04.674
And the other one is the
ma, many headed hydra.
00:50:05.904 --> 00:50:12.324
So they are kind of two related
books from, from historians, um,
00:50:12.384 --> 00:50:20.334
that analyze how in the beginnings
of capitalism, the process of
00:50:21.354 --> 00:50:22.944
primitive accumulation has started.
00:50:22.944 --> 00:50:23.124
Right?
00:50:23.124 --> 00:50:27.174
So the thing is that you had
some like common land that.
00:50:28.359 --> 00:50:33.759
It was for everyone to like to use
to, to enjoy and so on to, to work on.
00:50:34.599 --> 00:50:37.959
And then at some point what
happened is that you had this land
00:50:38.409 --> 00:50:43.179
getting privatized and, uh, owned
by less and less people, right?
00:50:43.179 --> 00:50:45.909
So you have this
accommodation process and.
00:50:46.539 --> 00:50:50.259
In England actually, you have a very
famous movement of people trying to
00:50:50.319 --> 00:50:52.539
resist to this, which is the digger.
00:50:52.869 --> 00:50:56.349
And, and there are like movements
of people trying to out, getting
00:50:56.349 --> 00:50:58.809
back the lands to the, to
the, to the common use, right?
00:50:59.499 --> 00:51:01.029
I think the same is happening.
00:51:01.059 --> 00:51:04.449
So one, once this happens
is very hard to undo it.
00:51:05.409 --> 00:51:08.559
And um, now I think the
same is happening with data.
00:51:09.249 --> 00:51:13.989
open AI and all these systems, they have
been built in public data, sometimes
00:51:13.989 --> 00:51:15.789
in private data with no permission.
00:51:16.734 --> 00:51:19.284
Um, and that's too late now.
00:51:19.374 --> 00:51:23.064
So it's like once they are there,
everyone is using the systems
00:51:23.244 --> 00:51:24.984
and there is no way to undo it.
00:51:24.984 --> 00:51:25.194
Right.
00:51:25.914 --> 00:51:32.484
Um, so So this kind of data
accumulation, it's happening now these
00:51:32.484 --> 00:51:35.184
days and we have to be aware of that.
00:51:35.946 --> 00:51:37.626
Chris: Yeah, I'm, I'm going
to check the last two out.
00:51:37.631 --> 00:51:39.756
I lo I lo love a history
book to be honest.
00:51:39.756 --> 00:51:41.256
So, um, yeah.
00:51:41.256 --> 00:51:41.796
Thank you.
00:51:41.946 --> 00:51:45.426
Um, alright, well, uh, Alberta,
that concludes today's episode.
00:51:45.426 --> 00:51:48.906
Thank you so much for coming on
and, and sharing your knowledge
00:51:48.906 --> 00:51:51.396
and um, yeah, being so open.
00:51:51.396 --> 00:51:53.796
Uh, I really appreciate
and thanks for your time.
00:51:54.179 --> 00:51:55.499
Alberto: Thank you, Chris for having me.
00:51:55.679 --> 00:51:56.334
It's been a pleasure.
00:51:56.561 --> 00:51:57.101
Chris: Thank you.
00:51:57.259 --> 00:51:57.479
Alberto: Bye.