Why Calling Your AI "Intelligent" Is a Leadership Mistake with Patrick Rooney, Founder of Leonis Strategy

WEBVTT
NOTE
This file was generated by Descript
00:00:04.789 --> 00:00:05.189
INTRO
00:00:08.219 --> 00:00:10.619
Today on the podcast, I'm
joined by Patrick Rooney.
00:00:10.679 --> 00:00:15.809
He has a background in cognitive science
going back to about 20 plus years where
00:00:15.809 --> 00:00:20.279
he worked on neural networks, precursors
to large language models as a student
00:00:20.279 --> 00:00:25.499
around 2005, and has been doing applied
AI work in business for over a decade
00:00:25.499 --> 00:00:27.839
now, and now runs Leona's strategy.
00:00:28.349 --> 00:00:30.149
The most interesting part
of this conversation was.
00:00:30.309 --> 00:00:33.909
Patrick draws a hard line between
two completely different things
00:00:33.909 --> 00:00:35.799
that we collapse into one word.
00:00:35.979 --> 00:00:40.329
He separates scripted autonomy,
which is what an AI agent does when
00:00:40.329 --> 00:00:44.229
it runs and does work for you while
you get coffee from personhood.
00:00:44.439 --> 00:00:47.114
Autonomy, which implies rights and will.
00:00:47.794 --> 00:00:53.404
And interiority, most founders, including
I in the past making this mistake, use
00:00:53.464 --> 00:00:56.044
personhood language for a scripted tool.
00:00:56.044 --> 00:00:59.644
And Patrick makes the case that
this isn't just an imprecise
00:00:59.644 --> 00:01:01.924
or a mistake in vernacular.
00:01:02.284 --> 00:01:05.674
It's actually a leadership problem that
shapes how your whole team relates to
00:01:05.674 --> 00:01:07.144
technology, and it's very fascinating.
00:01:07.144 --> 00:01:08.524
It's an actually really interesting point.
00:01:09.219 --> 00:01:12.129
Patrick came into this work from
a philosophical and theological
00:01:12.129 --> 00:01:15.219
background layered on top of a
technical training and actually
00:01:15.219 --> 00:01:19.629
using the world in ai, in real world
client services and delivering value.
00:01:19.929 --> 00:01:22.629
This is a combination you don't
really run into often in the
00:01:22.629 --> 00:01:24.609
conversation about AI adoption,
00:01:25.239 --> 00:01:28.689
he's genuinely not interested
in either the utopian take
00:01:28.689 --> 00:01:30.339
for AI or the fear narrative.
00:01:30.714 --> 00:01:34.284
And he wants to be precise about
what these systems actually do and
00:01:34.284 --> 00:01:36.534
what it means for humans using them.
00:01:36.954 --> 00:01:41.334
We get into why LLMs are plausibility
engines and not truth seekers.
00:01:41.664 --> 00:01:45.984
What founders should not outsource to the
model and know that does not mean your ip.
00:01:45.984 --> 00:01:48.414
There's actually something more they
shouldn't outsource to the model.
00:01:48.744 --> 00:01:49.344
And why?
00:01:49.344 --> 00:01:52.974
Doubling down an inhuman person
connection is actually the correct
00:01:52.974 --> 00:01:55.914
strategic response to AI at scale.
00:01:56.244 --> 00:01:56.514
Right?
00:01:56.519 --> 00:01:57.324
Let's get into it.
00:01:59.538 --> 00:01:59.938
PODCAST
00:01:59.938 --> 00:02:05.398
Raul: So one of the things also is we take
whatever the output is as gospel or truth.
00:02:05.398 --> 00:02:10.588
Can you help me kind of help founders
and help our teams as we work with LLMs?
00:02:10.588 --> 00:02:15.008
Like how do we navigate that and
how are LLMs actually structured
00:02:15.308 --> 00:02:19.418
as not truth seeking, but also
just to continue the conversation.
00:02:19.718 --> 00:02:20.948
Help us unpack that a little bit.
00:02:21.635 --> 00:02:24.845
Patrick Rooney: Yeah, I think this
is an incredibly important point for
00:02:25.475 --> 00:02:26.645
thinking about the difference between.
00:02:27.045 --> 00:02:30.255
Intelligence per human
intelligence and LLMs generally.
00:02:30.255 --> 00:02:34.125
So let me just mention one
point on intelligence broadly.
00:02:34.525 --> 00:02:40.015
LLMs, they're focused on plausibility,
right on continuation of conversation
00:02:40.120 --> 00:02:41.935
in a, in a way that makes sense.
00:02:41.995 --> 00:02:47.465
And often we're really fortunate that
the byproduct of that is the truth.
00:02:47.465 --> 00:02:50.645
It's the right answer for us practically,
it's the right answer factually.
00:02:51.045 --> 00:02:54.435
As humans, though, we also have
the ability to focus on the
00:02:54.435 --> 00:02:55.785
truth just for its own sake.
00:02:55.875 --> 00:03:00.915
Not because it's practical, not because
it's gonna get me food or it's going to
00:03:00.915 --> 00:03:04.365
improve my status, but because I really
want to understand, I want to know.
00:03:04.665 --> 00:03:06.255
My kids are a great example of this.
00:03:06.255 --> 00:03:10.035
Like they'll just go off and look at
books on geography, books on dogs.
00:03:10.435 --> 00:03:12.325
They're not gonna get them
better grades in class.
00:03:12.325 --> 00:03:14.815
They're not going to impress
their friends necessarily.
00:03:14.815 --> 00:03:17.545
It's, they really want to
know about these subjects.
00:03:18.265 --> 00:03:22.195
have an interest in what's out
there, what's true and what's false.
00:03:22.945 --> 00:03:27.445
That's distinctively human and I
think it's a distinctive aspect of any
00:03:27.445 --> 00:03:32.785
intelligent being, any intelligent agent
that it can, not that it always does,
00:03:32.785 --> 00:03:37.385
but it can focus on the truth for its
own sake and find that satisfying LLMs
00:03:37.448 --> 00:03:38.378
Raul: own reward, right?
00:03:38.855 --> 00:03:40.265
Patrick Rooney: Exactly, exactly.
00:03:40.685 --> 00:03:41.195
Lms.
00:03:41.195 --> 00:03:44.765
Instead, they're focused on
executing the tasks we give them.
00:03:45.155 --> 00:03:45.395
Right.
00:03:45.395 --> 00:03:46.085
So it's like, okay.
00:03:46.745 --> 00:03:50.195
How do I get to the most
reliable, plausible answer based
00:03:50.195 --> 00:03:51.665
on the conversation thus far?
00:03:52.385 --> 00:03:53.825
And usually
00:03:53.888 --> 00:03:54.758
Raul: of the time that works.
00:03:54.818 --> 00:03:55.268
That's good.
00:03:55.268 --> 00:03:55.418
Yeah.
00:03:55.535 --> 00:03:58.265
Patrick Rooney: on the training data,
based on the engineering most of the
00:03:58.265 --> 00:04:02.565
time, like 99 point something percent,
it's gonna give you a good answer, but
00:04:02.713 --> 00:04:03.003
Raul: Yeah.
00:04:03.003 --> 00:04:03.203
But.
00:04:03.225 --> 00:04:04.430
Patrick Rooney: thing as
caring about the truth itself.
00:04:05.148 --> 00:04:07.788
Raul: I mean, I haven't had my
agent say, Hey, I'll be right back
00:04:07.788 --> 00:04:09.768
taking a picnic just to enjoy.
00:04:10.098 --> 00:04:11.268
'cause there's like, there's truth.
00:04:11.733 --> 00:04:13.683
I mean, truth for its own
sake, it's its own reward.
00:04:13.683 --> 00:04:18.633
Appreciating art or something beautiful or
a landscape is its, it's its own reward.
00:04:19.043 --> 00:04:22.193
Or preaching something good about like
a, a good aspect of something like when
00:04:22.193 --> 00:04:25.823
you appreciate, I mean, it's gonna sound
super contemplative, but like you look
00:04:25.823 --> 00:04:30.183
at either like I have in the backdrop
of my, one of my screens here is Big
00:04:30.183 --> 00:04:33.993
Sur and just like the coast, like you,
you look at that and you appreciate it.
00:04:34.348 --> 00:04:35.728
Because it's just like, it's beautiful.
00:04:35.728 --> 00:04:36.268
It's really cool.
00:04:36.268 --> 00:04:37.858
And like that's its own reward.
00:04:38.108 --> 00:04:39.728
Or even for the Artemis flight crew.
00:04:39.728 --> 00:04:44.888
Like I, one of my other screensavers
or desktop photos is Earthrise or the
00:04:44.888 --> 00:04:46.928
earth from the moon's perspective.
00:04:47.378 --> 00:04:51.098
So it's kind of, it's kind of fascinating
and yeah, my agents are not taking
00:04:51.098 --> 00:04:53.798
vacations and leisurely wasting time.
00:04:54.445 --> 00:04:54.655
Patrick Rooney: Yeah.
00:04:54.715 --> 00:04:57.025
The advantage of the agents
is that they're so focused
00:04:57.025 --> 00:04:57.865
on the practical, right?
00:04:57.865 --> 00:04:59.875
They, they do exactly
what you tell 'em to do.
00:04:59.875 --> 00:05:01.435
They, you know, they don't get tired.
00:05:01.435 --> 00:05:03.835
Maybe they run out of
context, you know, but.
00:05:04.555 --> 00:05:05.125
That's it.
00:05:05.215 --> 00:05:06.145
That's their advantage.
00:05:06.145 --> 00:05:07.255
They're just practical.
00:05:07.615 --> 00:05:10.735
Our advantage as humans is that we
care about things for themselves.
00:05:11.065 --> 00:05:15.155
We have this thing called leisure, and
that just means caring about th things
00:05:15.155 --> 00:05:17.075
that are good or beautiful or true.
00:05:17.475 --> 00:05:17.895
Raul: Mm.
00:05:18.135 --> 00:05:21.945
I think it's a powerful thing to also
keep in mind as leaders to still flex
00:05:21.945 --> 00:05:25.065
that muscle and understand based on
everything that comes outta you in the
00:05:25.065 --> 00:05:28.455
array and what you're looking at and the
different docs or decisions to be made.
00:05:28.825 --> 00:05:33.445
Still like what you mentioned, owning
the output, owning what do we take, what
00:05:33.445 --> 00:05:37.075
do we discard, what do we act on, how
do we implement it and why we do it.
00:05:37.405 --> 00:05:37.870
I think that's important.
00:05:38.131 --> 00:05:39.871
Patrick Rooney: Yeah, and,
and that's a deeply humanizing
00:05:39.871 --> 00:05:41.101
aspect of your business, right?
00:05:41.101 --> 00:05:41.806
If you do bring it back to.
00:05:42.481 --> 00:05:45.961
Here's something we do that's grounded
in truth, that's grounded in beauty.
00:05:45.961 --> 00:05:50.391
If you're in of creative or if it's
grounded in goodness, if you're really
00:05:50.391 --> 00:05:55.951
mission driven, like those are your unique
contribution as an owner, as a business,
00:05:56.351 --> 00:05:58.721
Raul: It's one of the key things as a
leader, that's what you're gonna focus on.
00:05:58.721 --> 00:06:02.321
Where are you leading people to
the team and your clients, and
00:06:02.321 --> 00:06:03.521
what are the values that you share?
00:06:03.521 --> 00:06:05.741
And those are the probably the most
important pieces of your culture.
00:06:06.141 --> 00:06:06.651
Patrick Rooney: definitely.
00:06:07.244 --> 00:06:10.814
Raul: Patrick, before we dive into like
intelligence, perceived intelligence,
00:06:11.084 --> 00:06:16.974
out of my personal curiosity, what
drove you really to nudge more and
00:06:16.974 --> 00:06:19.734
not just accept LLMs at surface value?
00:06:20.034 --> 00:06:21.654
Is it a personal curiosity?
00:06:21.654 --> 00:06:24.354
Is it more that you want to
understand why these things work?
00:06:24.624 --> 00:06:27.594
Or is there a deeper, like
a deeper concern with how
00:06:27.594 --> 00:06:28.674
we're using it as society?
00:06:28.674 --> 00:06:32.794
What really drove you to understand
deeper the reasons behind intelligence,
00:06:32.794 --> 00:06:34.174
both human and artificial?
00:06:34.574 --> 00:06:35.624
Patrick Rooney: Yeah, great question.
00:06:35.624 --> 00:06:38.744
It's, it's been a long standing
interest of mine, so I can go back
00:06:38.744 --> 00:06:42.944
to, say my college days when I was
studying intelligence as a subject,
00:06:43.214 --> 00:06:44.864
cognitive science was my major.
00:06:45.464 --> 00:06:48.734
And that really attacks intelligence
from different angles, computer science,
00:06:48.794 --> 00:06:50.834
neuroscience, philosophy, et cetera.
00:06:51.494 --> 00:06:53.329
So even back then, like 20 years ago.
00:06:53.739 --> 00:06:56.469
Yeah, there was this notion,
Hey, we might be able to create
00:06:56.469 --> 00:07:00.879
something that's simulating or,
or rivaling human intelligence.
00:07:01.239 --> 00:07:05.449
Obviously it, it took a long time for that
to emerge, but already a long time ago.
00:07:05.449 --> 00:07:09.619
I, I found that fascinating and wanted
to understand is that perspective, right?
00:07:09.649 --> 00:07:10.909
Like, I mean that Right.
00:07:10.909 --> 00:07:15.324
I think for all of us, it's a little bit
unnerving at times to think Yeah, they're,
00:07:15.324 --> 00:07:19.524
they're thinking machines or thinking
software like right in front of me.
00:07:20.304 --> 00:07:23.004
It, you know, it threatens our
sense of uniqueness, our sense of
00:07:23.134 --> 00:07:24.754
potentially ownership or responsibility.
00:07:24.754 --> 00:07:29.054
So there are a lot of dimensions that I
think, engage all of us, and I've just
00:07:29.054 --> 00:07:32.144
been aware of them perhaps a little
bit longer than the average person
00:07:32.144 --> 00:07:37.274
because of my background both in work
doing AI the last several years and
00:07:37.964 --> 00:07:38.699
going all the way back to college.
00:07:39.261 --> 00:07:40.521
Raul: Yeah, IEX in the background.
00:07:40.521 --> 00:07:40.791
Yeah.
00:07:41.241 --> 00:07:44.111
Was there a moment where you
realized was it just like gradual
00:07:44.111 --> 00:07:47.201
over time or was there one moment in
time where your stance is a little
00:07:47.201 --> 00:07:48.791
bit, I wouldn't say controversial.
00:07:48.791 --> 00:07:51.221
It could be controversial depending on
who you talked to in Silicon Valley,
00:07:51.221 --> 00:07:53.981
but that LLMs aren't intelligent.
00:07:54.071 --> 00:07:57.921
Was it something that you just
realized from it or through like
00:07:57.921 --> 00:08:00.501
a conviction that you had over
time, or was it one moment in time?
00:08:00.901 --> 00:08:03.541
Patrick Rooney: That it's, it's been
a longstanding conviction, right?
00:08:03.541 --> 00:08:05.071
So as I just kind of suggested.
00:08:05.716 --> 00:08:06.976
It's a, it's a, an issue.
00:08:06.976 --> 00:08:10.706
I kind of had to start thinking through
years ago because I worked a little bit
00:08:10.736 --> 00:08:16.306
on stuff as a student in classwork, so
I'm not, not going to overstate it, but
00:08:16.396 --> 00:08:18.646
worked on what were precursors of LM.
00:08:18.676 --> 00:08:20.506
So like neural networks, right?
00:08:20.596 --> 00:08:22.696
The, that's some of the
architecture underlying it.
00:08:23.416 --> 00:08:28.186
And even back in, let's say 2005
ish, there were these claims that
00:08:28.246 --> 00:08:29.656
you can make a neural network.
00:08:30.091 --> 00:08:34.261
Decide to do something, it does X or Y
or it discriminates like it, it chooses
00:08:34.881 --> 00:08:37.671
between this stimulus and that stimulus.
00:08:37.671 --> 00:08:43.011
So a lot of the language used
decades ago was already very
00:08:43.041 --> 00:08:45.111
anthropomorphic, very human centered.
00:08:45.621 --> 00:08:48.951
And like, you know, back then you
could see there's a huge difference
00:08:49.101 --> 00:08:51.801
between what those systems were
doing and what we do as humans.
00:08:52.051 --> 00:08:53.851
So it was easier to push back.
00:08:53.851 --> 00:08:56.311
But over the years I've had
to kind of refine my thinking.
00:08:56.746 --> 00:08:59.831
As that gap seems to
get smaller and smaller.
00:09:00.231 --> 00:09:00.561
Raul: Hmm.
00:09:00.921 --> 00:09:05.511
One of the things, so the interesting
part that alluded us for, for recording
00:09:05.511 --> 00:09:08.781
this pod is that like your, your
thesis and statement, and I'm, I'm
00:09:08.781 --> 00:09:11.211
not disagreeing, but I just wanna
kind of work with this with you.
00:09:11.511 --> 00:09:14.481
If what we're using
isn't truly intelligent,
00:09:14.989 --> 00:09:15.409
Patrick Rooney: Mm-hmm.
00:09:15.471 --> 00:09:19.131
Raul: it's trained based on a
similar methodology that humans
00:09:19.131 --> 00:09:20.241
are trained, for example, like,
00:09:20.354 --> 00:09:20.594
Patrick Rooney: Yep.
00:09:21.381 --> 00:09:23.421
Raul: Through culture,
education, upbringing.
00:09:23.741 --> 00:09:26.261
Obviously there's a difference between
the 40 years though, like whatever,
00:09:26.261 --> 00:09:28.121
how old people are listening to this.
00:09:28.564 --> 00:09:28.864
Patrick Rooney: Yep.
00:09:29.027 --> 00:09:33.341
Raul: 60, 20, 30 years of human
training versus LLM training.
00:09:33.521 --> 00:09:36.011
But there's also the reality
that I'm also training and
00:09:36.011 --> 00:09:36.911
they're not doing it perfectly.
00:09:36.911 --> 00:09:39.831
Obviously, some of the outputs but
I'm also training like a fleet of
00:09:39.831 --> 00:09:43.741
agents and I have agents doing certain
things for me on a habitual basis.
00:09:43.741 --> 00:09:46.291
And some of the things are pretty
impressive and it's kind of.
00:09:46.996 --> 00:09:49.216
Mirrors what intelligence would look like.
00:09:49.464 --> 00:09:49.824
Patrick Rooney: Yep.
00:09:49.966 --> 00:09:52.726
Raul: I think there's even some
blurs of like, what is human versus
00:09:52.726 --> 00:09:55.466
what is what an LLM produced.
00:09:55.856 --> 00:09:56.126
What
00:09:56.194 --> 00:09:56.494
Patrick Rooney: Yep.
00:09:56.576 --> 00:09:59.856
Raul: of the, i, I don't know if
to use your words, but explanation
00:09:59.856 --> 00:10:03.276
for intelligent looking outputs
and implications of intelligence.
00:10:03.276 --> 00:10:06.276
Like how do we derive that
and have a hierarchical order?
00:10:06.276 --> 00:10:09.096
And then we'll bring it to practical
terms of like, why this question
00:10:09.366 --> 00:10:11.166
and this framework actually matters.
00:10:11.774 --> 00:10:12.104
Patrick Rooney: Yeah.
00:10:12.504 --> 00:10:12.744
Right.
00:10:12.744 --> 00:10:15.054
So what you're alluding to,
I'd say what are different?
00:10:15.134 --> 00:10:18.014
You could say variations
on the Turing test, right?
00:10:18.014 --> 00:10:19.814
That things, the Turing test.
00:10:20.234 --> 00:10:21.104
Probably a lot of folks.
00:10:21.869 --> 00:10:23.339
Listening or familiar with it, right?
00:10:23.339 --> 00:10:29.649
It was, it goes back to early 20th,
mid 20th century Alan Turing, we've
00:10:29.649 --> 00:10:34.579
got LL at that time just computers,
in theory, not yet in practice,
00:10:34.579 --> 00:10:37.459
potentially replicating human behaviors.
00:10:37.459 --> 00:10:40.219
Now you're, what you're describing
is we're actually seeing that.
00:10:40.219 --> 00:10:40.459
Right?
00:10:41.269 --> 00:10:45.169
So if it looks like what
we're doing, why isn't it?
00:10:45.634 --> 00:10:46.564
In fact intelligent.
00:10:46.624 --> 00:10:50.794
Like if it, if it walks like a duck,
duck and quacks like a duck, why
00:10:50.794 --> 00:10:52.834
isn't it a truly intelligent duck?
00:10:53.234 --> 00:10:56.604
And what you started talking about
there was like, training I think
00:10:56.604 --> 00:10:58.704
is a great angle to break in on.
00:10:58.704 --> 00:11:05.334
So if you look at the training of, of
an LLM, it's going to be distinctive,
00:11:05.334 --> 00:11:08.424
say from the training, the education,
the formation of a human being in
00:11:08.424 --> 00:11:11.334
that it is really heavily centered on.
00:11:11.919 --> 00:11:18.579
Text inputs, those are transformed into
numerical intermediaries, embeddings, and
00:11:18.579 --> 00:11:23.529
then ultimately you get, you'd get text
outputs from the system as it's tested
00:11:23.529 --> 00:11:25.179
and eventually after it's been trained.
00:11:25.809 --> 00:11:30.949
So the slice of the world that
it's ingesting and working with is
00:11:30.949 --> 00:11:32.659
actually, it's two things about it.
00:11:32.689 --> 00:11:34.099
Number one, it's really limited, right?
00:11:34.099 --> 00:11:37.219
So like, imagine that you had nothing
in your experience except books,
00:11:37.219 --> 00:11:38.689
like you didn't interact with.
00:11:39.184 --> 00:11:40.444
Actual humans, right?
00:11:40.444 --> 00:11:45.034
It didn't involve being a baby
and having all the, you know, the
00:11:45.154 --> 00:11:50.794
physical, visual, tactile stimuli,
but it, it's just focused on language.
00:11:51.364 --> 00:11:56.224
But your world would be totally
different and impoverished, right?
00:11:56.296 --> 00:11:57.676
Raul: on, based on the inputs.
00:11:57.681 --> 00:11:57.976
It, it
00:11:58.024 --> 00:11:58.294
Patrick Rooney: Yes.
00:11:58.306 --> 00:11:58.936
Raul: limited there.
00:11:58.936 --> 00:12:02.476
But, but I think the argument is that
some of the outputs in the, the real world
00:12:02.476 --> 00:12:06.496
value, or at least the economic value
that is creating, is somewhat comparable
00:12:06.496 --> 00:12:10.336
to like entry level workers or even for
junior assistants or even in some cases,
00:12:10.566 --> 00:12:10.896
Patrick Rooney: yeah.
00:12:10.903 --> 00:12:14.233
Raul: wouldn't say pure PhD, but some,
like, I've, like I've heard some stories
00:12:14.233 --> 00:12:15.583
of auto research going overnight.
00:12:15.583 --> 00:12:19.123
It what would take PhD seven
years, it would do in a week
00:12:19.123 --> 00:12:21.583
or in 36 hours, 48 hours.
00:12:21.883 --> 00:12:27.523
So I think it's understanding this is
important, but also the outputs are pretty
00:12:28.203 --> 00:12:30.693
I wouldn't say jaw dropping, but sometimes
it's like, it's pretty impressive.
00:12:30.693 --> 00:12:30.843
What's
00:12:30.866 --> 00:12:31.106
Patrick Rooney: Right.
00:12:31.526 --> 00:12:33.476
They're, they're massively impressive.
00:12:33.476 --> 00:12:35.126
Like, I, I don't want to understate that.
00:12:35.126 --> 00:12:35.366
Right.
00:12:35.426 --> 00:12:39.086
In addition to what you just mentioned,
we have mathematical proofs for
00:12:39.206 --> 00:12:44.036
unsolved problems that LMS are
either completing independently or
00:12:44.036 --> 00:12:45.746
at least assisting researchers with.
00:12:46.436 --> 00:12:48.716
So that's, you know, cutting
edge of intelligence.
00:12:48.806 --> 00:12:49.106
Right.
00:12:49.356 --> 00:12:51.901
So what, what's going on
that's different though?
00:12:51.901 --> 00:12:54.341
I think that there's, the one
thing I mentioned, kind of
00:12:54.341 --> 00:12:56.411
the, the text or often code.
00:12:56.771 --> 00:13:02.691
Heavy inputs very different from the
kind of experience that, that we have.
00:13:03.231 --> 00:13:07.191
And importantly that that doesn't
marry up with, it doesn't connect
00:13:07.221 --> 00:13:09.171
directly to things in the world.
00:13:09.201 --> 00:13:11.871
So I think that's kind of the
fundamental underlying point there.
00:13:12.621 --> 00:13:16.411
It's language, but it's
language dissociated from
00:13:16.811 --> 00:13:18.431
the world that we experience.
00:13:18.831 --> 00:13:22.461
What it, but it does a fantastic
job, right, of taking that language,
00:13:22.551 --> 00:13:25.911
transforming it into other language
or code as the case may be.
00:13:26.311 --> 00:13:27.331
It's reliant though.
00:13:27.571 --> 00:13:28.741
This goes, so this is the second point.
00:13:28.771 --> 00:13:34.751
It's reliant on basically millions, tens
of millions, maybe billions of people's
00:13:34.781 --> 00:13:37.031
input via the training data, right?
00:13:37.031 --> 00:13:42.186
The, the internet essentially for the
foundation models it's reliant on.
00:13:42.586 --> 00:13:44.476
Trainers who are shaping it.
00:13:44.666 --> 00:13:49.036
It could be reinforcement learning via
human feedback the engineers that are
00:13:49.036 --> 00:13:53.326
building the system, and ultimately
everyday users like you and me who are
00:13:53.326 --> 00:13:59.626
going to the system and directing it to do
certain things and correcting it if it's
00:13:59.836 --> 00:14:01.906
not acting in the way that we want it to.
00:14:02.236 --> 00:14:03.886
So we've got a whole network of people.
00:14:04.556 --> 00:14:07.106
And their experience that's
feeding into these systems.
00:14:07.136 --> 00:14:10.686
And somebody who's an AI researcher
at the forefront who has made a
00:14:10.686 --> 00:14:13.606
very similar observation is signing.
00:14:13.606 --> 00:14:16.876
She, I think I, I'm roughly
in the ballpark of his name
00:14:16.996 --> 00:14:18.406
with his pronunciation there.
00:14:19.126 --> 00:14:22.476
And you know, he's working
with Jan Koon, formerly from
00:14:22.476 --> 00:14:25.806
Meta on some novel AI research.
00:14:26.056 --> 00:14:29.866
He's made like these very points
in his discussions of whether
00:14:29.866 --> 00:14:31.576
AI is truly intelligent or not.
00:14:31.576 --> 00:14:34.936
So what I am offering is
not a unique perspective.
00:14:34.936 --> 00:14:37.606
It's not really unorthodox either.
00:14:37.606 --> 00:14:40.426
There, there are plenty of people out
there who are steeped in the field
00:14:40.936 --> 00:14:44.446
that would say, yeah, there are,
there are true distinctions between
00:14:44.646 --> 00:14:49.081
intelligence as we've understood and
defined it and what an AI is doing.
00:14:49.351 --> 00:14:52.861
And that's not to in any way downplay the
incredible achievements of the systems.
00:14:53.261 --> 00:14:54.701
Raul: Just to see if for what it is.
00:14:54.701 --> 00:14:57.431
Because also the other things too
that are implications of intelligence
00:14:57.431 --> 00:15:01.221
to what you also shared with me was
that is there a will or autonomy?
00:15:01.221 --> 00:15:02.001
Is there personhood?
00:15:02.001 --> 00:15:02.751
Is there rights?
00:15:03.104 --> 00:15:03.404
Patrick Rooney: Right.
00:15:03.501 --> 00:15:06.561
Raul: an AI have the right to
control a human vice versa?
00:15:06.591 --> 00:15:07.311
Eventually.
00:15:07.551 --> 00:15:09.621
And I think this also has
other implications for
00:15:09.621 --> 00:15:11.061
legal, ethical, and economic.
00:15:11.311 --> 00:15:14.371
Right now, obviously we're leveraging
it as a, as a, as a thought
00:15:14.371 --> 00:15:17.401
partner, as a tool to be able to
grow and even some business models.
00:15:17.581 --> 00:15:19.441
AI is the business model
and that's actually.
00:15:19.441 --> 00:15:21.811
Of the key things that I'm seeing
and working with clients with to help
00:15:21.811 --> 00:15:23.551
them, you know, evolve into the future.
00:15:23.801 --> 00:15:27.771
But I think having this discussion to
understand fully and also understand the
00:15:27.771 --> 00:15:32.011
limits and implications so that we know
what to do and also what not to do to
00:15:32.011 --> 00:15:33.721
bring it more practical and down to earth.
00:15:33.721 --> 00:15:35.731
Like based on what you're saying,
what are some of the things that
00:15:35.731 --> 00:15:38.611
founders should not be doing with ai?
00:15:39.301 --> 00:15:42.601
We, we may or may not like
this is probably a litany list
00:15:42.644 --> 00:15:42.864
Patrick Rooney: Yes.
00:15:43.141 --> 00:15:47.851
Raul: based on this thesis and this
framework of AI looks intelligent,
00:15:48.571 --> 00:15:49.741
it's not really intelligent.
00:15:49.741 --> 00:15:50.851
It's a tool we crafted.
00:15:51.301 --> 00:15:54.541
However, it per it, it works
and it has significant outputs
00:15:54.541 --> 00:15:56.431
that generate economic value.
00:15:56.741 --> 00:15:59.471
It helps, I mean, some of my clients
went in some of the biggest deals with
00:15:59.471 --> 00:16:01.511
some of the work that we did co with ai.
00:16:01.511 --> 00:16:04.961
So it's, a real thing
and a real matter to it.
00:16:04.961 --> 00:16:06.221
Not to get too caught up in that, but
00:16:06.364 --> 00:16:06.634
Patrick Rooney: Yeah.
00:16:07.034 --> 00:16:09.434
Raul: Push away the technology
as if it's just another
00:16:09.434 --> 00:16:11.324
Photoshop or just another iPhone.
00:16:11.324 --> 00:16:13.334
Like it's not, it's
more powerful than that.
00:16:13.754 --> 00:16:16.844
But what are some of the things
that we should be cognizant of not
00:16:16.844 --> 00:16:19.994
doing as founders leveraging this,
00:16:20.234 --> 00:16:20.534
Patrick Rooney: yeah.
00:16:20.894 --> 00:16:23.114
So there, there are a few things
you started mentioning there
00:16:23.114 --> 00:16:26.004
that I think if AI were really
intelligent, it would fall out.
00:16:26.004 --> 00:16:30.354
That it's, it has these characteristics,
so things like, yeah, it's
00:16:30.354 --> 00:16:31.824
got will or autonomy, right?
00:16:31.824 --> 00:16:34.224
It, it's, it's its own person.
00:16:34.434 --> 00:16:37.794
So there are rights and
responsibilities associated with that.
00:16:38.214 --> 00:16:41.874
That is we, there's a certain
way we should be treating LLMs
00:16:42.264 --> 00:16:45.224
like human beings and yeah.
00:16:45.284 --> 00:16:46.124
And then potentially.
00:16:46.131 --> 00:16:48.831
Raul: classes, by the way, by,
there are some classes of thought.
00:16:49.101 --> 00:16:50.121
'cause I started breaking down.
00:16:50.121 --> 00:16:53.301
Like I think there are
different metaphysics of ai.
00:16:53.301 --> 00:16:55.761
Some people do believe that it
has rights and do believe that
00:16:55.769 --> 00:16:56.189
Patrick Rooney: Mm-hmm.
00:16:56.481 --> 00:16:58.401
Raul: a personhood and
is treated a certain way.
00:16:58.979 --> 00:16:59.269
Patrick Rooney: Yeah.
00:16:59.511 --> 00:17:03.051
Raul: so there, there are multiple
thoughts around this perspective
00:17:03.051 --> 00:17:06.261
and this, this could be a con,
like I, I alluded in the beginning.
00:17:06.261 --> 00:17:11.541
This podcast itself could be like a
controversial statement said in Silicon
00:17:11.541 --> 00:17:12.651
Valley to some people, depending
00:17:12.734 --> 00:17:13.034
Patrick Rooney: Yeah,
00:17:13.041 --> 00:17:13.051
Raul: are.
00:17:13.604 --> 00:17:15.529
Patrick Rooney: so that's a
great point you make there.
00:17:16.019 --> 00:17:19.324
All the things that we are discussing
right now, these are like summarized
00:17:19.324 --> 00:17:22.924
versions of what, what are actually
very profound debates and, and
00:17:22.924 --> 00:17:25.444
discussions about personhood about.
00:17:25.819 --> 00:17:27.859
What is the nature of
thinking and intelligence?
00:17:28.489 --> 00:17:32.149
What is the nature of an artifact
versus a natural thing, right?
00:17:32.149 --> 00:17:35.429
And, and some of these go back
centuries, if not millennia.
00:17:36.149 --> 00:17:39.239
So, so we're not new to these topics,
and you could, we could spend hours.
00:17:39.779 --> 00:17:43.029
I what I just want to, I would want to
point out what I would want people to
00:17:43.029 --> 00:17:47.379
take away is if you hear a perspective
like that, that this is a person,
00:17:47.409 --> 00:17:51.789
this, this LLM or this instance,
this chat bot is a person, is that.
00:17:52.209 --> 00:17:58.059
That comes with a bunch of assumptions
that a, a whole load of baggage behind it.
00:17:58.619 --> 00:18:02.009
So it's not something that
should be said glibly or naively.
00:18:02.099 --> 00:18:05.099
And I think in fact, there, there are
good reasons to argue the opposite,
00:18:05.879 --> 00:18:11.259
but I, I wouldn't expect in like that
20, 30 minutes to be able to convince
00:18:11.259 --> 00:18:16.309
somebody that there's no reason to
at all to believe that there's some
00:18:16.309 --> 00:18:17.929
personality or personhood behind these.
00:18:18.329 --> 00:18:18.869
Raul: That's fair.
00:18:19.619 --> 00:18:20.219
the things that we shouldn't.
00:18:20.219 --> 00:18:23.939
So one of the things, maybe we shouldn't
say that it's, this is my, I mean, we
00:18:23.939 --> 00:18:25.169
are using that different right now.
00:18:25.259 --> 00:18:26.999
This is my AI coworker or my
00:18:27.067 --> 00:18:27.487
Patrick Rooney: Yes.
00:18:28.079 --> 00:18:28.739
Raul: or we
00:18:28.897 --> 00:18:29.317
Patrick Rooney: Right.
00:18:29.399 --> 00:18:30.239
Raul: like who is we?
00:18:30.439 --> 00:18:32.719
So what are other, other
things to be cognizant aware,
00:18:33.057 --> 00:18:33.387
Patrick Rooney: Yeah.
00:18:33.919 --> 00:18:38.069
Raul: for, in practical terms, like in
the field building agents with ai, with
00:18:38.162 --> 00:18:38.942
Patrick Rooney: So, so, yeah.
00:18:38.999 --> 00:18:41.219
Raul: some sort of autonomy,
but they're scripted autonomy,
00:18:41.647 --> 00:18:42.247
Patrick Rooney: Yes.
00:18:42.397 --> 00:18:42.667
Yeah.
00:18:42.667 --> 00:18:42.967
Right.
00:18:42.967 --> 00:18:44.287
That term autonomy, right.
00:18:44.842 --> 00:18:47.722
Has different layers and
different meanings to it.
00:18:48.142 --> 00:18:51.262
So if you say it works autonomously,
you mean like, Hey, I can walk away and
00:18:51.262 --> 00:18:52.912
it'll do something while I get coffee.
00:18:53.332 --> 00:18:53.962
That's fine.
00:18:53.962 --> 00:19:00.052
But it autonomy has this kind of other
connotation that it, it's an independent
00:19:00.442 --> 00:19:04.852
person with rights and that's what I
would, I would've people shy away from
00:19:05.422 --> 00:19:07.042
those really anthropomorphic terms.
00:19:07.732 --> 00:19:10.622
Yeah, like the really anthropomorphic
terms, you know, if you use it.
00:19:10.942 --> 00:19:14.542
Ca you know, casually, Hey, it's
an AI assistant, not a big deal,
00:19:14.542 --> 00:19:17.242
but deeply personifying your ai.
00:19:17.492 --> 00:19:22.382
We've seen instances where that actually
has, that has led to mental health issues.
00:19:22.382 --> 00:19:22.922
And that's a,
00:19:23.172 --> 00:19:23.202
Raul: hm.
00:19:23.600 --> 00:19:24.740
Patrick Rooney: something
of a tangent, right.
00:19:24.740 --> 00:19:26.660
But that's, that's it's
made national news.
00:19:26.660 --> 00:19:30.650
People who have become fixated on
that founders, I would think are less
00:19:30.650 --> 00:19:32.880
likely to to develop that particular,
00:19:32.902 --> 00:19:33.832
Raul: as leadership,
00:19:34.260 --> 00:19:34.560
Patrick Rooney: yeah.
00:19:34.732 --> 00:19:34.852
Raul: with
00:19:34.980 --> 00:19:35.160
Patrick Rooney: Alright.
00:19:35.212 --> 00:19:35.362
Raul: too.
00:19:35.362 --> 00:19:35.872
Like they, they,
00:19:35.895 --> 00:19:36.105
Patrick Rooney: Yeah.
00:19:36.105 --> 00:19:39.135
But to, to model it, I think for,
for their teams in particular
00:19:39.135 --> 00:19:40.845
is it's just healthy behavior.
00:19:41.595 --> 00:19:44.835
But a really positive thing, I
would say is to continue to take
00:19:44.925 --> 00:19:49.975
ownership of what's going into the
ai, how it's working and the outputs.
00:19:49.975 --> 00:19:50.185
Right.
00:19:50.185 --> 00:19:52.945
So not so much the Yeah.
00:19:53.395 --> 00:19:57.955
Downplaying what the AI AI can do,
but instead emphasizing your role
00:19:58.495 --> 00:20:00.865
as a human and as an owner of.
00:20:01.375 --> 00:20:06.445
This tool and this process, like,
hey, I need to ensure that the inputs
00:20:06.445 --> 00:20:09.265
I'm putting in here for my clients,
that they're, they're clean, that
00:20:09.265 --> 00:20:15.125
they're accurate that I've handled
them responsibly, that the tool handles
00:20:15.125 --> 00:20:19.775
them responsibly, data securities,
forced that whatever comes out also
00:20:19.775 --> 00:20:21.310
has gone through appropriate scrutiny.
00:20:21.485 --> 00:20:25.205
Like it, it's a huge time saver,
but that doesn't give us you know,
00:20:25.865 --> 00:20:30.265
the re any reason or any, it doesn't
clear us with the responsibility to
00:20:30.595 --> 00:20:32.005
take ownership of what's going on.
00:20:32.405 --> 00:20:34.915
Raul: Yeah, the term that comes to mind,
this is easier 'cause it outsourced
00:20:34.915 --> 00:20:39.005
thinking sometimes you're doing, I mean,
on a good day I'm personally able to do
00:20:39.005 --> 00:20:41.780
maybe six to 13 things in a day like.
00:20:42.043 --> 00:20:42.133
Patrick Rooney: Mm-hmm.
00:20:42.875 --> 00:20:46.415
Raul: Big things, like things that
matter with my agents and like a
00:20:46.415 --> 00:20:49.175
good day is like 70, maybe 80 things.
00:20:49.175 --> 00:20:49.985
Like actual things.
00:20:49.985 --> 00:20:51.245
Not just decisions made.
00:20:51.575 --> 00:20:52.445
Not just ideas
00:20:52.608 --> 00:20:52.898
Patrick Rooney: Yeah.
00:20:53.615 --> 00:20:57.515
Raul: So with that volume, obviously,
I mean, what's that derivative?
00:20:57.515 --> 00:21:00.515
Six to seven, maybe like
a 10 to five X volume.
00:21:00.915 --> 00:21:03.465
There are some things like
the time saving, sometimes the
00:21:03.465 --> 00:21:06.615
output you do accept as fact, but
you'll have to think critically.
00:21:06.615 --> 00:21:10.635
And this, I'm literally using more brain
power at the end of the day now just to.
00:21:10.905 --> 00:21:15.195
Cognitively work my way through
and identify this or that,
00:21:15.195 --> 00:21:16.365
or weigh the information.
00:21:16.365 --> 00:21:19.095
And I think that's one of the key
fallacies that you just mentioned, is
00:21:19.395 --> 00:21:22.635
not just taking ownership of what goes
in data security, privacy, et cetera.
00:21:23.025 --> 00:21:27.085
I wanna talk to you about that in a second
for local, but also knowing that the
00:21:27.085 --> 00:21:30.655
output, you also need to think, and when
people say outsource your own thinking, I
00:21:30.655 --> 00:21:34.945
think this is where the real outsourcing
might happen, where the outputs or
00:21:34.945 --> 00:21:38.845
the formulas or the strategies or
whatever it produces, you could take as.
00:21:39.195 --> 00:21:39.945
As truth.
00:21:39.945 --> 00:21:43.125
And I think you just have to counter
argue like, okay, did you look at this?
00:21:43.125 --> 00:21:43.785
Did you do that?
00:21:43.815 --> 00:21:44.595
This research?
00:21:44.745 --> 00:21:47.535
Is there any other implication that I
would be right based on my assumption,
00:21:47.535 --> 00:21:48.795
based on my experience and have a
00:21:49.053 --> 00:21:49.533
Patrick Rooney: Exactly.
00:21:50.095 --> 00:21:53.625
Raul: And usually depending on your
setup I thankfully don't have to argue as
00:21:53.625 --> 00:21:56.445
much anymore 'cause I'm, it's, it's much
more embedded the way that I'm working.
00:21:56.755 --> 00:22:01.315
It's, it's, I mean, for better or for
worse, it's training like 2000 documents
00:22:01.315 --> 00:22:04.535
of mine, like, for like actual writing
and thinking and stuff like that.
00:22:04.535 --> 00:22:07.825
But, But in some cases, if I'm just
using I'm not gonna name the app's
00:22:07.825 --> 00:22:12.145
names, but if I'm just using like an
l an LLM model, frontier model on,
00:22:12.150 --> 00:22:13.845
on one of my my apps on the phone.
00:22:14.245 --> 00:22:16.255
I do have to argue, hey, is
that the right assumption?
00:22:16.375 --> 00:22:17.215
Hey, is that the right thing?
00:22:17.215 --> 00:22:20.395
Can you audit your work and your
research and do some strong man
00:22:20.395 --> 00:22:23.575
and strong man arguments, argue
against yourselves, present to
00:22:23.575 --> 00:22:25.285
me, and then validate like there's
00:22:25.293 --> 00:22:25.593
Patrick Rooney: Right.
00:22:25.975 --> 00:22:29.305
Raul: to think about before you
just accept the output is true.
00:22:29.305 --> 00:22:30.235
So I think that's one key
00:22:30.573 --> 00:22:30.783
Patrick Rooney: Yeah.
00:22:31.135 --> 00:22:36.075
Raul: And one of the, one of the the
biggest rebuttals I have, and I hear, and
00:22:36.075 --> 00:22:40.605
I think it's legit, but I also, I mean
at this point it's, may not be as legit.
00:22:41.015 --> 00:22:44.585
Is like, well, if I just train AI
on all my ip, I lose everything.
00:22:44.895 --> 00:22:46.305
There's a level of truth there.
00:22:46.305 --> 00:22:49.285
But how would you argue that, and
how would you think about that based
00:22:49.285 --> 00:22:53.165
on your actual use case and what
you're seeing even now that you're
00:22:53.165 --> 00:22:54.875
working on your, on your firm?
00:22:54.875 --> 00:22:54.905
I.
00:22:55.305 --> 00:22:57.765
Patrick Rooney: Well, if, yeah, if
you train it, first of all, I think
00:22:57.765 --> 00:23:00.585
that there's a, there's a, a kind
of technological answer to that,
00:23:00.585 --> 00:23:05.345
which maybe is too simple, which if
you, if you own the model, if it's
00:23:05.345 --> 00:23:09.465
locally, you know, run right, you,
you can keep your IP private, right?
00:23:09.465 --> 00:23:11.235
You're, you're not just giving it away.
00:23:11.347 --> 00:23:14.167
Raul: if you downloaded everything,
you got the Mac studio, you put Kimmy
00:23:14.167 --> 00:23:19.087
or GLM, like you're hosting your own
LLM, so like you don't really host
00:23:19.087 --> 00:23:20.197
it to the cloud, so you're good.
00:23:20.597 --> 00:23:23.427
Patrick Rooney: Yeah, so it and
there, you know, on top of that,
00:23:23.427 --> 00:23:26.997
right, there are protections you
can have with cloud-based models.
00:23:27.052 --> 00:23:31.902
I'm not telling anybody to go out and put
all your IP into any of those, but it's
00:23:31.902 --> 00:23:35.892
not simply going to show up, you know,
like across the ocean because you, you've.
00:23:36.612 --> 00:23:38.082
Uploaded a couple documents.
00:23:38.332 --> 00:23:43.632
The but bigger, I mean, the, the other
thing I would emphasize though is the,
00:23:44.322 --> 00:23:49.852
the thought that went into crafting those
materials and the ownership, like the
00:23:49.852 --> 00:23:53.752
actual thought that goes into deciding
that yes, this output is the right output,
00:23:53.752 --> 00:23:56.152
like affirming it, that's still yours.
00:23:56.362 --> 00:23:58.397
So you're the one deciding, hey.
00:23:58.797 --> 00:24:02.517
I, I want to upload this set
of documents, these sets of
00:24:02.517 --> 00:24:04.867
files to train the, the system.
00:24:05.617 --> 00:24:07.297
It's not deciding that autonomously.
00:24:07.687 --> 00:24:10.627
Now, of course, you can get systems
that are more and more capable, right?
00:24:10.627 --> 00:24:14.557
That of grabbing your files and, and
selecting, picking, and choosing.
00:24:14.557 --> 00:24:17.647
But you're still, at some point
you're making that decision.
00:24:18.047 --> 00:24:21.767
You are also ultimately going to
be the one with the, the yes no.
00:24:21.767 --> 00:24:25.387
Say regarding whether it's good
enough for your business, whether
00:24:25.387 --> 00:24:28.507
it's good enough for your clients and
what I've, you know what I'm saying?
00:24:29.077 --> 00:24:32.617
On both the philosophical level and
then on the very practical level is
00:24:33.007 --> 00:24:34.957
don't give up that responsibility.
00:24:35.257 --> 00:24:37.207
I know sometimes it
can be tempting, right?
00:24:37.207 --> 00:24:38.677
Because hey, it's, it's easier.
00:24:39.337 --> 00:24:39.667
Yeah.
00:24:40.387 --> 00:24:40.787
You can still.
00:24:40.934 --> 00:24:45.074
Raul: at the end of the day, legit
at the end of the day where I
00:24:45.074 --> 00:24:47.444
have to make decision because I,
I mean, it's a thought partner.
00:24:47.444 --> 00:24:50.844
Like I've, I've had like a, I have
like nine agents now working with,
00:24:50.844 --> 00:24:52.134
like, I run the business on ai
00:24:52.534 --> 00:24:52.804
Patrick Rooney: Yeah.
00:24:53.021 --> 00:24:55.271
Raul: at the end of the day when
I'm making decisions, sometimes
00:24:55.271 --> 00:24:56.351
it says the recommendations.
00:24:56.351 --> 00:24:58.331
And I'm like, okay, just come
with your recommendation.
00:24:58.331 --> 00:25:02.471
I'm like burnt out on thinking and
this is not, 'cause, I mean, I think
00:25:02.471 --> 00:25:05.051
for a living, like legitimately
and the people say, look, AI's
00:25:05.051 --> 00:25:06.611
gonna come for your job, I think.
00:25:06.931 --> 00:25:07.771
Yes and no.
00:25:08.071 --> 00:25:11.221
It will for like simple things, but for
your, what you just mentioned, what gets
00:25:11.221 --> 00:25:15.361
pushed to reality based on your unique
experience, based on your empirical
00:25:15.361 --> 00:25:19.531
like life experience, not just what's
on text, what's good for the client?
00:25:19.711 --> 00:25:22.741
The judgment call, the trust
call, the relationship call.
00:25:22.981 --> 00:25:26.071
Those are things that aren't gonna
be just swiped away from you.
00:25:26.471 --> 00:25:27.371
Patrick Rooney: Yeah, and, and people.
00:25:28.136 --> 00:25:29.636
I think there are a
couple dimensions of that.
00:25:29.666 --> 00:25:33.956
People still value performance,
like even a podcast.
00:25:33.956 --> 00:25:35.846
There's I think some,
00:25:35.988 --> 00:25:36.408
Raul: Matter.
00:25:36.896 --> 00:25:38.216
Patrick Rooney: excitement and engagement.
00:25:38.216 --> 00:25:40.226
And these are real people.
00:25:40.256 --> 00:25:41.426
They're doing something live.
00:25:41.456 --> 00:25:45.236
Yes, there can be flaws and foibles,
but we're seeing them kind of
00:25:45.636 --> 00:25:49.386
perform for us and produce, you
know, hopefully interesting output.
00:25:49.786 --> 00:25:51.216
Likewise there.
00:25:51.616 --> 00:25:53.536
You just on that, that ownership side.
00:25:53.536 --> 00:25:53.746
Right.
00:25:53.746 --> 00:25:55.666
I think people really value that.
00:25:56.206 --> 00:26:01.286
I can go to a human, I can go to Raul and,
and hold, you know, if I need to hold him
00:26:01.286 --> 00:26:05.756
responsible for, you know, this business
engagement we had, he's the one who's
00:26:05.756 --> 00:26:10.196
going to follow up, you know, and ensure
that things head down the right path.
00:26:10.596 --> 00:26:12.876
This LLM doesn't actually have.
00:26:13.401 --> 00:26:15.351
A, an inherent investment
in what I'm doing.
00:26:15.411 --> 00:26:18.486
Like it, you know, at the end of
the day, it doesn't make money, it
00:26:18.486 --> 00:26:20.241
doesn't feel, it doesn't feel shame.
00:26:20.241 --> 00:26:23.831
It you know, it can simulate
all these things, but it, it is
00:26:23.831 --> 00:26:25.571
not a locus of responsibility.
00:26:26.111 --> 00:26:29.921
So practically, I want that
person on the other side.
00:26:30.361 --> 00:26:33.601
Even if, you know, he uses
AI for 90% of his work.
00:26:33.601 --> 00:26:37.201
I want that person who's responsible
for any high stakes engagement.
00:26:37.601 --> 00:26:39.461
Raul: Yeah, to take the
outcome, take the ownership.
00:26:39.861 --> 00:26:41.091
is, this is really good to understand.
00:26:41.091 --> 00:26:42.381
Again, it's not intelligence.
00:26:42.591 --> 00:26:45.861
One of the things also like maybe
feel free to kinda argue me this,
00:26:45.974 --> 00:26:46.274
Patrick Rooney: Yeah.
00:26:46.281 --> 00:26:48.891
Raul: We were at a dinner table
and we were chatting about this and
00:26:48.891 --> 00:26:52.431
one of the other guys is that was
there, business owner, great guy.
00:26:52.731 --> 00:26:53.721
Actually my connected two.
00:26:54.081 --> 00:26:58.701
But, he, he does a lot of database work
and then puts on LMS on top of that.
00:26:58.701 --> 00:27:02.271
So he works, he works at the data layer
and then the, the model layer and then
00:27:02.271 --> 00:27:04.321
the application layer, for businesses.
00:27:04.411 --> 00:27:06.991
And he was saying, 'cause he,
he like same perspective, like
00:27:06.991 --> 00:27:07.921
this stuff is not intelligent.
00:27:07.921 --> 00:27:08.761
I know how it works.
00:27:09.181 --> 00:27:12.421
However we perceive it as
culture that it is intelligent
00:27:13.141 --> 00:27:14.311
it's passed the terrain test.
00:27:14.701 --> 00:27:18.271
And then when we all believe
that it's reached a GI.
00:27:19.066 --> 00:27:20.146
Whatever that means.
00:27:20.424 --> 00:27:20.814
Patrick Rooney: Yep.
00:27:20.956 --> 00:27:24.506
Raul: agi, we did define terms, but
if we perceive that and we behave
00:27:24.506 --> 00:27:28.736
that way, then regardless if it did or
didn't, it did because we're behaving
00:27:29.516 --> 00:27:34.316
if it did as culture, even though
the truth underlying may be not.
00:27:34.316 --> 00:27:36.236
So what's your take on that?
00:27:36.836 --> 00:27:38.276
I mean, based on everything that we just
00:27:38.434 --> 00:27:42.214
Patrick Rooney: Well, I, you know,
I'm gonna maybe, I put some words
00:27:42.214 --> 00:27:45.934
in his mouth, but he's basically
saying if it, if it tricked me into,
00:27:45.956 --> 00:27:47.036
Raul: the argument No, I'm, I'm
00:27:47.104 --> 00:27:47.344
Patrick Rooney: yeah.
00:27:47.366 --> 00:27:48.206
Raul: argument that,
00:27:48.664 --> 00:27:48.994
Patrick Rooney: Yeah.
00:27:49.016 --> 00:27:50.426
Raul: culture, if we, we
believe because he's saying
00:27:50.524 --> 00:27:51.064
Patrick Rooney: Okay.
00:27:51.116 --> 00:27:51.506
Raul: told him, well,
00:27:51.604 --> 00:27:51.874
Patrick Rooney: Yeah.
00:27:52.646 --> 00:27:53.366
Raul: that it's there.
00:27:53.396 --> 00:27:54.446
We're gonna behave as if it's
00:27:54.574 --> 00:27:54.724
Patrick Rooney: Okay.
00:27:54.866 --> 00:27:56.006
Raul: So therefore it's there.
00:27:56.406 --> 00:27:59.206
Patrick Rooney: Yeah, no, well that
actually, that's kind of the, the
00:27:59.206 --> 00:28:02.746
original premise of the Turing test,
that if it can deceive you, it's real.
00:28:02.896 --> 00:28:03.166
Right?
00:28:03.166 --> 00:28:05.596
That if it can deceive
you, that it's intelligent.
00:28:05.596 --> 00:28:06.346
It really is.
00:28:07.096 --> 00:28:08.356
And what I think.
00:28:08.911 --> 00:28:13.561
We want to go back to this foundational
distinction in just a history of thought,
00:28:13.561 --> 00:28:15.931
which is appearance and reality, right?
00:28:15.931 --> 00:28:21.421
So if I can, if wax sculpture, for
example, can trick me into thinking
00:28:21.481 --> 00:28:24.911
Michael Jackson is, you know, right
behind my shoulder, is Michael
00:28:24.911 --> 00:28:26.021
Jackson really behind my shoulder?
00:28:26.021 --> 00:28:31.221
And is he, is he there until I, I reach
out and feel it and realize, oh no,
00:28:31.221 --> 00:28:32.721
this is not, this is not a human being.
00:28:33.081 --> 00:28:34.431
This is a sculpture.
00:28:34.831 --> 00:28:36.001
The answer is clearly no.
00:28:36.061 --> 00:28:36.541
You know it.
00:28:37.051 --> 00:28:42.391
The fact that it was deeply convincing
that it was maybe based on a detailed
00:28:42.391 --> 00:28:46.981
study of like every feature of his
face, his posture, his clothing.
00:28:47.213 --> 00:28:48.323
Raul: to that future though, we're
00:28:48.481 --> 00:28:48.541
Patrick Rooney: Yeah.
00:28:48.563 --> 00:28:52.163
Raul: to that future world consensus
of the, of society and culture.
00:28:52.283 --> 00:28:55.593
Even though right now in US, the sentiment
is really low for using these tools.
00:28:55.593 --> 00:28:56.673
Hopefully, we'll, we'll flip that.
00:28:57.033 --> 00:29:01.023
But we're going to that future where,
where there's gonna be a culture
00:29:01.143 --> 00:29:03.093
alignment saying like, yes, this is it.
00:29:03.393 --> 00:29:06.273
This wax statue is real,
even though it's not real.
00:29:06.603 --> 00:29:08.733
How do we, how are you
gonna lead as a founder?
00:29:08.733 --> 00:29:10.413
What are the things you're
gonna work with your clients on?
00:29:10.413 --> 00:29:11.643
How are you going to reframe?
00:29:11.643 --> 00:29:14.463
Like, what's, what are the real
implications you're gonna take?
00:29:14.463 --> 00:29:17.253
'cause I think that's the future we're
gonna head in because no one's gonna
00:29:17.253 --> 00:29:20.673
be able to judge it in my opinion,
like if they reached it or not.
00:29:21.033 --> 00:29:23.043
'cause they're, everyone's saying,
we reached a GI, it's here.
00:29:23.341 --> 00:29:23.671
Patrick Rooney: Right.
00:29:23.913 --> 00:29:26.283
Raul: That, that was like leaked
two weeks ago, three weeks ago.
00:29:26.683 --> 00:29:30.593
whether it's there or
not, one, does it matter
00:29:30.986 --> 00:29:31.406
Patrick Rooney: Mm-hmm.
00:29:32.123 --> 00:29:34.733
Raul: Two, if it does matter,
what do we do about it?
00:29:35.133 --> 00:29:38.013
And three, based on what we do
about it, how do we practically
00:29:38.013 --> 00:29:39.063
implement that in business?
00:29:39.716 --> 00:29:43.076
Patrick Rooney: Yeah, I mean,
it's a fantastic question and I'll
00:29:43.076 --> 00:29:44.576
throw out a, a couple thoughts.
00:29:45.266 --> 00:29:49.966
I think it, it, it's doubling
down on human connection on
00:29:50.086 --> 00:29:51.736
in-person human connection, right?
00:29:51.736 --> 00:29:56.556
And I'm not saying this to undermine,
you any online businesses like I've,
00:29:56.676 --> 00:29:58.626
I've worked remotely for several years.
00:29:58.626 --> 00:30:02.436
I'll continue to engage in business
online, but I think maintaining that
00:30:02.436 --> 00:30:07.626
human contact is a, a strong reminder to
us of what we actually can do in person
00:30:07.626 --> 00:30:08.826
when it's like, oh, I can see that.
00:30:08.946 --> 00:30:11.166
Yeah, your, your coffee cup just spilled.
00:30:11.166 --> 00:30:12.186
Let me help you clean that up.
00:30:12.186 --> 00:30:12.366
Right?
00:30:12.366 --> 00:30:17.076
It's tiny things that make a lot
of difference when we have those
00:30:17.076 --> 00:30:18.636
tangible elements in our relationship.
00:30:18.636 --> 00:30:19.031
So I would say.
00:30:19.671 --> 00:30:23.061
To the extent you're able to do that,
even if it's up not in work, but
00:30:23.061 --> 00:30:24.981
personal relationships cultivate that.
00:30:25.651 --> 00:30:29.981
I think that's, that will then having
that as a primary foundation will then
00:30:30.221 --> 00:30:33.851
kind of spill over into the honor line
relationships and I think ground them.
00:30:33.911 --> 00:30:38.751
So my, my view is having those kind
of tangible in-person relationships
00:30:39.201 --> 00:30:41.721
helps make the online site stronger.
00:30:42.351 --> 00:30:44.541
It's a, it's a, it's having
the right foundation.
00:30:45.441 --> 00:30:49.706
The, yeah, the other analogy, I
just, it's kinda like video games
00:30:49.706 --> 00:30:53.856
versus, you know, playing sports and
video games versus in person, right.
00:30:53.976 --> 00:30:57.906
You can, you can pick up some interesting
coordination skills, et cetera,
00:30:57.906 --> 00:31:02.466
playing whatever tennis like used
to be we tennis like long time ago.
00:31:02.466 --> 00:31:02.706
And
00:31:02.728 --> 00:31:02.938
Raul: yeah.
00:31:02.938 --> 00:31:03.088
Yeah.
00:31:03.088 --> 00:31:03.478
That was fun.
00:31:03.576 --> 00:31:03.936
Patrick Rooney: Yeah.
00:31:04.618 --> 00:31:04.798
Raul: Then
00:31:04.806 --> 00:31:05.106
Patrick Rooney: And it's,
00:31:05.128 --> 00:31:06.058
Raul: VR glasses.
00:31:06.366 --> 00:31:09.096
Patrick Rooney: you can get, you can
get better at it, but you, you're
00:31:09.096 --> 00:31:10.836
still gonna f find out there's a gap.
00:31:10.866 --> 00:31:12.471
There's a huge skill gap when
you go out and you touch.
00:31:12.671 --> 00:31:14.141
Tennis racket in the real world.
00:31:14.891 --> 00:31:20.061
So do, do those, like get grounded
in those, those activities?
00:31:20.121 --> 00:31:25.041
I think that, again, they, they'll spill
back over and help you with all the,
00:31:25.441 --> 00:31:27.361
what I'll call derivative activities.
00:31:27.361 --> 00:31:29.311
That is, they're, they're
built on that foundation.
00:31:29.311 --> 00:31:32.611
So the online, the
simulated stuff with LLMs.
00:31:33.011 --> 00:31:33.731
Those are all good.
00:31:33.731 --> 00:31:37.361
They're great tools for building
business connections for for building
00:31:37.361 --> 00:31:40.241
projects, but don't ever let that become
00:31:40.448 --> 00:31:40.728
Raul: value.
00:31:41.501 --> 00:31:42.041
Patrick Rooney: from the
00:31:42.433 --> 00:31:42.943
Raul: Yeah.
00:31:43.096 --> 00:31:43.376
Patrick Rooney: concrete.
00:31:43.776 --> 00:31:44.856
Raul: Just having that recognition.
00:31:44.856 --> 00:31:48.396
I think the, the practical takeaway to
a GI here, not regardless, relationships
00:31:48.396 --> 00:31:53.126
matter, but also even in my own, like
going deep, like real deep, trusting
00:31:53.126 --> 00:31:57.536
your own human based intuition
and judgment based on experience.
00:31:57.596 --> 00:32:01.256
Not only just what the model outputs,
even if the model is super compelling.
00:32:01.656 --> 00:32:05.256
The other approach is the other, the
other key thing to keep in mind is
00:32:05.256 --> 00:32:10.026
also you, you have the ability to know
what's right and to know what to do.
00:32:10.416 --> 00:32:12.546
So don't plaque on your own reasoning.
00:32:12.974 --> 00:32:13.214
Patrick Rooney: Yep.
00:32:13.296 --> 00:32:15.516
Raul: something to keep in import,
like how, I forget how much
00:32:15.516 --> 00:32:19.366
we invested into, trillions of
dollars into training these models.
00:32:19.366 --> 00:32:24.276
Now I'm pretty sure like cumulatively
yet your brain getting, getting cost a
00:32:24.276 --> 00:32:28.196
trillion and, it's kind of fascinating
when you think about it that way, that,
00:32:28.474 --> 00:32:28.564
Patrick Rooney: Yes.
00:32:29.006 --> 00:32:31.926
Raul: we take we take our own
brain thinking for granted.
00:32:31.926 --> 00:32:34.236
So you can also reason your way through.
00:32:34.944 --> 00:32:35.234
Patrick Rooney: Yeah.
00:32:35.646 --> 00:32:36.966
Raul: and then also vernacular matters.
00:32:36.966 --> 00:32:41.046
The way that you talk to it, the way
that you personified or not personified
00:32:41.046 --> 00:32:44.386
and the way that you train your team
is important because your team may
00:32:44.386 --> 00:32:45.616
not be having these discussions.
00:32:45.616 --> 00:32:48.076
You may not be thinking about this day
in and day, and it's easy to go with the
00:32:48.076 --> 00:32:50.366
flow and to get sucked into whatever.
00:32:50.446 --> 00:32:52.846
Whatever's being pushed
in terms of ideology.
00:32:52.846 --> 00:32:54.946
So it's just good to have your own
stance and your own perspective.
00:32:54.946 --> 00:32:57.316
Even if you disagree, that's great,
but at least you thought about that
00:32:57.616 --> 00:33:00.866
and you thought about your own thinking
which I think very few people do.
00:33:00.896 --> 00:33:03.056
'cause thinking about your own
thinking is called writing.
00:33:03.296 --> 00:33:04.706
It's definitely right about how you think.
00:33:05.099 --> 00:33:05.579
Patrick Rooney: Yes.
00:33:05.906 --> 00:33:09.956
Raul: I think is a good loop on
where can people find you if you're
00:33:09.956 --> 00:33:12.776
writing or posting more about this,
and also sharing your insights.
00:33:13.056 --> 00:33:14.916
Where's the best place
for people to go to?
00:33:15.316 --> 00:33:17.926
Patrick Rooney: Yeah, right now
LinkedIn is a great place to find me.
00:33:18.226 --> 00:33:20.626
I'm happy to share
contact details with you.
00:33:20.926 --> 00:33:24.586
I've also got a, a business
website, leonis strategy that's,
00:33:24.706 --> 00:33:25.696
that's a little more limited.
00:33:25.696 --> 00:33:28.846
So I'd say that the, the LinkedIn
side of things is gonna be perfect
00:33:28.846 --> 00:33:29.836
for anybody who wants to connect.
00:33:30.236 --> 00:33:30.746
Raul: Sounds good.
00:33:30.746 --> 00:33:32.696
We'll put those links in the show notes.
00:33:32.696 --> 00:33:34.646
Patrick, thank you again for being on.
00:33:35.046 --> 00:33:36.276
Patrick Rooney: Thanks so
much for having me, Raul.
00:33:38.526 --> 00:33:40.036
Thank you for listening to the podcast.
00:33:40.056 --> 00:33:42.266
If you'd like to get more of
these podcasts, please go ahead
00:33:42.266 --> 00:33:45.446
and subscribe to wherever you're
watching or listening to the pod.
00:33:45.866 --> 00:33:49.016
If you have any suggestions or questions
that you want me to answer on the
00:33:49.016 --> 00:33:51.996
pod, or you have a guest that you want
me to interview, please go ahead and
00:33:52.016 --> 00:33:56.926
email me all of your suggestions and
questions at podcast at do good work.
00:33:56.956 --> 00:34:00.206
io that's podcast at do good work.
00:34:00.236 --> 00:34:01.186
io.
00:34:01.626 --> 00:34:04.386
If you'd like to give me public
feedback, you can go on Apple
00:34:04.406 --> 00:34:07.986
Podcasts, and from there, you
can leave up to a 5 star review.
00:34:08.226 --> 00:34:09.746
I would greatly appreciate that.
00:34:10.206 --> 00:34:13.936
If you'd like more free resources where
you can get my best strategies to help you
00:34:13.946 --> 00:34:18.196
increase your company's performance and
scale profitably, where you can get hand
00:34:18.196 --> 00:34:22.156
picked articles to propel your growth, and
you can get trainings and discussions that
00:34:22.156 --> 00:34:27.201
I give freely online, as well as low cost
resources, Such as my books and guides.
00:34:27.511 --> 00:34:30.771
You can get all of that
by going to dogoodwork.
00:34:30.821 --> 00:34:36.231
io forward slash resources 90 percent
of all these resources are not behind
00:34:36.231 --> 00:34:40.551
any email opt ins you can get instant
access by going to dogoodwork.
00:34:40.661 --> 00:34:46.176
io forward slash resources Now, if
you'd like to accelerate your progress
00:34:46.226 --> 00:34:50.646
and shorten the gap between information
and action and start seeing results
00:34:50.646 --> 00:34:54.356
in your business, let's work together
to increase your company's performance
00:34:54.386 --> 00:34:57.716
and scale profitably and serve
more clients without the overwhelm.
00:34:58.226 --> 00:35:01.916
You can request a free clarity call
to see how we can best support you to
00:35:01.916 --> 00:35:04.376
reach your goals by going to DoGoodWork.
00:35:04.416 --> 00:35:06.206
io forward slash apply.
00:35:06.386 --> 00:35:08.206
Again, that's DoGoodWork.
00:35:08.206 --> 00:35:09.616
io forward slash apply.
00:35:10.056 --> 00:35:13.556
As always, it's an honor to be
a small part in your journey.
00:35:13.856 --> 00:35:15.006
This is Raul Hernandez.
00:35:15.456 --> 00:35:15.886
Do good work.