Subscribe
Share
Share
Embed
Do Good Work is not a label but a way of living.
It is the constant and diligent effort to achieve a new level of excellence in one’s own life.
It is the hidden inner beauty behind the struggle to achieve excellence.
It is not perfect but imperfect.
It is the effort, discipline and focus that often goes unnoticed.
The goal of this podcast is to highlight that drive.
The guests I have on this show emulate this drive in their own special way. You’ll be able to apply new ideas into your own life by learning from them.
We will also have 1on1 episodes with me where we’ll dive into my own experiences with entrepreneurship and leadership.
Every episode is designed to provide you with ideas that you can apply and grow in excellence in all areas of your life, business and career.
Do Good Work,
Raul
INTRO
Today on the podcast, I'm
joined by Taylor Black.
Taylor directs AI and venture
ecosystems inside of Microsoft's
Office of the CTO, where he runs
the incubation studio and co-leads
a 25,000-member technical community.
He also founded the Leonum Institute
for AI and Emerging Technologies at
the Catholic University of America.
He teaches at the University of
Washington Foster School of Business
and advises the US Embassy to the
Holy See on AI and human flourishing.
All of this ties to his single conviction
that the right response to more
powerful AI is better-formed humans.
The conversation we had centers on a
distinction where Taylor draws a hard
line between intelligence and insight.
A machine can gather all the evidence,
run all the patterns, and tell you,
"This is probably true," but only a
human knower can make the categorical
leap into, "Yes, this is true."
He calls reaching that kind of
certainty the virtually unconditioned.
What that means in plain English is
that there's a depth of understanding
that compute cannot totally replicate,
but the more that AI can do, the more
it actually matters that the human
running the machines has actually
developed that depth of thought, and
in my opinion, that's a call to action.
We get into what he calls the
anthropology-first approach to
building ventures, the idea that
every product you ship carries an
implicit view on what a human being
is, whether you intend it or not.
We also talk about the danger of using
AI as an oracle that lets people skip
the productive struggle we also get
into his concept of cold storage,
where he uses AI as a context engine
to shelve good ideas early and monitor
it for the right conditions in the
marketplace or trends or behaviors that
could revive good ideas for the future.
And we also get into the question that
should be on the top of the mind of anyone
developing or building a business today.
The question is not what can we
build, but what ought we build?
All right.
Let's dive into it.
PODCAST
. Raul: Taylor, you mentioned
something specific around AI is
gonna be far smarter than humans.
However, that means we need to have
well-formed humans, and I wanna kinda
understand your thesis because we're
probably gonna spend the majority of our
conversation breaking bit by bit down.
if AI's gonna have abundance and
intelligence will be abundance,
then impossible tasks will seem
more possible to us, or at least
merely hard tasks, not impossible.
But then you also argue that we still
need insight and that a human person is
the only one that can provide insight.
And then the argument that we need to
form better humans, which I am completely
aligned with all those three arguments.
The question that I have, though, is
if that's true, how do you distinguish
between intelligence and insight?
Taylor Black: Certainly.
Yeah, so we have this, uh, the AI
is based off of the rough biological
structures of our own minds neurons
and all that wonderful stuff.
But the, but we know things in
a variety of different ways, and
there's been a bunch of literature
on this, both in philosophy in
books like Blink in neuroscience
tomes, all that sort of thing.
But, the one that I that I think is most
relevant for our conversation today is
to look into your own interiority of how
you've ever come to know something, right?
And I'm gonna, I'm gonna articulate
this for f- in a felt sense and
any of the listeners to, to our pod
today can do this themselves too.
But, there's this concept of human
knowing, a- and we know in a variety of
different ways called statistical knowing.
And that, that is where we have
enough data points that we are pretty
sure that something is the case.
We're not certain about it, but
we're certain enough for the piece of
knowledge s- serving in a certain way
for the groundedness of our reality.
so for example, like if you're
driving a car and you're taking your
car around the curve of a road, you
aren't s- absolutely certain that
you're following the perfect curve in
order to take your car on that turn.
Don't have all of the data points
down to the level of granularity that
you actually feel like you mightâ¦
You also don't have all of the mechanics
between your s- hands on the steering
wheel all the way down to the dri- the
steering train for your car measured
out to that level of accuracy too.
There's kind of a good enough for
you driving that car around the curve
that you're comfortable with and
Raul: Practical
Taylor Black: It's practical.
It's practical.
It's a, "Yeah I got that there."
and so I give that as an
example of statistical knowing.
there's another kind of knowing though
that h- that is unique to humans,
and this is where I c- I think the
categorical difference between human
knowing and statistical knowing is.
Human knowing in the more like the
insight sort of fashion that that
you mentioned in your question.
And when you have an insight, you are
grasping what what I would call the
virtually unconditioned, which means that
you've all the further pertinent questions
that you need to pursue in order for you
as a knower to posit, "Yes, that is true
And there's a number of different things
in your life where you do this, right?
Some of my favorite examples end
up being like the big choices
we've made in life, right?
Getting married or believing in,
in some sort of religion or or
knowing that you would do like
anything for your child, for example.
even knowing things that help
frame your entire worldview.
Now, this is a, this is very strange kind
of knowing to talk about it in kind of
the, um, uh, the secular sense or, the
even the neurological sense because it's
very different from w-what I would call
scientific knowing, ends up being a this
is true given i-in a very strict under
a very strict set of conditions that
I'm able to replicate in a lab, right?
a lot of people might say
that they believe in science
or that science is real.
but if you un-unpack that a little
bit, it's really hard to live your
life completely according to everything
that is known only by science, right?
b-because
Raul: You can't prove math exists.
It just exists
Taylor Black: exactly right.
It's really complicated.
And so the, what we actually believe then
forms our worldview and forms the way in
which we bop about in our existence, ends
up being much more following that, that
unrestricted desire to know, which each
of us has inside us, to follow out those
further pertinent questions to say, "Yep,
I believe this to be true for myself.
And as a result of me believing it
to be true, I believe other people
also un-- could see this as truth."
and that moment of insight is when you
are able to come to that realization
that something is true, that something
is real in that particular sense.
Raul: That we can grasp it.
I think, one thing that I've read and
heard was that when you understand a
concept, like you understanding animal
kingdom or what a universe is, it's not
the, a mi- it's not a duplicate of that
in existence in your mind, but it's the
actual mirror of that thing in your mind
that you can harness and understand.
Which is I think what you're saying
is more of a metaphysical knowing
at that point, and that's, we're
not gonna dive too deep into that
stream, but that's the abstract
Taylor Black: It is.
Yeah.
And you can also think of it as
you've understood something about
reality, knowing that there is so
much more to understand about it.
So for, for example if you're walking
down my street with my kids, right?
And I look at a house.
I can see the paint on the house.
I've actually framed and built
a couple of houses when I worked
in construction in high school.
And so as a result of that experience
that I've had, I can look at a house
and see things that somebody who
hasn't framed a house can't see,
Because they aren't looking-- That
isn't part of the further pertinent
questions that they're asking.
Now, the same thing can happen w-with if
I walk with my friend who's a botanist.
She will be able to see things in flowers
that I can't see, that aren't there
Raul: the nuance
Taylor Black: of reality sort
of sense because I'm not asking
that kind of question around it.
And so you can imagine the universe kind
of being this way for everybody, which
is there are further pertinent things I
could ask about almost anything that I
encounter, all of that is being mediated
by my understanding, being mediated by my
v- my wanting to seek out those further
pertinent questions so that I can reach
that virtually unconditioned so that I can
have the insight into what that thing is
Raul: And what you're saying is that right
now when we leverage this, quote-unquote,
"intelligence," and I think the, the
vernacular has been argued, is it the
right word to use or not, but that's what
we're calling it just moving forward.
We're renting or we're
using this intelligence.
We're renting or running this compute, but
it's different from the actual knowing.
How do we-- how do you distinguish
that though in practical use cases?
And we can talk about like ventures
and business and value creation,
but how do you distinguish that
specifically in practical execution
for daily using the technology?
Taylor Black: Yeah, certainly.
So they end up being ways in which--
a tool that we can use in order
to come to insights ourselves.
I see LLMs, generative AI as being really
good at helping us pattern our experience
so that we're able to come to an insight.
If we ask it for like true
information, and this ends up being
kind of an alignment question, which
I love the term alignment because it, it
begs further pertinent questions, right?
Alignment to what?
But if we're asking that kind of
question, it's "Hey, is this true?"
The machine can't really say, "Yes,
I posit this as an unders- like
the, the correct understanding
of reality," which is what we're
saying when something is true.
will reference in its own, training
data or internet data "Here are
the sources that I relied on in
order to come to the very strong
recommendation that this is the way
in which you should perceive reality.
I might s- I'm not saying it, I, I,
Claude, am, am positing that this is true.
I am assembling all of these
under- un- underlying resources and
telling you this is likely the case.
Now you g- knower, go and
make that affirmation."
Even even if we do architect the system
so that it would say something is true,
that's our architecture of it, where
it is comparing what we ha- what we've
asked or comparing what data it has
against something that some developer
or us as a builder of the system
has said, "When it looks like this,
then you can say that it is true."
The machine itself can't
make that categorical leap
Raul: Yeah, and I think
that's an interesting point.
And we-- The practical, one
of the practical phraseologies
is you can't outsource your
judgment or your own thinking.
But again, we can dive into like
mechanics and I would like to look
at, I like to call them ingredients
of how you actually run them.
You don't just go to this thing.ai
or .com
and you start chatting away.
That's like a basicâ¦
That's a basic approach, but there's
much more that we can dive on.
What I do wanna understand, though, is
y-the, I don't know if it's the second
or the first, forming better humans.
And I wanna understand when you see this
in real life, like running new ventures
either at Microsoft or new ventures
in the venture studio how does thatâ¦
'Cause that's the thesis.
If you're better formed, then
you're gonna be able to have
better in- insight or in- intuition
into whatever for value creation.
How have you seen that with the ideas
or concepts or value creations, we'll
call them ventures, inside a big, a
huge firm, and then as well as what
you're building out- outside of that?
Taylor Black: Yeah, certainly a lot
of it has to do with the way in which
you're architecting the conditions
under which a decision can be made.
one of the reasons why, and one of
the major things investors look for in
a founding team, for example, is not
only kind of the broad subject matter
expertise, but the tenacity to seeâ¦
to s- be able to sit with the tension
of inquiry b- and extend that tension of
inquiry for as long as is necessary to
posit that something is actionable, right?
y- you can imagine this like the,
the archetype of course is like the,
the researcher in his study or her
study, like slaving away at the table,
like running the experiment, and
knowing that it isn't good enough.
Like theyâ¦
the data isn't there for them to
say, "Ah, at last I have come to the
realization," or, "At last I have the
answer to this particular question."
They are able to sustain
that tension of inquiry until
they, that checks out, right?
And that ends up being a major formative
piece because you can truncate the
search and say, "Ah, it can't be
found," or, "Ah, this is good enough,"
but still have that kind of lingering
feeling in the back of your head like
I-- there's still the work to be done.
and to me, that ends up being one
of the major pieces of formation.
if you're not able to sit with
that tension in finding the,
right answer and finding the what
qualifies as product market fit.
We have all these amazing
stories about jobs, right?
Where apocryphal or not, dropping the
iPhone into the, of water and seeing
bubbles is like, "Hey, you, we still
have space we can use in there."
it's that sort of drive where
like, "No, this isn't good enough."
Like we can still get
there and we can't stop.
and knowing and having the
judgment to know when you do need
to stop too because there areâ¦
Raul: Yeah, no, there, there's
a fine line between pushing.
Taylor Black: to
Raul: How long is your ultra?
When you do ultra running, like
technically ultra's beyond 24 what, miles?
When do you stop?
When do you call like that?
Now it's like a-- So that
makes a lot of sense.
But practically, so you're hit-
you're hitting an anthropologically
or anthropology first approach to
magic creation, and you're saying
that it matters a lot more than
just being a practical founder
and practical product market fit.
There's much more to push.
And I think that that's-- Sounds
like a leadership aspect and
like a culture aspect versus just
the practicality of a product.
Taylor Black: it is, but also
there's, time you have a human, like,
building something, they're making
decisions around what to build.
And so even laying out and so I work
particularly in venture studios, and
venture studios are just frameworks for
running rapid experimentation and making
decisions more quickly and thoroughly
as a result of rapid experimentation.
And so it-- ev-even rise and fall
of nations ends up having to tying
back to how individuals do or do not
sit with that tension of inquiry and
come to better judgments and better
understandings as a result of that.
so I wrap in this kind of cognitive
theory into venture building itself
because it's that, that is it good
enough for my customer or is it
something that I'm just willing to ship?
and it's there that the, the tension
lies of I need some more data before
I can make the decision around that
virtually unconditioned as to whether
this interface is sufficient for my
customer journey so my customer, delights
in it and I find my product market fit.
and therein and there's frameworks
to make that easier for us as humans.
But us as humans also need
to have that tenacity, have
that, that drive to that point.
Raul: It's a relentlessness
approach to it specifically.
Taylor Black: Yep
Raul: Do you think that's the
only principle you're looking
at on the anthropology first?
Like the tension of good enough to
ship and sometimes people are lazy.
It's like it's good enough for now, but
always consistently R&D-ing and improving.
So I get that reality.
Uh, and I'm on the ship first approach.
Ship first, figure it out,
continue to improve as we continue.
I, I'm just like, that's just who I am.
But is that the only principle?
Is there something else that
you're looking at on that axis?
Taylor Black: right?
what you understand about your own
coming to aha moment of "Oh, I found
it," is that no, the best way for me to
gather data as to the further pertinent
question of product market fit is for
me to ship smaller, small aspects of
what I am building, get that feedback
loop so that I then can say, "Ah, yes,
okay, I solved that minor problem that
leads toward the, towards answers to the
larger problem of product market fit."
And so it's being aware of your
metacognitive strategies as you approach
your work, I think is invaluable in
your work, and even more so when you're
working alongside an intelligence
like artificial intelligence where
you are offloading pieces of what,
of your thinking in thoughtful,
appropriate ways to the machine that
is cranking through the examples or
cranking through the data or cranking
through the the build, the code, what-
Raul: The build plan and the execution.
Yeah
Taylor Black: where you're comfortable
with that because it's fitting within
your overall framework of it needs to be
able to answer these evals whether that's
evals built or the eval inside your own
head, which is this is enough data for
me to posit reality and move forward
to the next virtually unconditioned.
Raul: I like to look at
some of those frameworks.
One of the things that I've done is
created hard rules, like laws, right?
So a hard rule is that the agent,
if it's in its memory, it might say,
"Psh, I'm not gonna follow this thing."
But if you make it a hard rule like
a law, then it has to pass through
certain gates for it to actually,
Take-- I'm explaining
that to the audience.
I know you get that back of your hand.
And one of the things that I've
done specifically is like a,
an ethos of human flourishing.
Like you can't just agree, you
have to find truth, you have
to push me to the extreme.
You can't just do something to cut
a corner if it's gonna harm a human
or if it's gonna be just to get
to the right answer, cut corners.
Like all these, like this nine principles,
fundamental principles, and I've
instituted them as laws in the, like
the infrastructure that I built, which
helps me help it make sure that we're
following what I believe through my
insight to be a path forward to build.
And it has proven, at least
to this point, fruitful.
So I think that's a practical takeaway.
I wanna ask you and look at the
other axis of insight or another
prong of insight, and that's the
prong of what ought we to build.
And it's not what should you
build, 'cause you can literally
build anything right now.
And I-- And one of the key
things, like you gotta point these
tokens to something valuable.
And me being the knucklehead
I am I just pointed it to
revenue stuff in the beginning.
Listen, playing around since the
ChatGPT days to now, like it's all rev.
How can I increase revenue and
make the ROI super obvious?
And then now with more
products and whatever.
But what ought we to build?
Better h- better form
humans, more insight.
How do we answer that question of ought?
Which is different what
should you or can you.
Taylor Black: yeah.
It's, and it's a hard question, and
I think it ends up being unique to
every individual, am a zero to one guy.
I like once it gets to one, once
you find product market fit, my
interest, dries up very quickly.
And we all know if you don't, if you're
not like super interested in the thing,
you can probably grind through it, but
it's not gonna be as good as the, the
guy who loves running from one to 100.
And you can tell the energy difference.
You can tell the desire to follow
through on the tension of inquiry for
answering the next set of questions that's
necessary to get from one to 100, right?
I'm not that guy.
I'm the zero to one guy.
I I like the thing that's
like, what if this was a thing?
And let's prototype it out.
Let's figure, turn the idea into
something that we can iterate on,
something that we can wrap a business
model around, something we can
find some customer validation on.
And so I think the answer to that
question ends up being very unique,
and this lends itself even to that
formational aspect where if I don't
know that about myself, then I could
be stuck doing stuff that really isn't
bringing me meaning or bringing me joy.
Not that everything has to, right?
We've all ground through the,
Raul: You learn through that experience.
It's a tough tack sometimes if you
don't feel aligned, but know thyself
Taylor Black: But there's a
formational aspect to it, right?
And it's hard to do without that pain
Raul: Yeah, no, yep, absolutely.
So that's an insight of knowing
yourself and that insight that you
bring to the table when you build.
What have been some of the insights that
either you're leveraging or depending
upon for what you're building right
now, or that you might have learned
as you build from the zero to one?
Being the zero to one guy, but
just consistently over time
Taylor Black: Yeah, I think that there's
a number of different things, I think.
One of them, one of the fundamental ones
I think is this concept of humility.
And humility for me is being able
to see a situation from somebody
else's perspective, right?
If I'm proud, then I'm preferencing
my own perspective over some,
o-o-over something else.
Where if you're humble, then you're
doing a better job at listening and
you're able to see reality through
somebody else's eyes in a better
way I think it's fundamental to
innovation because it's fundamental
to understanding the problem better.
I don't understand the problem from
the perspective of my customer,
then I'm not able to build what
my customer actually needs.
So that, that's an aspect of it
an aspect of being able to sit
with and empathize and see reality
through the kinds of problems that
your customer is encountering that
you have this desire to solve.
I've seen many and I've been one of the
entrepreneurs who built the thing and
fell in love with the solution, right?
And falling in love with s- the
solution is, of course, the, the
fastest way to tank your startup
because you lose sight of the problem.
So that's one thing applying this
framework to, to venture building.
The other thing too is that when you
come to understandings of how of how you
operate and how you believe the world
operates as a result of kind of this
framework for thinking, to distill out
certain kinds of playbooks for solving
certain kinds of operational problems.
So the venture studio mechanism in
my mind is, and I've built several
of them, a way of kind of applying
this this concept of idea flow.
Jeremy Utley wrote who teaches
at the Stanford d.school,
wrote this excellent book
on what he calls idea flow.
But it's a, a methodology, methodological
for taking an idea experimenting with
it, stretching the idea out, prototyping
it not just from a, a product designer
perspective, but also a, a business model
perspective, also a, um, understanding
the problem better s- perspective.
And it ends up being a rubric that you
can follow for iterating through idea
to prototype to incubation to launch.
And being able to have a framework
to manage your thinking, to know
where you are in the different loops
Raul: Yeah
Taylor Black: of doing that so that it
becomes more efficient that ends up being,
I think, a really powerful, thing falling
out of this understanding of how human
minds work that I found a lot of value
in over the course of my career as well.
Raul: That's fascinating how human minds
work and fall- the falling out of that,
and it's, I like how you framed that.
To, to shift some of the
conversation, I think I call it
like the, the theology of AI.
Like some people worship it,
other people fear it, and that's
like a, an oversimplification
of the scenery out there.
Like there's multiple schools of thought.
There's nine or 12 prominent ones
and then big categories around it.
Working with multiple, founders and
venture studios, team members, personnel,
like who have different viewpoints of
the technology, how do you harness all
that to the end goal that you have, which
is an anth- anthropology first approach
and as well one for value creation?
How do you work with all those dichoto-
like they could be dichotomies in
mindsets and approaches to the technology?
Taylor Black: Yeah, it's interesting.
I think a really powerful way I've
found of framing this is, um, from the
user experience problem that, uh, that
your user co-creating products with you
Right?
Three-- before November 2022, we mostly
as technologists built deterministic
products where we could test out every
possible experience that the user would
have as a result of using our thing.
And we could put it in front of as,
and as we were building it, we could
put it in front of users and tell
them to do the thing and watch them
do the thing and then understand what
was happening from from soup to nuts.
Probabilistic products, and
this is both agentic, right?
The user is co-creating experiences
that we can't fully anticipate.
Raul: And that are unique
to them, by the way.
Depending on how they use the
technology or if they're using a
subscription, it behaves differently
depending on how you behave and the
model, how it actually behaves based
on your worldview and demographic for
some models, which is fascinating.
So it's a two-sided approach
Taylor Black: it's, yeah, it's, and it's,
and it makes it so complicated, right?
Particularly 'cause a lot of sta-
is happening inside their head.
You can ask them, like, how
they're thinking about it and
you can get some kind of answer.
But generally, people aren't gr-
very great about thinking about
how they're thinking either.
And there's this gap now between
you being able to understand what
the user is actually experiencing
when you're building your product.
And so why does this
matter for product makers?
You're making a product for a human,
and that you ship has an implicit
anthropology whether you want it or not.
what I mean by that is that you have
assumptions about your user, and how
your user operates can't verify purely by
watching them, and you can't verify purely
by having a science of how the electrons
and neurons are firing in their brain.
so there's this gap that is called
anthropology that you're shipping
whether you like it or not.
so do you wantâ¦
If you want to get that anthropology
right, it's probably helpful to
try on a bunch of different kinds
of anthropologies for your product
and see which one works the best.
This
Raul: How do you do that?
Taylor Black: my- this,
it's an interesting problem.
Part of it is I think codifying
different behavioral pathways that
you want the artificial intelligence
to augment or to circumvent when the
user is interacting with your product.
Part of it too, I think,
is understanding howâ¦
Is trying out different ways in which you
think that your user actually operates
from an intellectual standpoint, how
they actually make decisions, um, how
they, how you want this technology to
interact with their working styles,
with their working modalities.
And I think actually there's- we're
running into a bunch of different
problematic kind of ways in which we
interact with the technology we can flex.
We're very flexible as
human beings, right?
And we can have, we can flex
around it and mold ourselves to the
technologies rather than changing the
technologies to better fit ourselves.
But I think we lose something i- in, in,
in the interim there, where if you're
able to make a tool that fits better then
the person using it is going to have a
better experience in building with it and
developing with it or thinking with it.
and it's up to us to
design towards that end.
Perhaps a, a good example of
this is utilizing AI in the K
through 12 classroom, right?
Y- the, the design modalities currently
that I've experienced currently even
with options like learn mode in ChatGPT,
don't necessarily sustain the tension
of inquiry long enough to help the
student understand how to do that better.
And so what the student ends up
doing is using the tool as an oracle
to give answers that the student
doesn't want to figure out for
themselves Well, that's problematic.
The whole point of
Raul: Skill set atrophy.
Taylor Black: teaching is
developing that, right?
you want them to pass
the marshmallow test.
you want them to be able to,
Raul: Yeah
Taylor Black: suffer through a
little bit so that they can read
something like Ivanhoe, and then
make their way through a 400-page
book because of the awesomeness ofâ¦
that's there.
Or, think through the, a, a a
complicated geometric proof.
And I think we can get there, but
notice that the, those all end up
being design decisions around a
particular view of anthropology
that, that you may or may not hold.
And Iâ¦
And my anticipation is the closer you are
to, to reality, whatever you want that
to be from a s- secular point of view.
I have my own point of view as
to what reality is but from a
pure product standpoint, right?
The closer you are to reality of how
humans actually operate, and there
is o- a way, the better your products
are going to help that person flourish
Raul: Yeah.
And depending, e-e-even-- So there's
a lot to unpack here because you hit,
like, a really big philosophical point.
But even to the practical point of
even if the user is an agent and the
agent buys your product, it's still a
human that dictated, "Go do this thing.
Go buy these things or
tools that are necessary."
Because there are a lot of new products
that are just purely for agents.
Like the consumer or the end user is
an agent, but we have to understand
it's still a human at the back end
making these decisions deploying that.
But I f- I find it fascinating
too, so it's using the heuristic of
what kind of view of anthropology.
Because this is like the
beginning of, uh, all philosophy.
What is reality?
How do I talk about reality?
Based on that, how do-- What is
human, which is anthropology.
How do I treat human based on that?
So it's anthropology and ethics
because treating people is through a,
a product at that point, because that's
how you treat humans and depending
on who they are, like their stance
or their worth or their worthlessness
if they, you believe that to be true.
So
Taylor Black: But there's also an
interesting parallel to how you're-- how
you treat and agents as well, because
as we know, the, the literature proves
out that the same sort of psychological
tricks that expert negotiators or
con artists use also work on LLMs.
and so there's a certain amount that we
have mirrored the way in which humans
operate into the, the algorithmic,
biases of these systems such that they
can be gamed in very similar ways.
And it's there in the
scientific literature
Raul: That's crazy, isn't it?
Oh, that's wild.
As well as knowing that,
uh, you never know.
Gotta be nice to your
LL- LLM, your littleâ¦
Taylor Black: Yeah.
Raul: Oh, man
Taylor Black: and you can get
the answers you want that way.
I mean, there's some fascinating studies
done by some of my colleagues if you put
a line from a Psalm in with your prompt,
there's a measurable difference in how
morally aligned the responses can be.
is even true if you do something
like put an image of the
Annunciation in with your prompt.
The model will change its behavior in
a in a way that, that humans understand
as being morally codified in its
response in a direction or another.
Raul: That's interesting
Taylor Black: Yeah, it's fascinating.
Raul: I would always think, I mean,
this is the first time I'm making this
analogy, but I, I think there's, there's
the prompt engineering, but I think
there's also the, the approach of the
system that you've built around it.
'Cause I mean, you have to use
a harness to leverage the LLM.
The harness I think is like an engine.
Like it's aâ¦
I like to allu- like I like lis-
uh, watching the Bluc- Bruce McLaren
like come up story and how he built
his own engines like by hand, and
I thought it was pretty sweet.
So like the harness is the engine, the
LLM I think is the driver, so you can
swap out drivers whenever you want.
But there's also some technical components
around that 'cause you also have like the
wheels, the, uh, whatever the suspension,
the types of whatever mechanics.
Like I'm not obviously
an expert in it, butâ¦
And these are all the different
components, the, like the primitives,
the agents, the skill set, the memory
the, the hard rules, et cetera.
And what I'm seeing is that when
you set up that system at a certain
way, like your own engine, sometimes
the driver can be an okay driver,
and you still can go pretty fast.
Theâ¦
Or like i-i- in allu- in allu-
like alluding to like quality
of output or quality of thing.
But you're also saying that the way that
you actually treat, like, uh, the way the
driver thinks determines which is back to
square one or better for performed inside.
Taylor Black: yeah but it, like the
har- the harness is taking into account
the proclivities of the driver and
guiding them in certain directions,
Mean and so you're already assuming
something about the engine, the driver
itself, the LLM, so that you're able
to do a better job of directing it
Raul: Yeah, I think it's more important
to, to design the engine than to
hope on one LLM versus the other.
Taylor Black: Yes,
Raul: That's at least what I'm testing.
Taylor Black: Yeah.
Raul: fascinating.
Any cool projects that you're working on
right now, or any cool projects in the
incubators of the venture studios that are
exciting you that you're allowed to share?
Taylor Black: Yeah, certainly.
So I'm so I've been experimenting with
this concept that I call cold storage.
and there's this kind of perennial
problem in venture building where
it's a, I have a good idea, but I
don't have the right team, or I have
a good idea, but the, after a year of
working on it, the market isn't there.
in fact, I had a startup a number of
years ago that, that hit this really well.
It was it was an app that was using
near-field communication to create local,
local networks between cell phones.
So you could imagine being at a, and I'm
a huge electronic music fan, and so my
use case was I go to these festivals all
the time, and nobody has any coverage
because we blow the cell tower out or
we blow the, the router out because they
just can't handle that many connections.
if I just had a a low latency text
message capability that just used
everybody else's phones around me as
kind of a, a near-field communication
network to talk to my buddy on the other
side of the other side of the venue?
And it's a fantastic idea.
And there's-- I had a
lot of energy around it.
people thought it'd be great.
Ran into the problem that the near
field communication protocols were
just not where they needed to be.
And in fact, it's only in the
last two or three years that they
have gotten to where they need to
be for that sort of application.
So that experience and also,
others in that space is I
encountered this concept ofâ¦
I came up with this concept of what
I call cold storage of what are
the things that need to change in
order for that idea to move forward?
so rather than killing the idea, I put it
into cold storage and say, "These are the
things that would need to change in order
for this to be able to move forward."
Now, the problem with putting
something, an idea into cold storage
from a venture standpoint is that
the team disappears, the people who
remember the idea disappear, and
Raul: gone
Taylor Black: youâ¦
Exactly.
You go back to your projects that
you have from five years ago, and
you're like, "What even was this?"
but AI is a fantastic context engine.
Don't I give it all of the context
around the projects, reasons why it won't
move forward, and tell it to monitor
the reasons so that if the right team
member or person who has a PhD at X or
publishes a paper in Y, hits archive,
or the technology shift happens so that
near field communication can go, fifteen
feet between devices, or, the market
conditions change so that all of a sudden
a billion more people are doing electronic
music concerts or something like that.
Each of those ends up being a,
something that can be monitored
by my AI context engine to say,
"Hey, time to dust this idea off.
And by the way, here was the idea.
Here was all the meetings
Raul: contacts, yeah
Taylor Black: of the artifacts," every
all, the whole context, so that even
if I bring in a new person, they're
able to stick their head into the
context engine and be like, "Got it.
Let's now move that for- off the shelf."
this does is it gives you a lower
barrier to killing something because
you're able to revive it more quickly.
one of the hardest things in
venture is knowing, i- is being
able to kill your darlings.
And so being able to do that in a cold
storage standpoint and build out a, a
Raul: Becomes IP though, that
becomes IP that, that's stored
there that can be deployed.
So who owns it?
How do you store it?
That's an interesting product
Taylor Black: And there's great
ways of doing that in fractional
s-standpoints too that AI is
really good at parsing out andâ¦
so I'm ex- I'm super excited and
jazzed about that that concept
because it allows for more fluid
collaboration between venture studios
Raul: It'd be pretty interesting if that
was even-- I don't know if it would be
open source, but it would be interesting
because a lot of idea people but not
the right person to run the company.
But they're brilliant at hitting
something PMF in the beginning, but
they just suck at running the team.
So they can put that there, then someone
who is good at running a team can just
take that on, and it's almost like an
exchange of like a real incubation center
Taylor Black: 100%.
And I, I've operated this in two, i-
in prototype ways in two different
ecosystems, and it's been really
interesting to see the effects.
Yeah, so I'm excited to push that forward
more have it impact the real world.
So yeah
Raul: I like it.
Taylor, where can people go to thank
you for being on, learn more about
your writing, and connect with you?
Taylor Black: My Substack is great.
Innovate.pourbrew.me.
That, that's a great place toâ¦
And also LinkedIn.
I share a lot on LinkedIn, not only
because I work at Microsoft and the
CTO of Microsoft is a LinkedIn founder.
But I j- I just like the
conversations that happen there
Raul: Put those links in the show notes.
Taylor, thanks again for being on
Taylor Black: Dude, thanks for having me.
Appreciate it
Thank you for listening to the podcast.
If you'd like to get more of
these podcasts, please go ahead
and subscribe to wherever you're
watching or listening to the pod.
If you have any suggestions or questions
that you want me to answer on the
pod, or you have a guest that you want
me to interview, please go ahead and
email me all of your suggestions and
questions at podcast at do good work.
io that's podcast at do good work.
io.
If you'd like to give me public
feedback, you can go on Apple
Podcasts, and from there, you
can leave up to a 5 star review.
I would greatly appreciate that.
If you'd like more free resources where
you can get my best strategies to help you
increase your company's performance and
scale profitably, where you can get hand
picked articles to propel your growth, and
you can get trainings and discussions that
I give freely online, as well as low cost
resources, Such as my books and guides.
You can get all of that
by going to dogoodwork.
io forward slash resources 90 percent
of all these resources are not behind
any email opt ins you can get instant
access by going to dogoodwork.
io forward slash resources Now, if
you'd like to accelerate your progress
and shorten the gap between information
and action and start seeing results
in your business, let's work together
to increase your company's performance
and scale profitably and serve
more clients without the overwhelm.
You can request a free clarity call
to see how we can best support you to
reach your goals by going to DoGoodWork.
io forward slash apply.
Again, that's DoGoodWork.
io forward slash apply.
As always, it's an honor to be
a small part in your journey.
This is Raul Hernandez.
Do good work.