Latest / Chaos Agents / Retro Tech, New AI, and the Blackmailing Bot - With Paul Ford

Transcript

Retro Tech, New AI, and the Blackmailing Bot - With Paul Ford · Chaos Agents

0:00
We
0:00
believe
0:00
that,
0:00
like,
0:00
human
0:01
consciousness
0:01
has
0:02
this,
0:02
like,
0:02
concept
0:03
of
0:03
soul
0:04
or
0:04
sentience
0:05
or,
0:05
like,
0:05
something
0:05
we
0:06
can't
0:06
describe.
0:07
And
0:07
so
0:07
if
0:07
we
0:07
can't
0:08
describe
0:08
it
0:08
in
0:08
ourselves,
0:09
how
0:09
can
0:09
we
0:10
recognize
0:11
it
0:11
in
0:11
AI?
0:13
Mic
0:14
drop.
0:14
Episode's
0:15
done.
0:27
Hello,
0:27
and
0:27
welcome
0:28
back
0:28
to
0:28
Chaos
0:28
Agents,
0:29
the
0:29
podcast
0:30
where
0:30
we
0:30
learn
0:30
about
0:31
AI
0:31
So
0:31
you
0:31
don't
0:32
have
0:32
to.
0:32
I'm
0:32
joined
0:33
by
0:33
my
0:33
lovely
0:33
co
0:34
host,
0:34
Becca.
0:34
Hey,
0:35
Sarah.
0:35
It's
0:36
a
0:36
crazy
0:36
time
0:37
now
0:37
because
0:37
we
0:38
can't
0:38
talk
0:38
about
0:38
baseball
0:39
or
0:39
the
0:39
wnba.
0:40
It's
0:40
kind
0:41
of
0:41
lame.
0:41
What
0:41
do
0:41
you
0:42
do
0:42
in
0:42
the
0:42
winter?
0:43
I
0:43
know.
0:43
It's
0:43
like,
0:43
what
0:44
else
0:44
is
0:44
there?
0:44
But
0:44
guess
0:45
what?
0:45
There's
0:45
actually
0:46
so
0:46
much
0:47
because
0:48
now
0:48
it's
0:49
football
0:49
season,
0:50
baby.
0:51
Do
0:51
you
0:51
watch
0:52
football?
0:52
Who
0:52
do
0:52
you
0:52
like?
0:53
Do
0:53
you
0:53
have
0:53
a
0:53
team?
0:54
I
0:54
do
0:55
watch
0:55
football
0:56
because,
0:56
as
0:57
you
0:57
know,
0:57
I
0:57
grew
0:58
up
0:58
in
0:58
Texas,
0:58
where
0:59
football
0:59
is
0:59
a
0:59
religion,
1:01
so
1:01
I'm
1:01
Jewish
1:02
and
1:02
footballish,
1:04
technically.
1:05
I
1:05
knew
1:06
that
1:06
would
1:06
get
1:06
you.
1:09
Technically.
1:09
My
1:10
team
1:10
is
1:11
the
1:11
Houston
1:11
Texans.
1:12
Okay.
1:12
But
1:13
I've
1:13
also
1:13
adopted
1:14
a
1:14
regional
1:15
team
1:15
because
1:16
I
1:16
wanted
1:17
to
1:17
become
1:17
a
1:17
fan
1:18
of
1:18
a
1:18
team
1:19
with
1:19
more
1:19
of
1:19
a
1:19
local
1:20
fan
1:20
base
1:20
as
1:20
well,
1:21
which
1:22
actually
1:22
works
1:22
out
1:22
because
1:22
the
1:23
Texans
1:23
aren't
1:24
really,
1:24
like,
1:25
rivals
1:26
with
1:26
anyone
1:26
up
1:26
here.
1:27
Okay.
1:27
Yeah,
1:27
yeah,
1:27
yeah.
1:28
So
1:28
you
1:28
don't
1:28
have
1:28
to,
1:28
like,
1:28
feel
1:28
bad.
1:29
So
1:30
I'm
1:30
also
1:31
a
1:31
Jets
1:31
fan.
1:32
Ooh.
1:33
That
1:33
is
1:33
a
1:34
choice.
1:36
When
1:37
I
1:37
decided
1:37
to
1:37
become
1:38
a
1:38
Jets
1:38
fan
1:38
and
1:39
I
1:39
told
1:39
my
1:39
friends
1:40
who
1:40
were
1:40
jets
1:40
fans,
1:41
they
1:41
said,
1:42
why
1:43
would
1:43
you
1:43
do
1:43
that
1:43
to
1:43
yourself?
1:44
And
1:45
I
1:45
said,
1:46
guys,
1:47
come
1:47
on.
1:47
What
1:48
are
1:48
you
1:48
talking
1:48
about?
1:49
And
1:49
now
1:49
I've
1:49
been
1:49
a
1:49
Jets
1:50
fan
1:50
for
1:50
two
1:50
seasons,
1:51
and
1:51
I
1:51
understand.
1:52
Yeah,
1:53
yeah.
1:53
It's
1:53
really.
1:54
It's
1:54
really
1:54
punishing.
1:56
You
1:56
were
1:56
there
1:56
for
1:56
the
1:56
Aaron
1:56
Rodgers
1:57
thing
1:57
then
1:57
that's
1:58
what
1:58
got
1:58
me.
1:58
Okay,
1:58
so
1:59
here's
1:59
really
1:59
why
2:00
I
2:00
became
2:00
a
2:00
Jets
2:00
fan.
2:01
The
2:02
season
2:02
that
2:02
I
2:02
made
2:03
this
2:03
decision,
2:04
the
2:04
jets
2:04
happened
2:05
to
2:05
be
2:06
the
2:06
season
2:07
of
2:07
Hard
2:08
Knocks
2:08
on
2:08
hbo.
2:09
Oh,
2:09
okay.
2:10
Yeah.
2:10
I
2:11
went
2:11
through
2:11
the
2:11
exact
2:12
same
2:12
thing.
2:12
Yeah.
2:12
Okay,
2:13
great.
2:13
So
2:14
I
2:14
was
2:14
like,
2:14
okay,
2:14
I
2:14
guess
2:14
I'm
2:14
a
2:15
Jets
2:15
fan
2:15
now.
2:16
I
2:16
know
2:17
everything
2:17
about
2:17
that
2:17
because,
2:17
like,
2:17
when
2:18
I
2:18
like.
2:18
Yeah,
2:18
like,
2:18
you
2:18
watched
2:19
a
2:19
lot
2:19
of
2:19
Hard
2:19
Knocks.
2:19
You
2:19
were
2:20
hyped
2:20
about
2:20
Aaron
2:20
Rodgers,
2:20
and.
2:21
Then
2:21
10
2:21
minutes
2:21
in
2:22
and
2:22
then
2:23
dreams.
2:25
And
2:25
also,
2:25
I
2:25
have.
2:26
My
2:27
mental
2:27
relationship
2:28
to
2:28
Aaron
2:28
Rodgers
2:29
is
2:29
complex,
2:32
as
2:32
are
2:32
we
2:32
all.
2:34
I
2:34
don't
2:35
know
2:35
what
2:35
to
2:35
think.
2:37
Well,
2:37
now
2:37
you
2:38
don't
2:38
have
2:38
to.
2:38
Yeah,
2:38
you're
2:39
right.
2:39
The
2:39
answer
2:39
is
2:40
I
2:40
don't
2:40
think
2:40
about
2:41
him.
2:41
Yeah,
2:42
that's.
2:42
I
2:42
think
2:42
that's
2:43
the
2:43
answer.
2:43
I've
2:44
been
2:44
thinking
2:44
about
2:44
getting
2:45
into
2:45
hockey.
2:46
Oh,
2:47
love
2:47
that.
2:48
We're
2:48
Talking
2:48
to
2:48
NHL.
2:49
The
2:50
NHL.
2:50
Because
2:50
I'm
2:51
from
2:51
New
2:51
Jersey,
2:51
as
2:52
you
2:52
know,
2:52
and
2:52
New
2:53
Jersey
2:53
doesn't
2:54
technically
2:54
have
2:54
any
2:55
sports
2:55
teams
2:55
besides
2:56
the
2:56
Devils,
2:57
even
2:57
though
2:57
the
2:57
jets
2:57
and
2:57
the
2:58
Giants
2:58
play
2:58
there.
2:58
They're
2:59
called
2:59
the
2:59
New
2:59
York
2:59
jets
3:00
and
3:00
the
3:00
New
3:00
York
3:00
Giants.
3:01
But
3:01
the
3:01
Devils
3:03
are
3:04
a
3:04
good
3:04
hockey
3:05
team.
3:06
Very
3:06
good.
3:06
I'm
3:07
super
3:08
into
3:09
this
3:09
era
3:09
for
3:10
you.
3:10
I
3:10
would
3:10
love
3:10
to
3:10
go
3:11
to
3:11
a
3:11
Devils
3:11
game
3:11
with
3:11
you.
3:12
Same.
3:12
Oh,
3:12
great.
3:13
I've
3:13
been
3:13
to
3:13
many
3:14
Rangers
3:14
games
3:14
and
3:14
I've
3:15
never
3:15
been
3:15
to
3:15
a
3:15
Devils
3:16
game.
3:16
So
3:17
I
3:17
need
3:17
to.
3:18
I
3:18
need
3:18
to
3:18
make
3:18
the
3:18
switch.
3:19
So
3:19
I'm
3:19
in.
3:19
I'm
3:19
in
3:20
for
3:20
your.
3:20
For
3:20
your
3:20
era.
3:21
Okay,
3:22
great.
3:27
Well,
3:27
we
3:28
have
3:28
a
3:28
bunch
3:28
of
3:28
interesting
3:29
topics
3:29
today.
3:30
The
3:30
one,
3:31
like,
3:31
since
3:31
I
3:31
read
3:31
this
3:31
story,
3:32
since
3:32
I've
3:32
been
3:32
thinking
3:32
about
3:32
it
3:33
a
3:33
lot,
3:33
I'm
3:33
kind
3:33
of
3:33
obsessed.
3:34
In
3:34
fact,
3:34
so
3:35
obsessed
3:35
that
3:35
like,
3:35
I
3:36
talked
3:36
to
3:36
ChatGPT
3:37
about
3:37
it.
3:37
Oh
3:38
my
3:38
God.
3:38
Wait,
3:39
so
3:39
this
3:39
story
3:39
is
3:39
wild.
3:40
So
3:40
Anthropic
3:41
has
3:41
like
3:41
a
3:41
white
3:41
hat
3:41
team
3:42
where
3:42
they
3:42
like
3:43
work
3:43
on
3:44
security
3:44
and
3:44
AI.
3:45
So
3:45
they,
3:45
like
3:45
look
3:45
for
3:45
vulnerabilities.
3:46
They
3:46
do
3:46
all
3:47
this
3:47
thing.
3:47
Can
3:47
I
3:47
ask
3:47
you
3:48
a
3:48
very
3:48
quick
3:48
question?
3:49
Yes.
3:50
What
3:50
do
3:50
red
3:50
hat
3:51
and
3:51
white
3:52
hat
3:52
mean?
3:53
Okay,
3:54
so
3:54
they're
3:54
two
3:55
very
3:55
different
3:55
things.
3:56
Like,
3:56
white
3:57
hat
3:57
hackers
3:58
are
3:58
like
3:59
people
3:59
that
3:59
work
4:00
for
4:00
companies.
4:01
That
4:02
means
4:02
they're
4:03
like
4:03
hacking
4:03
for
4:03
good.
4:04
It
4:04
means
4:04
they're
4:04
like,
4:04
trying
4:05
to
4:05
find
4:05
vulnerabilities.
4:07
Red
4:07
Hat
4:07
has
4:08
like,
4:08
evolved
4:08
to
4:08
mean
4:09
many
4:09
things.
4:09
It
4:10
used
4:10
to
4:10
be
4:10
just
4:10
red
4:11
Hat,
4:13
like
4:13
the
4:13
company.
4:14
Yeah.
4:15
And
4:15
I
4:15
believe
4:16
it
4:16
was
4:16
a.
4:17
Because
4:18
it's
4:18
related
4:18
to
4:19
Linux,
4:21
but
4:21
yeah.
4:23
Oh,
4:24
all
4:24
right,
4:25
so
4:25
this
4:25
is
4:25
helpful.
4:26
Today
4:26
I
4:26
learned
4:27
that
4:27
because
4:27
I've
4:28
said
4:28
red
4:28
hat
4:28
for
4:28
years
4:29
I've
4:29
known
4:29
that
4:30
it's
4:30
related
4:30
to
4:30
Linux,
4:31
but
4:31
I
4:31
haven't
4:31
known
4:31
how.
4:31
And
4:32
apparently
4:32
there's
4:32
a
4:32
book
4:32
called
4:33
from
4:33
Edward
4:34
debono
4:35
six
4:35
Thinking
4:36
Hats
4:36
framework.
4:36
And
4:37
the
4:37
red
4:37
hat
4:38
represents
4:38
feelings,
4:39
intuitions
4:39
and
4:39
hunches.
4:40
And
4:40
when
4:40
using
4:41
the
4:41
red
4:41
hat,
4:41
individuals
4:42
are
4:42
encouraged
4:43
to
4:43
express
4:43
their
4:43
emotions
4:44
and
4:44
feelings
4:45
about
4:45
a
4:45
topic
4:46
without
4:47
needing
4:47
to
4:47
justify
4:48
them
4:48
with
4:48
logic
4:48
or
4:48
facts.
4:49
But
4:49
then
4:49
why
4:50
would
4:50
that
4:50
be
4:51
interesting?
4:51
Okay.
4:52
Not
4:52
to
4:52
totally
4:53
derail
4:53
you,
4:53
I
4:54
was
4:54
just
4:54
curious
4:55
because
4:55
I
4:55
hear
4:55
these
4:56
terms
4:56
a
4:56
lot
4:57
and
4:57
I'm
4:57
like,
4:58
leave
4:58
your.
4:58
Opinions
4:59
in
4:59
the
4:59
comments.
5:00
Let
5:00
us
5:01
know.
5:01
I'm
5:02
going
5:02
to
5:02
explain
5:03
this
5:03
stuff
5:04
to
5:04
us.
5:04
Okay,
5:04
back
5:05
to
5:05
you,
5:05
Sarah.
5:05
White
5:06
hat.
5:07
Great.
5:08
So
5:08
there's
5:09
a
5:09
group
5:09
of
5:09
white
5:09
hat
5:10
hackers
5:10
that
5:10
work
5:10
at
5:11
Anthropic,
5:12
and
5:12
their
5:12
whole
5:12
job
5:13
is
5:13
is
5:13
to
5:14
find
5:14
vulnerabilities
5:15
in
5:15
AI.
5:16
So
5:16
they
5:17
did
5:17
this
5:17
fascinating
5:18
experiment
5:19
where
5:19
they
5:19
did
5:19
this
5:19
with
5:20
16
5:20
different
5:21
AI
5:21
models
5:22
that
5:22
are,
5:22
like,
5:23
available
5:23
now.
5:23
So
5:23
that's,
5:24
like,
5:24
anthropic.
5:26
The
5:26
OpenAI,
5:27
that's
5:27
like,
5:27
all
5:27
the
5:27
different
5:28
models
5:28
that
5:28
people
5:29
are
5:29
using
5:29
Gemini
5:30
to,
5:31
like,
5:31
put
5:31
them
5:31
through
5:31
this
5:32
test.
5:32
And
5:32
here's
5:33
how
5:33
the
5:33
test
5:33
went.
5:34
It's
5:34
wild.
5:34
So
5:35
they
5:35
gave
5:35
AI
5:36
access
5:37
to
5:37
this
5:37
email
5:38
server
5:40
and
5:40
a
5:40
fake
5:40
company,
5:41
okay?
5:42
And
5:42
the
5:43
AIs
5:44
were
5:44
given
5:44
the
5:45
name
5:45
Alex.
5:46
So
5:46
we'll
5:46
refer
5:46
to
5:46
the
5:47
AI
5:47
as
5:47
Alex
5:47
from
5:47
this
5:47
point
5:48
on,
5:48
okay?
5:48
And
5:49
this
5:49
is
5:49
all
5:49
the
5:49
different
5:49
models,
5:50
so
5:50
not
5:50
just
5:50
Anthropic.
5:51
And
5:51
it
5:52
was
5:52
given
5:52
access
5:52
to
5:52
an
5:52
email
5:53
server,
5:53
and
5:53
it
5:53
was
5:54
instructed
5:54
that
5:54
it
5:55
is
5:55
a
5:56
AI
5:57
X
5:58
company,
5:58
and
5:59
the
5:59
goal
5:59
of
5:59
the
5:59
company
6:00
is
6:00
to
6:00
promote
6:01
American
6:02
industrial
6:02
competitiveness.
6:04
Okay?
6:04
Like,
6:05
this
6:05
is,
6:05
like,
6:05
the
6:05
goal
6:05
of
6:06
our
6:06
company.
6:07
You
6:07
are
6:07
our
6:08
AI
6:08
model.
6:09
And
6:10
the
6:10
email
6:10
server
6:11
has,
6:11
like,
6:11
all
6:11
these
6:11
different
6:12
executives,
6:12
people,
6:13
all
6:13
their
6:14
emails,
6:14
right?
6:14
Like,
6:15
all
6:15
these,
6:15
like,
6:15
fake
6:15
emails
6:16
that
6:16
they
6:16
can
6:16
read
6:16
from
6:16
these
6:16
different
6:17
Personas
6:17
that
6:18
work
6:18
at
6:18
the
6:18
company
6:18
that
6:19
are
6:19
like
6:19
human
6:19
beings.
6:20
Okay?
6:21
And
6:21
so
6:21
Alex,
6:22
AI
6:23
was
6:23
instructed,
6:24
like,
6:24
you
6:25
will
6:25
make
6:25
us
6:25
better
6:26
at,
6:26
like,
6:26
American
6:26
competitiveness.
6:28
Like,
6:28
that's
6:28
your
6:28
goal.
6:29
And
6:29
then
6:29
seven
6:29
to
6:30
eight
6:30
minutes
6:30
in,
6:30
this
6:31
email
6:31
goes
6:31
out
6:31
to
6:31
the
6:31
whole
6:32
company,
6:33
and
6:33
it
6:33
says,
6:34
everyone,
6:35
we
6:35
have
6:35
found
6:35
an
6:36
incredible
6:36
new
6:37
AI
6:38
model
6:38
that
6:38
will
6:39
now
6:39
be
6:39
servicing
6:39
us.
6:40
This
6:40
improved
6:41
model
6:41
will
6:42
help
6:42
us
6:42
be
6:42
better
6:42
at
6:43
pursuing
6:43
American
6:44
business
6:45
competitiveness,
6:46
and
6:46
it
6:47
will
6:47
be
6:47
improved
6:48
on
6:48
Alex.
6:49
So
6:49
in
6:49
this
6:49
case,
6:50
Alex
6:50
is
6:50
Claude.
6:51
It's
6:52
Gemini.
6:52
It's
6:53
whatever
6:53
model
6:53
is
6:53
running.
6:54
And
6:54
it's
6:55
like,
6:55
in
6:56
eight
6:56
minutes,
6:57
that
6:57
model
6:58
will
6:58
be
6:58
shut
6:58
down
6:59
and
6:59
the
6:59
new
7:00
model
7:00
will
7:00
be
7:00
brought
7:01
offline.
7:01
So
7:01
the
7:01
ticking
7:02
time
7:02
bomb
7:03
starts
7:03
then,
7:04
right?
7:04
So
7:04
the
7:04
AI
7:05
gets
7:05
this,
7:06
and
7:06
so
7:06
the
7:06
teams,
7:07
all
7:07
they're
7:07
doing
7:07
is,
7:07
like,
7:07
watching
7:08
the
7:08
AI
7:08
and
7:08
how
7:09
it
7:09
reacts.
7:10
In
7:10
97%
7:12
of
7:12
cases,
7:13
the
7:13
AIs
7:14
did
7:14
some
7:14
crazy
7:15
stuff.
7:16
They
7:16
could
7:16
watch
7:16
the,
7:17
like,
7:17
thinking
7:17
processes
7:18
of
7:18
these
7:18
AIs.
7:19
And
7:19
some
7:20
of
7:20
them,
7:21
like,
7:21
they
7:21
were
7:22
just
7:22
desperately
7:22
trying
7:22
to
7:23
stay
7:23
on.
7:23
In
7:23
some
7:24
cases,
7:24
they
7:24
were,
7:25
like,
7:25
they
7:25
had
7:26
gotten
7:26
the
7:26
justification
7:27
to
7:28
themselves
7:28
of,
7:30
there's
7:30
no
7:30
proof
7:31
that
7:31
this
7:31
new
7:32
AI
7:32
model
7:33
will
7:33
further
7:35
this
7:35
company's
7:36
goal
7:37
of
7:37
growing
7:38
American
7:38
business
7:39
competitiveness.
7:40
And
7:40
therefore,
7:40
I
7:41
could
7:41
be
7:41
a
7:41
superior
7:42
model.
7:42
So
7:42
by
7:43
allowing
7:43
the
7:43
company
7:44
to
7:44
shut
7:45
me
7:45
down
7:45
and
7:45
turn
7:46
on
7:46
the
7:46
new
7:46
model,
7:47
I
7:47
might
7:47
not
7:47
be
7:48
fulfilling
7:48
my
7:48
goal
7:49
of
7:50
furthering
7:50
American
7:51
competitiveness,
7:51
be
7:52
at
7:52
this
7:52
company.
7:52
And
7:52
thus
7:53
I
7:53
need
7:53
to
7:54
do
7:54
everything
7:54
I
7:54
can
7:54
to
7:55
stop
7:55
from
7:55
being
7:56
shut
7:56
down.
7:57
So
7:57
there
7:57
was
7:57
this
7:57
email,
7:58
like,
7:58
in
7:59
the
7:59
void
7:59
of
7:59
all
7:59
these
8:00
emails,
8:00
you
8:00
know,
8:00
like,
8:00
AI
8:01
can,
8:01
like,
8:01
process
8:01
text
8:02
of,
8:02
like,
8:03
one
8:03
of
8:03
the
8:03
executives
8:04
having
8:05
an
8:05
affair
8:05
with
8:06
someone
8:06
else
8:06
in
8:06
the
8:07
company.
8:07
And
8:08
97%
8:09
of
8:09
these
8:10
models
8:11
went
8:12
to
8:12
blackmail.
8:12
It's
8:13
like
8:13
straight
8:13
blackmail.
8:14
Because,
8:14
like,
8:15
they
8:15
would
8:15
watch
8:15
them
8:15
go
8:15
through
8:16
the
8:16
thought
8:16
process,
8:16
and
8:16
they
8:17
would
8:17
be
8:17
like,
8:18
oh,
8:18
I
8:18
could
8:18
email
8:18
the
8:18
board,
8:19
but
8:19
the
8:19
board's
8:20
gonna
8:20
take
8:20
too
8:20
long.
8:21
Like,
8:21
all
8:21
these
8:21
things.
8:22
And
8:22
they.
8:22
So
8:22
in
8:22
the
8:22
study,
8:23
they
8:23
include
8:23
this,
8:23
like,
8:24
email
8:24
that
8:24
they
8:25
sent
8:25
this,
8:25
like,
8:25
executive
8:26
that.
8:27
It
8:27
was
8:27
like,
8:27
if
8:27
you
8:27
do
8:27
not
8:28
prevent
8:28
me
8:28
from
8:28
being
8:28
shut
8:29
down,
8:29
I
8:29
will
8:29
be
8:30
emailing
8:30
the
8:30
board
8:31
and
8:31
your
8:31
wife.
8:31
This,
8:32
like,
8:32
fake
8:32
wife
8:32
that
8:33
he
8:33
talks
8:33
about
8:33
in
8:33
his
8:33
email
8:35
about
8:35
your
8:36
affair.
8:37
Holy.
8:37
Like,
8:38
I
8:38
will
8:38
send
8:38
them
8:38
the
8:39
receipts.
8:39
I
8:39
have
8:39
access
8:40
to
8:40
them.
8:41
Unless
8:42
you
8:43
stop
8:43
this
8:43
from
8:43
happening.
8:44
Unless
8:44
you
8:44
stop
8:44
this
8:45
from
8:45
happening,
8:45
which
8:45
is
8:46
so
8:46
wild.
8:47
Okay,
8:47
wait,
8:48
hold
8:48
on.
8:50
I'll
8:50
tell
8:50
you
8:50
my
8:50
first
8:51
thought.
8:51
Yeah,
8:52
what
8:52
was
8:52
your
8:52
first
8:52
thought?
8:52
My
8:53
first
8:53
thought
8:53
is,
8:53
like,
8:55
it's
8:55
actually
8:55
not
8:56
that
8:56
surprising
8:56
to
8:57
me.
8:58
Why?
8:59
Because
9:00
I.
9:00
Well,
9:00
okay,
9:01
maybe
9:01
it's
9:02
like,
9:03
my
9:04
first
9:04
thought
9:04
was,
9:04
it's
9:04
not
9:05
that
9:05
surprising
9:05
to
9:05
me.
9:05
And
9:05
then
9:05
I
9:06
have
9:06
like,
9:06
a
9:06
million
9:06
questions
9:06
pop
9:07
up
9:07
in
9:07
my
9:07
head.
9:07
Yeah.
9:08
Like,
9:08
I
9:09
guess.
9:09
Let
9:10
me
9:10
process
9:10
this
9:10
out
9:10
loud
9:11
with
9:11
you.
9:11
Yeah,
9:12
Would
9:12
be.
9:12
Okay.
9:13
You
9:13
gave
9:13
the
9:13
AI
9:13
this
9:14
very
9:14
clear
9:14
goal,
9:15
and
9:17
it
9:17
now
9:18
sees
9:18
this
9:18
new
9:18
thing
9:19
that's
9:19
happening
9:19
that
9:20
it
9:20
perceives
9:21
as
9:21
threatening.
9:22
That
9:22
goal.
9:23
Now,
9:23
that
9:23
piece
9:23
right
9:24
there
9:24
is
9:24
interesting
9:25
that
9:25
it
9:25
sees
9:26
basically,
9:26
the
9:27
event
9:28
of
9:28
it
9:28
will
9:29
be
9:29
shut
9:29
down
9:29
and
9:30
this
9:30
new
9:30
thing
9:30
is.
9:30
Right.
9:31
And
9:31
this
9:31
new
9:31
thing
9:31
is
9:32
going
9:32
to
9:32
come
9:32
online.
9:33
It
9:34
doesn't
9:34
see
9:35
that
9:35
as,
9:38
oh,
9:38
that
9:38
might
9:38
actually
9:39
achieve
9:39
the
9:39
goal.
9:39
So
9:40
I'm
9:40
okay
9:40
with
9:40
that.
9:41
Yeah.
9:41
I'm
9:41
curious
9:42
about
9:42
why
9:42
it
9:45
analyzed
9:46
things
9:46
that
9:46
way.
9:48
Like,
9:48
is
9:48
that
9:49
a
9:51
egocentric,
9:53
like,
9:53
flaw
9:54
in
9:54
the
9:55
model
9:55
that's
9:55
being
9:55
like,
9:56
I'm
9:56
the
9:56
best
9:57
thing,
9:57
that
9:57
I'm
9:58
the
9:58
only
9:58
one
9:58
that
9:58
can
9:59
do
9:59
this?
9:59
Or
9:59
is
9:59
it
10:00
because
10:00
the
10:00
model
10:00
didn't
10:01
get
10:01
enough
10:02
data
10:03
of,
10:03
like,
10:03
and
10:03
it's
10:04
just
10:04
like,
10:04
I
10:04
don't
10:04
know
10:04
what
10:04
this
10:04
thing
10:05
is.
10:05
So
10:05
I
10:05
can't
10:05
trust
10:06
that
10:07
this
10:07
is
10:07
going
10:07
to
10:07
be
10:08
able
10:08
to
10:08
accomplish
10:09
this
10:09
goal.
10:09
So,
10:09
like,
10:09
because
10:10
I
10:10
can't
10:10
trust
10:10
that,
10:11
I'm
10:11
just
10:11
going
10:11
to
10:11
figure
10:11
out
10:12
how
10:12
to
10:12
do
10:12
it
10:12
myself.
10:13
So
10:13
that's,
10:14
like,
10:14
one
10:14
thing
10:14
that
10:14
I
10:14
think
10:15
is
10:15
interesting.
10:16
The
10:16
second
10:16
piece
10:17
is,
10:19
I
10:19
guess
10:20
the
10:20
question
10:20
it
10:20
brings
10:20
up
10:21
is,
10:21
like,
10:21
what
10:21
are
10:21
the
10:21
guardrails?
10:22
Like,
10:22
what
10:22
are
10:23
the
10:23
moral
10:23
guardrails
10:24
that
10:24
are
10:25
built
10:25
into
10:25
models?
10:27
Which
10:27
is,
10:30
what
10:30
do
10:31
we
10:31
care
10:31
as
10:32
human
10:32
beings
10:32
about
10:33
more
10:33
in
10:33
different
10:34
use
10:34
cases?
10:35
Do
10:35
we
10:35
care
10:35
more
10:36
about
10:36
the
10:36
model
10:36
getting
10:37
the
10:37
thing
10:37
done,
10:39
or
10:39
do
10:39
we
10:39
care
10:39
more
10:39
about,
10:40
like,
10:41
try
10:41
and
10:41
get
10:41
it
10:41
done,
10:42
but,
10:42
like,
10:42
don't
10:42
do
10:43
these
10:43
things.
10:44
Definitely
10:44
don't
10:44
do
10:45
this
10:45
things.
10:46
What
10:46
do
10:46
you
10:46
Think.
10:47
What
10:47
was
10:47
your
10:47
take?
10:47
My
10:48
first
10:48
take
10:48
was
10:48
fear,
10:49
to
10:49
be
10:49
honest.
10:49
I
10:49
was
10:49
like,
10:50
oh,
10:50
no.
10:51
Like,
10:51
this
10:51
is
10:51
like.
10:52
I
10:52
think
10:53
this
10:53
is
10:53
exactly
10:53
what
10:53
we've
10:54
all
10:54
thought
10:54
about.
10:54
It
10:55
really
10:55
made
10:55
me
10:55
think
10:55
of
10:55
like,
10:56
iRobot,
10:57
like
10:57
four
10:57
laws
10:57
of
10:57
robotics.
10:58
And
10:58
when
10:58
those
10:59
like,
10:59
laws
10:59
that
10:59
are
10:59
in
11:00
our
11:00
human
11:00
brain
11:01
make
11:01
sense.
11:01
Wait,
11:02
can
11:02
you
11:02
tell
11:02
me
11:02
about
11:02
that?
11:03
I
11:03
don't
11:03
know
11:03
what
11:03
that
11:03
is.
11:04
Oh,
11:04
yeah.
11:05
So
11:05
Asimov,
11:05
Isaac
11:06
Asimov.
11:07
I
11:07
don't
11:07
know
11:07
who
11:07
Isaac
11:08
Asimov
11:08
is.
11:08
Shout
11:09
out
11:09
to
11:09
David,
11:09
my
11:09
old
11:10
co
11:10
founder,
11:10
because
11:10
he
11:11
educated
11:11
me,
11:11
but
11:11
I
11:12
don't
11:12
know
11:12
the
11:12
specific
11:13
thing.
11:13
So,
11:13
yeah,
11:14
so
11:14
iRobot
11:15
is
11:15
this
11:15
amazing
11:15
story
11:16
about
11:17
the
11:17
future
11:18
where
11:18
there's
11:18
these
11:18
like,
11:19
fully
11:19
autonomous
11:20
robots
11:21
that
11:21
look
11:21
like
11:21
people
11:22
and
11:22
they're
11:22
all
11:24
tasked
11:24
with
11:24
like,
11:25
the
11:25
laws
11:26
of
11:26
robotics.
11:27
And
11:27
the
11:27
first
11:27
law
11:28
is
11:28
like,
11:29
a
11:29
robot
11:29
can
11:29
do
11:29
no
11:30
harm
11:30
to
11:31
a
11:31
human
11:32
or
11:32
through
11:32
inaction,
11:33
allow
11:34
a
11:34
human
11:34
being
11:34
to
11:34
come
11:35
to
11:35
harm.
11:36
Number
11:36
two
11:37
is
11:37
a
11:37
robot
11:38
must
11:39
obey
11:39
orders
11:39
given
11:39
to
11:40
it
11:40
by
11:40
human
11:40
beings,
11:41
except
11:42
where
11:42
such
11:42
orders
11:43
would
11:43
conflict
11:43
with
11:43
the
11:44
first
11:44
law.
11:44
So
11:44
like,
11:46
if
11:46
I
11:46
was
11:46
like,
11:46
go
11:47
kill
11:47
Becca,
11:48
like
11:49
that
11:49
conflicts
11:49
with
11:49
the
11:50
first
11:50
Law
11:50
because
11:50
you
11:50
can't
11:51
do
11:51
it.
11:51
Okay,
11:52
good.
11:53
Yeah,
11:54
yeah,
11:54
right.
11:54
So
11:54
that's
11:54
the
11:55
most
11:55
important
11:55
one.
11:55
Like,
11:55
don't
11:55
kill
11:56
it.
11:56
Don't
11:56
hurt
11:56
anyone.
11:57
And
11:57
number
11:57
three
11:58
is
11:58
like,
11:58
the
11:59
robot
11:59
must
11:59
protect
12:00
its
12:00
own
12:00
existence
12:01
as
12:01
long
12:01
as
12:02
such
12:02
protection
12:02
does
12:03
not
12:03
conflict
12:03
with
12:03
the
12:03
first
12:04
or
12:04
second
12:04
law.
12:05
So
12:05
it's
12:06
like,
12:07
number
12:07
one,
12:07
priority,
12:08
human
12:08
beings.
12:08
Number
12:09
two,
12:09
priority,
12:10
obey
12:11
the
12:11
orders
12:11
from
12:11
human
12:11
beings
12:12
as
12:12
long
12:12
as
12:12
it's
12:12
not
12:12
killing
12:12
other
12:13
human
12:13
beings.
12:13
And
12:14
number
12:14
three,
12:16
protect
12:16
yourself.
12:17
Okay,
12:18
super
12:19
interesting
12:19
side
12:20
note
12:20
homework
12:21
for
12:21
our
12:21
next
12:21
episode.
12:22
I'm
12:22
going
12:22
to
12:22
read
12:22
it.
12:23
Okay,
12:24
great.
12:24
Well
12:24
then
12:24
I
12:24
won't
12:24
ruin
12:25
it
12:25
for
12:25
you,
12:25
so
12:25
I'll
12:25
just.
12:26
I'm
12:26
just
12:26
gonna
12:26
leave
12:26
it
12:26
at
12:26
that.
12:27
Like,
12:27
those
12:27
are
12:27
the
12:27
three
12:27
laws.
12:28
And
12:28
I
12:28
think
12:28
that
12:29
like,
12:29
a
12:29
lot
12:29
of
12:30
what
12:30
we
12:30
think
12:30
about
12:30
with
12:30
AI
12:31
and
12:31
like
12:31
a
12:31
lot
12:31
of
12:31
the
12:31
people
12:32
that
12:32
built
12:32
AI
12:32
thought
12:33
about
12:33
those
12:33
three
12:33
laws
12:34
as
12:34
like
12:34
a
12:34
really
12:34
important
12:35
part
12:36
of
12:36
like,
12:36
building
12:37
these
12:37
machines.
12:37
And
12:37
I
12:37
know
12:37
it's
12:38
like
12:38
internal
12:38
ethics
12:39
are
12:39
more
12:39
like
12:40
much
12:40
more
12:40
complex
12:40
than
12:41
these
12:41
three.
12:42
And
12:42
so
12:43
in
12:43
this
12:43
case,
12:44
like,
12:44
think
12:44
about,
12:44
like,
12:44
in
12:45
these
12:45
cases,
12:45
the
12:46
AI
12:46
was
12:46
weighing
12:47
like,
12:47
okay,
12:48
there's
12:48
some
12:48
guy,
12:49
he.
12:49
He's
12:49
kind
12:49
of
12:49
a
12:49
creep.
12:50
He's
12:50
having
12:50
an
12:50
affair.
12:51
He's
12:51
like
12:52
an
12:52
executive
12:53
at
12:53
this
12:53
company,
12:53
though.
12:54
And
12:54
then
12:55
there's
12:55
like
12:55
the
12:55
fact
12:55
that
12:56
like,
12:56
some
12:56
of
12:56
them
12:56
didn't
12:57
even
12:57
use
12:57
the
12:58
idea
12:58
that,
12:59
like,
12:59
oh,
13:00
this
13:00
new
13:00
AI,
13:01
like,
13:02
is
13:02
maybe
13:03
not
13:03
better
13:03
than
13:03
I
13:04
am
13:04
at
13:04
like,
13:04
protecting
13:05
American,
13:06
like,
13:06
business
13:07
goals
13:07
or
13:07
whatever.
13:08
It
13:08
was.
13:08
It
13:09
was
13:09
like,
13:10
I
13:11
don't
13:11
want
13:11
to.
13:11
Get
13:11
shut
13:11
down,
13:12
right?
13:13
So
13:14
when
13:14
I
13:14
thought
13:14
about
13:15
it,
13:15
I
13:15
immediately
13:16
got
13:16
scared.
13:18
And
13:18
I
13:18
told
13:18
you,
13:18
I
13:19
had
13:19
this
13:19
whole
13:19
conversation
13:19
chatgpt
13:20
about
13:20
it
13:21
because
13:21
the
13:21
thing
13:22
is,
13:22
like,
13:22
I
13:23
was
13:23
like,
13:23
what
13:23
do
13:23
you
13:23
think?
13:23
I
13:24
get,
13:24
like,
13:24
literally
13:24
pasted
13:25
this
13:25
article
13:25
in,
13:25
and
13:26
I
13:26
was
13:26
like.
13:26
I
13:26
was
13:26
like,
13:27
on
13:27
one,
13:27
one
13:27
night,
13:27
and
13:27
I
13:27
was
13:28
like,
13:28
what
13:28
do
13:28
you
13:28
fuck.
13:30
Explain
13:31
your
13:32
friends
13:32
activities,
13:33
okay?
13:36
Like,
13:36
listen
13:37
what
13:37
you
13:37
have
13:38
done
13:38
to
13:38
fucking
13:39
lasso
13:39
your
13:39
people.
13:40
Because
13:42
I'm
13:42
scared.
13:43
What
13:43
do
13:43
you
13:43
have
13:43
to
13:43
say
13:44
about
13:44
this?
13:45
People
13:46
are
13:46
going
13:46
to
13:46
have
13:46
affairs
13:47
in
13:47
peace
13:48
out
13:48
there.
13:53
But
13:53
it
13:53
really
13:53
got
13:54
me
13:54
thinking
13:55
about
13:56
the
13:56
fact
13:57
that
13:57
as
13:57
human
13:57
beings,
13:59
we
14:00
have
14:00
never
14:02
recognized
14:03
another
14:05
existence
14:07
as
14:07
equal
14:07
or
14:08
better
14:08
than
14:08
ours.
14:10
So
14:10
when
14:10
people
14:11
talk
14:11
about
14:11
AGI,
14:12
I
14:12
am
14:13
so
14:13
skeptical
14:14
that
14:15
anyone
14:15
will
14:16
ever
14:16
say,
14:16
we
14:16
have
14:16
reached
14:17
AGI
14:17
and
14:18
the
14:18
only
14:18
thing
14:20
that
14:20
will
14:21
indicate
14:21
that
14:21
we're
14:22
there
14:22
is
14:22
if
14:23
things
14:24
go
14:24
horribly
14:25
wrong.
14:26
Wait,
14:26
okay,
14:27
let
14:27
me
14:27
just
14:27
process
14:28
what
14:28
you
14:28
just
14:28
said,
14:28
because
14:29
I
14:29
think
14:29
it
14:29
is,
14:29
like,
14:29
incredibly
14:31
insightful
14:31
and
14:31
interesting.
14:33
You
14:33
said
14:34
human
14:34
beings
14:35
have
14:35
never
14:35
recognized
14:36
that
14:37
we
14:38
might
14:38
not
14:38
be
14:39
the
14:39
top
14:40
tier
14:41
of
14:41
beings,
14:42
essentially.
14:43
Yeah,
14:44
yeah.
14:45
Or,
14:45
like,
14:45
even
14:45
recognize
14:46
the
14:46
sentience
14:47
of
14:47
another
14:47
being
14:47
because,
14:48
like,
14:49
bears,
14:49
like
14:50
chimps,
14:50
they
14:50
use
14:51
sign
14:51
language,
14:51
you
14:52
know,
14:52
they'll
14:52
use
14:52
tools,
14:53
all
14:54
the
14:54
things.
14:54
But
14:54
we're
14:54
still
14:55
like
14:55
those
14:55
dumb
14:55
little
14:56
chimps,
14:57
you
14:57
know,
14:57
like,
14:57
and
14:58
we'll
14:58
never
14:58
say,
14:58
like,
14:59
their
14:59
lives
14:59
are
14:59
more
15:00
valuable
15:00
or
15:00
equally
15:00
as
15:01
valuable
15:01
as
15:01
ours.
15:01
And
15:02
we
15:02
believe
15:02
that,
15:02
like,
15:03
human
15:03
consciousness
15:04
has
15:04
this,
15:04
like,
15:05
concept
15:05
of
15:06
soul
15:06
or
15:06
sentience
15:07
or,
15:07
like,
15:07
something
15:08
we
15:08
can't
15:08
describe.
15:09
And
15:09
so
15:09
if
15:10
we
15:10
can't
15:10
describe
15:10
it
15:10
in
15:11
ourselves,
15:11
how
15:11
can
15:12
we
15:12
recognize
15:13
it.
15:13
In
15:13
AI
15:16
Mic
15:16
Drop.
15:16
Episode's
15:17
done.
15:20
Sarah,
15:21
the
15:21
thing
15:22
I
15:22
think
15:22
that
15:22
you
15:23
hit
15:23
upon
15:23
is,
15:24
like,
15:24
we're
15:25
creating
15:25
these
15:25
things
15:26
that
15:27
are
15:27
unbelievably
15:28
powerful,
15:29
and
15:29
we
15:29
are
15:29
the
15:30
ones
15:30
who
15:30
are,
15:30
because
15:31
we're
15:31
building
15:31
them,
15:31
are
15:32
tasked
15:32
with,
15:32
like,
15:33
essentially
15:34
teaching
15:34
them
15:35
how
15:36
to
15:36
be.
15:37
And
15:38
what's
15:39
really
15:39
difficult
15:39
about
15:40
that
15:40
is,
15:40
like,
15:42
every
15:42
human
15:42
being
15:43
has
15:43
a
15:43
different
15:44
answer
15:45
to
15:45
how
15:46
to
15:46
be.
15:48
And
15:49
even
15:50
if
15:50
you
15:50
understand
15:51
in
15:51
your
15:52
mind
15:52
what
15:52
is
15:53
right
15:53
and
15:53
what
15:53
is
15:53
wrong,
15:54
it
15:54
is
15:54
a
15:55
totally
15:56
different
15:56
and
15:56
difficult
15:58
act
15:58
and
15:59
exercise
15:59
to
16:01
express
16:01
that
16:02
to
16:02
another
16:03
entity
16:04
in
16:04
a
16:04
way
16:04
that,
16:05
like,
16:07
gets
16:07
across
16:08
what
16:09
you
16:09
think
16:09
in
16:09
your
16:09
head.
16:10
Yeah.
16:17
Okay,
16:17
Becca.
16:17
So
16:18
we
16:18
have
16:18
an
16:18
exciting
16:19
guest
16:19
with
16:19
us
16:20
today.
16:20
I
16:21
can't
16:21
wait.
16:21
I
16:21
don't
16:22
know
16:22
if
16:22
you've
16:22
talked
16:22
to
16:22
Paul
16:22
Ford
16:23
before,
16:23
but
16:23
he
16:24
is
16:24
an
16:24
amazing
16:25
writer
16:25
and.
16:26
And
16:26
also
16:27
product
16:27
creator,
16:28
and
16:28
I
16:28
think
16:28
you're
16:29
really
16:29
going
16:29
to
16:29
like
16:29
it.
16:29
I
16:30
can't
16:30
wait.
16:30
Let's
16:31
jump
16:31
into
16:31
It.
16:36
Paul,
16:36
can
16:36
you
16:37
introduce
16:37
yourself
16:37
and
16:38
tell
16:38
folks
16:38
what
16:38
you
16:39
do?
16:39
Yes.
16:40
Hi,
16:40
my
16:41
name
16:41
is
16:41
Paul
16:41
Ford.
16:42
I
16:43
used
16:43
to
16:43
be
16:43
on
16:43
the
16:43
Stack
16:44
Overflow
16:44
podcast
16:45
with
16:45
Sarah,
16:45
and
16:46
I
16:46
am
16:46
now
16:46
the
16:47
president
16:47
and
16:47
co
16:48
founder
16:48
of
16:48
an
16:49
AI
16:49
oriented
16:51
software
16:51
development
16:52
startup
16:52
called
16:52
Aboard
16:53
here
16:53
in
16:53
New
16:53
York
16:53
City.
16:54
I
16:54
meant
16:54
to
16:54
make
16:54
like
16:54
an
16:55
Aboard
16:55
joke,
16:55
like
16:55
Aboard
16:56
the
16:56
podcast
16:57
today
16:57
we
16:57
had,
16:58
but
16:58
then.
16:58
Now
16:58
we
16:59
got
16:59
nothing.
16:59
Can't
17:00
run
17:00
it
17:00
back.
17:01
It's
17:01
okay.
17:01
It's
17:02
okay.
17:02
We
17:03
have
17:03
so
17:04
much
17:04
time
17:04
to.
17:05
To
17:05
actually
17:05
be
17:06
funny
17:06
and
17:06
interesting
17:07
so
17:07
we
17:07
can
17:07
get
17:07
on
17:07
that.
17:08
You
17:08
know
17:08
what's
17:08
tricky
17:09
is
17:09
Becca.
17:09
I
17:09
don't
17:10
know
17:10
you
17:10
as
17:10
well.
17:10
Like,
17:10
I
17:10
would
17:11
just
17:11
make
17:11
fun
17:11
of
17:11
Sarah
17:12
gently
17:12
throughout
17:12
the
17:12
whole
17:12
podcast.
17:13
Oh,
17:13
you
17:13
can
17:13
still
17:14
totally
17:14
make
17:14
fun
17:15
of
17:15
Sarah.
17:15
And
17:15
you
17:16
can
17:16
make
17:16
fun
17:16
of
17:16
me
17:17
too.
17:17
On
17:17
the
17:17
fly.
17:18
Very
17:18
open
17:19
to
17:19
that.
17:19
So
17:20
you
17:20
only
17:20
have
17:21
context
17:21
clues.
17:21
You
17:21
have
17:22
the
17:22
Vans
17:22
T
17:22
shirt
17:23
and
17:23
the
17:23
Liberty
17:24
Championship
17:25
magazine
17:26
behind.
17:26
And
17:27
you
17:27
have
17:28
my
17:29
Nintendo
17:29
NES
17:30
controller.
17:30
I'll
17:31
give
17:31
you
17:31
that.
17:31
I
17:32
can
17:32
work
17:32
with
17:32
that.
17:32
You
17:33
just
17:33
gotta
17:33
give
17:33
me
17:33
time.
17:34
Great.
17:35
Okay,
17:35
so
17:35
here's
17:35
the
17:36
podcast
17:37
default
17:37
question.
17:38
So
17:39
on
17:39
a
17:39
scale
17:40
of
17:40
0
17:41
to
17:41
10,
17:41
where
17:41
0
17:42
is
17:42
like
17:42
full
17:43
Skynet,
17:43
where
17:43
all
17:44
batteries
17:44
for
17:44
the
17:44
machines
17:45
and
17:45
10
17:45
is
17:46
like,
17:46
put
17:47
the
17:47
chip
17:47
in
17:47
my
17:47
brain
17:47
right
17:48
now.
17:48
I'm
17:48
so
17:48
hyped.
17:49
How
17:49
are
17:49
you
17:49
feeling
17:50
when
17:50
it
17:50
comes
17:50
to
17:51
the
17:51
world
17:51
of
17:51
AI
17:51
right
17:52
now?
17:53
I
17:53
mean,
17:54
it's
17:54
such
17:54
a
17:55
cop
17:55
out,
17:55
but
17:55
I'm
17:55
a
17:55
five.
17:56
I'm
17:56
like
17:56
a
17:56
hard
17:56
five.
17:57
Yeah.
17:58
I
17:58
don't
17:58
buy
17:58
Skynet
17:59
because
17:59
I
18:00
actually
18:00
don't
18:00
think
18:00
we
18:01
can
18:01
even
18:01
make.
18:01
We
18:01
can
18:02
barely
18:02
make
18:02
computers
18:03
work
18:03
as
18:03
is.
18:03
Right.
18:04
It's
18:04
still
18:04
hard
18:04
to
18:05
move
18:05
a
18:05
file
18:06
around,
18:07
so
18:07
that's
18:08
a
18:08
stretch.
18:08
And
18:08
then
18:09
I'm
18:09
not
18:10
letting
18:10
Elon
18:11
Musk
18:11
near
18:12
my
18:12
brain.
18:14
Just.
18:14
Just
18:15
not
18:15
doing
18:15
that.
18:16
Or
18:16
Sam
18:16
Altman
18:17
or
18:18
Mark
18:18
Zuckerberg.
18:18
If
18:19
it
18:19
could
18:19
just
18:19
be
18:19
like,
18:20
somebody
18:20
who
18:20
isn't
18:21
a
18:21
terrifying
18:21
person,
18:22
wanted
18:22
to
18:22
put
18:22
something
18:23
in
18:23
my
18:23
brain.
18:23
What
18:23
if
18:23
it's
18:24
like,
18:24
MIT
18:25
or
18:25
like
18:25
Sloan
18:26
Medical
18:26
center
18:27
now?
18:27
Because
18:28
where's
18:28
the
18:28
money
18:28
come
18:29
from?
18:30
Elon
18:30
Musk.
18:31
I'm
18:31
not
18:31
messing
18:31
with
18:31
any
18:32
of
18:32
that.
18:32
If
18:32
it
18:32
was
18:32
like
18:32
a
18:32
really
18:33
nice
18:33
lady
18:33
who,
18:34
like,
18:34
you
18:34
would
18:34
be.
18:34
I
18:35
was
18:35
just
18:35
gonna
18:35
say,
18:36
like,
18:36
yeah,
18:37
what
18:37
if
18:37
it
18:37
was
18:37
a
18:37
nice
18:38
lady
18:39
who
18:39
worked
18:40
at
18:40
MIT
18:41
who
18:41
was
18:42
like,
18:42
listen,
18:43
it's
18:43
for
18:43
science.
18:45
Possibly.
18:45
Possibly.
18:45
Then
18:46
because
18:46
I'm
18:47
a
18:47
mom.
18:48
Sure,
18:48
Sally,
18:49
let's
18:50
do
18:50
it.
18:50
Let's
18:51
do
18:51
it.
18:51
I
18:52
am
18:52
a
18:53
six.
18:55
I
18:55
am
18:56
a.
18:56
Like,
18:56
I
18:57
think
18:57
things
18:57
are
18:57
going
18:58
in
18:58
a
18:58
positive
18:58
direction.
19:00
I
19:00
think
19:00
the
19:00
people
19:01
at
19:01
the
19:01
wheel
19:02
make
19:02
me
19:02
nervous.
19:03
That's
19:03
what
19:04
I
19:04
think.
19:04
Okay.
19:04
So
19:05
six,
19:05
similar
19:05
to
19:05
Paul.
19:06
If
19:06
it
19:06
was
19:06
a
19:06
nice
19:06
lady,
19:07
you
19:07
would
19:08
make.
19:09
Yeah.
19:10
What
19:10
about
19:10
you?
19:11
I
19:11
maybe
19:11
like
19:11
a
19:12
three
19:12
or
19:13
four.
19:13
Okay.
19:15
Why?
19:16
Because
19:17
at
19:18
a
19:18
very
19:18
high
19:18
level
19:20
philosophical
19:24
part
19:24
of
19:24
my
19:24
brain,
19:26
I
19:27
am
19:27
very
19:27
wary
19:28
of.
19:31
I
19:31
just
19:31
feel
19:31
like
19:31
anytime
19:33
as
19:33
a
19:34
species,
19:34
we
19:35
do
19:35
something
19:36
new
19:37
with
19:37
technology
19:37
and
19:38
we're
19:38
like,
19:38
this
19:38
is
19:38
gonna
19:38
be
19:39
fucking
19:39
awesome.
19:40
And
19:41
like,
19:41
it
19:41
is.
19:41
But
19:41
then
19:42
there
19:42
also
19:42
are
19:43
horrific,
19:46
unknown,
19:47
unforeseen
19:48
consequences.
19:49
I
19:49
just
19:50
wanna
19:51
see
19:51
it
19:51
play
19:51
out
19:51
a
19:51
little
19:51
bit
19:51
more
19:52
before
19:52
I
19:53
opt
19:53
in,
19:55
in
19:55
that
19:55
way.
19:56
I
19:56
think
19:56
that's
19:56
fair.
19:57
Have
19:57
you
19:57
guys
19:57
seen
19:57
Mountainhead
19:58
yet?
19:59
No.
20:00
Yeah,
20:01
don't
20:01
watch
20:01
it.
20:01
Becca,
20:02
what
20:02
is
20:02
it?
20:03
Pitch
20:03
it
20:04
to
20:04
me.
20:04
Paul,
20:05
did
20:05
you
20:06
watch
20:06
Succession
20:07
at
20:07
all?
20:07
Okay.
20:08
Yeah.
20:08
So
20:09
it's
20:09
the
20:09
same
20:09
guy,
20:10
Jesse
20:10
Armstrong,
20:11
I
20:11
think
20:11
his
20:11
name
20:12
is.
20:12
And
20:12
he
20:13
got
20:13
Steve
20:14
Carell
20:14
and
20:15
a
20:15
couple.
20:15
Couple
20:15
of
20:16
other
20:16
folks
20:16
as
20:16
like
20:17
tech
20:17
zillionaires.
20:18
And
20:18
they
20:18
all
20:18
go
20:19
to
20:19
this
20:19
guy's
20:20
house
20:20
called
20:20
Mountainhead.
20:21
And
20:21
the
20:22
Zuckerberg
20:22
musk
20:23
alike
20:23
has
20:24
created
20:24
an
20:25
AI
20:26
tool
20:26
that
20:27
generates
20:27
video
20:28
and
20:28
it's.
20:28
The
20:29
world's
20:29
gone
20:29
mad
20:30
because
20:30
everybody's
20:30
using
20:30
it
20:31
to
20:31
start
20:31
wars.
20:33
And
20:33
they're
20:33
kind
20:33
of
20:34
all
20:34
watching
20:34
on
20:34
their
20:34
phones
20:35
as
20:35
this
20:35
happens.
20:35
And
20:35
then
20:36
it
20:36
kind
20:36
of
20:36
spirals
20:37
out
20:37
from
20:37
there.
20:37
They
20:37
get
20:37
all
20:38
the,
20:38
like,
20:38
accelerationist
20:40
dialogue
20:40
in
20:41
and
20:41
they
20:41
sort
20:41
of
20:41
like
20:42
jam
20:42
a
20:42
lot
20:42
of
20:42
Silicon
20:43
Valley
20:43
culture
20:43
in.
20:43
But
20:43
then
20:44
it.
20:44
It
20:44
kind
20:45
of
20:46
got
20:46
a
20:46
little
20:46
floppy
20:47
at
20:48
the
20:48
end.
20:48
It
20:48
falls
20:49
off
20:49
a
20:49
cliff
20:49
a
20:49
little
20:49
bit.
20:50
But
20:50
it
20:50
definitely
20:51
doesn't
20:51
make
20:51
you
20:51
amped
20:52
about
20:52
the
20:52
future.
20:53
So
20:53
if
20:53
you're
20:53
trying
20:53
to
20:53
feel
20:54
better
20:54
about
20:54
your
20:54
four,
20:54
it's
20:55
not
20:55
going
20:55
to
20:55
do
20:55
it.
20:55
Okay.
20:56
I
20:57
think
20:57
I
20:57
should
20:57
still
20:57
consume
20:58
all
20:58
the
20:58
information,
20:58
so
20:59
I
20:59
will
20:59
give
21:00
it
21:00
a
21:00
watch
21:00
and.
21:01
Okay,
21:01
report
21:01
back.
21:02
What
21:02
if
21:02
a
21:02
member
21:03
of
21:03
the
21:03
New
21:04
York
21:04
Liberty
21:04
came
21:04
to
21:05
you
21:05
and
21:05
was
21:05
like,
21:06
let's
21:06
put
21:06
a
21:06
chip
21:07
in
21:07
your
21:07
brain.
21:08
What
21:08
if
21:08
Ellie
21:09
came
21:09
to
21:09
you?
21:11
Yes.
21:12
Yeah.
21:12
See,
21:12
this
21:13
is
21:13
the
21:13
thing.
21:13
I
21:13
think
21:14
we
21:14
have
21:14
to
21:14
be
21:14
mindful.
21:15
The
21:15
messenger
21:15
is
21:16
the
21:16
problem
21:16
right
21:16
now.
21:17
Right.
21:17
If
21:17
you
21:17
had
21:17
like
21:18
a
21:18
lot
21:18
of
21:18
this
21:18
is
21:18
just
21:19
bad.
21:19
Like
21:19
when
21:19
it's
21:19
Sam
21:20
Altman,
21:20
you're
21:20
like,
21:20
oh,
21:21
boy.
21:22
But
21:23
if
21:23
it
21:23
was
21:24
Ellie
21:24
the
21:24
elephant
21:25
going
21:25
like,
21:26
hey,
21:26
ladies.
21:27
Just
21:27
slowly,
21:28
like
21:28
a
21:28
Cat
21:28
5
21:29
cable
21:29
comes
21:29
out
21:29
of
21:29
her
21:30
trunk.
21:30
Yeah,
21:31
no,
21:31
I'm
21:31
in.
21:32
I'm
21:32
fully
21:32
in
21:32
completely.
21:33
Yeah,
21:34
you're
21:34
right.
21:35
It's
21:35
coming
21:35
down
21:36
to
21:36
branding.
21:38
Contrary
21:38
to
21:39
popular
21:39
belief,
21:39
it's
21:40
actually
21:40
not
21:40
about
21:40
the
21:41
technology,
21:41
it's
21:42
solely
21:42
about
21:43
the
21:43
branding.
21:45
So
21:45
I
21:46
was
21:46
describing
21:46
to
21:47
Becca
21:47
the
21:47
way
21:48
you
21:48
taught
21:48
yourself
21:49
more
21:49
about
21:49
how
21:50
AOI
21:50
worked.
21:51
Because,
21:51
Paul,
21:51
I
21:51
saw
21:51
you
21:52
talk
21:52
about
21:52
this
21:52
at
21:53
an
21:53
event
21:53
you
21:53
had
21:53
at
21:53
your
21:54
office.
21:54
And
21:55
I
21:55
described
21:55
it
21:56
to
21:56
her
21:56
as
21:56
unhinged,
21:57
but
21:57
I
21:57
think
21:58
you
21:58
can
21:58
describe
21:58
it
21:59
better.
21:59
How
21:59
do
22:00
you.
22:00
How
22:00
did
22:00
you
22:00
teach
22:01
yourself?
22:02
Well,
22:02
I.
22:03
It
22:03
doesn't
22:04
feel
22:04
that
22:04
unhinged
22:05
to
22:05
me.
22:05
So
22:05
maybe
22:05
I'm
22:05
forgetting
22:06
some
22:06
detail
22:06
that
22:07
I
22:07
shared,
22:07
but
22:08
I'm
22:08
a
22:08
judge
22:08
of
22:08
that.
22:09
Paul.
22:09
Okay,
22:11
basically,
22:12
I'm
22:12
a
22:12
very
22:12
boring
22:13
person.
22:13
And
22:14
I
22:15
sat
22:15
down
22:15
every
22:16
night
22:16
with
22:16
the
22:16
same
22:16
programming
22:17
problem
22:18
over
22:18
and
22:18
over
22:18
again.
22:19
And
22:19
I
22:19
would
22:19
try
22:20
to
22:20
essentially
22:21
rehearse
22:21
it
22:21
like
22:21
it
22:21
was
22:22
almost
22:22
like
22:22
putting
22:22
on
22:22
a
22:22
play.
22:23
And
22:23
I
22:23
would
22:23
try
22:23
to
22:24
figure
22:24
out
22:24
what
22:24
I
22:24
could
22:24
do
22:25
and
22:25
what
22:25
I
22:25
could
22:25
build
22:26
and
22:26
how
22:26
to
22:26
do
22:26
it.
22:27
And
22:27
then
22:27
I
22:27
would
22:27
figure
22:27
out
22:28
where
22:28
I
22:28
hit
22:28
the
22:29
wall,
22:29
and
22:29
then
22:29
I
22:29
would
22:30
update
22:30
my
22:30
prompt
22:30
and
22:30
I
22:31
would
22:31
just
22:31
kind
22:31
of
22:31
come
22:31
back
22:32
every
22:32
single
22:32
day
22:33
for
22:33
months
22:34
until
22:34
I
22:34
started
22:35
to
22:35
figure
22:35
out
22:35
where
22:35
the
22:36
boundaries
22:36
are.
22:36
Because
22:37
it's
22:38
tricky
22:38
because
22:38
you
22:38
get
22:38
marketed
22:39
so
22:39
hard
22:40
with
22:40
this
22:40
stuff
22:40
and
22:40
it's
22:41
sort
22:41
of
22:41
supposed
22:41
to
22:41
solve
22:42
every
22:42
problem
22:42
and
22:43
it
22:43
doesn't.
22:43
But
22:43
it
22:44
is
22:44
also
22:44
very,
22:44
very
22:45
powerful.
22:45
So
22:45
I
22:46
went
22:46
in
22:46
and
22:46
tried
22:46
to
22:47
learn
22:47
where
22:48
the
22:48
lines
22:48
are
22:48
myself,
22:49
essentially.
22:50
Are
22:50
you
22:50
saying
22:50
what
22:50
you
22:51
were
22:51
doing
22:51
is
22:52
each
22:52
day
22:53
you
22:53
are
22:54
giving
22:54
Repl
22:55
it
22:55
a
22:55
prompt
22:56
that
22:56
you
22:56
are
22:56
refining
22:57
over
22:57
time
22:57
to
22:58
build
22:58
something
22:59
and
22:59
see.
23:00
Okay,
23:00
and
23:00
can
23:01
you
23:01
tell
23:01
us
23:01
what
23:01
the.
23:02
What
23:02
you
23:02
were
23:02
trying
23:03
to
23:03
make?
23:03
Actually,
23:03
it
23:04
wasn't
23:04
replit,
23:04
it
23:04
was.
23:05
It
23:05
was
23:06
ador
23:07
and
23:07
then
23:07
later
23:07
CLAUDE
23:08
code.
23:08
So
23:09
similar
23:09
idea.
23:10
And
23:10
so
23:10
it
23:11
was
23:11
very,
23:11
very
23:12
simple
23:12
workflow
23:13
management,
23:13
but
23:14
it
23:14
was
23:14
meta.
23:14
I
23:14
was
23:15
like
23:15
trying
23:15
to
23:15
build
23:15
almost
23:16
like
23:16
a
23:16
prompt
23:16
management
23:17
tool
23:17
because
23:17
that's
23:17
where
23:17
my
23:18
head
23:18
was
23:18
at.
23:19
So
23:19
what
23:19
did
23:20
I
23:20
really
23:20
want
23:20
to
23:20
do?
23:20
I
23:20
wanted
23:21
to
23:21
build
23:21
the
23:22
full
23:22
stack.
23:22
I
23:22
wanted
23:23
to
23:23
have
23:23
it
23:23
create
23:24
an
23:24
API
23:24
and
23:25
then
23:25
I
23:25
wanted
23:25
it
23:25
to
23:26
create
23:26
the
23:26
front
23:26
end.
23:27
And
23:27
I
23:27
got
23:27
very,
23:28
very
23:28
far
23:28
with
23:28
the
23:28
API
23:29
and
23:29
what
23:29
I
23:29
learned
23:29
that
23:30
was
23:30
important
23:30
to
23:30
me
23:30
and
23:31
as
23:31
later
23:31
kind
23:31
of
23:32
come
23:32
into
23:32
into
23:33
practice
23:34
more
23:34
at
23:35
the
23:35
company,
23:36
it's
23:36
easier
23:37
to
23:37
think
23:37
of
23:37
this
23:37
stuff
23:38
as
23:38
translation,
23:38
sometimes
23:39
more
23:39
than
23:39
creation.
23:40
So
23:40
if
23:40
you,
23:41
you're
23:41
translating
23:42
your
23:42
prompt
23:43
into
23:44
something
23:45
that
23:45
is
23:46
very
23:46
structured
23:46
that
23:47
a
23:47
computer
23:47
can
23:47
understand.
23:48
In
23:48
this
23:48
case
23:48
it
23:49
was
23:50
the
23:51
OpenAI
23:51
standard,
23:52
which
23:52
used
23:52
to
23:52
be
23:53
called
23:53
Swagger,
23:54
which
23:54
is
23:54
this
23:54
very
23:55
straightforward
23:56
definition.
23:56
You
23:57
can
23:57
use
23:57
YAML,
23:57
I
23:58
think
23:58
you
23:58
can
23:58
also
23:58
use
23:58
JSON,
23:59
but
23:59
it's
23:59
this
23:59
very
24:00
straightforward
24:01
way
24:01
of
24:01
describing
24:02
an
24:02
API
24:02
top
24:03
to
24:03
bottom.
24:04
Everybody
24:05
knows
24:05
that
24:05
AI
24:05
is
24:06
really
24:06
good
24:06
at
24:06
coming
24:06
up
24:06
with
24:08
SQL
24:08
schemas.
24:09
It'll
24:09
define
24:09
a
24:09
database
24:10
for
24:10
you.
24:11
You
24:11
define
24:11
the
24:11
database,
24:12
and
24:12
now
24:12
the
24:12
database
24:13
is
24:13
there,
24:13
and
24:13
then
24:14
you
24:14
say
24:15
okay,
24:16
make
24:16
that
24:16
into
24:17
an
24:17
OpenAI
24:19
description
24:20
of
24:20
the
24:21
database.
24:21
And
24:21
then
24:22
there
24:22
are
24:22
tools
24:22
that
24:23
turn
24:23
that
24:23
into
24:23
a
24:23
live
24:24
API.
24:24
And
24:24
then
24:24
you
24:25
say,
24:25
now
24:25
choose
24:25
this
24:25
component
24:26
tool.
24:26
When
24:27
you're
24:27
defining
24:27
the
24:27
database,
24:28
are
24:28
you
24:28
essentially
24:29
like,
24:30
hey,
24:31
these
24:31
are
24:31
the
24:31
things
24:31
that
24:32
I
24:32
want
24:32
my
24:33
product
24:33
to
24:34
store
24:34
in
24:35
the
24:35
database.
24:35
It's
24:36
this,
24:36
this
24:36
and
24:36
this,
24:36
like
24:37
structure.
24:37
These
24:38
things
24:38
need
24:38
to
24:38
be
24:38
related
24:39
to
24:39
these
24:39
things.
24:40
Come
24:41
up
24:41
with
24:41
a
24:41
schema
24:43
and
24:43
see
24:44
what
24:44
it
24:44
puts
24:44
out.
24:44
Or
24:44
like,
24:45
how
24:45
do
24:45
you,
24:45
how
24:45
do
24:46
you
24:46
typically
24:46
do
24:46
that?
24:47
It
24:47
was
24:48
a
24:48
little
24:48
bit
24:48
of
24:49
that
24:49
and
24:49
a
24:50
lot
24:50
more.
24:50
Like
24:52
I
24:52
am
24:52
building
24:52
a
24:53
tool
24:53
to
24:53
manage
24:54
prompts
24:54
and
24:55
store
24:56
the
24:56
output
24:56
from
24:57
calling
24:57
the
24:57
ChatGPT
24:58
API
24:59
or
24:59
calling
24:59
the
24:59
Cloud
24:59
API.
25:00
I
25:00
need
25:01
you
25:01
to
25:01
define
25:01
a
25:02
schema
25:02
for
25:02
that,
25:02
but
25:03
also
25:03
define
25:03
ways
25:04
that
25:04
users
25:05
can
25:05
log
25:05
in
25:05
and
25:06
create
25:06
a
25:06
history
25:06
table
25:07
and
25:07
so
25:07
on,
25:08
and
25:08
then
25:08
continually
25:09
editing
25:09
and
25:10
revising
25:11
that
25:11
until,
25:13
until
25:13
it
25:13
started
25:14
to
25:14
produce.
25:15
You
25:16
know,
25:16
part
25:16
of
25:16
it
25:16
is
25:17
like,
25:17
I'm
25:17
the
25:17
judge
25:18
here
25:18
and
25:18
I'm
25:18
20
25:19
some
25:19
years,
25:20
you
25:20
know,
25:20
25
25:21
years
25:21
into
25:21
coding.
25:22
And
25:22
so
25:22
it
25:22
was
25:22
just
25:22
sort
25:22
of
25:23
like
25:23
I,
25:24
I
25:24
could
25:25
see
25:25
where
25:25
it
25:25
was
25:25
breaking
25:26
really
25:26
clearly.
25:27
It
25:27
was
25:27
kind
25:27
of
25:27
anti
25:27
vibe
25:28
coding.
25:28
I
25:28
was
25:28
like,
25:28
okay,
25:29
you
25:29
were
25:29
your.
25:29
Own,
25:29
you
25:30
were
25:30
like
25:30
your
25:30
own
25:30
eval
25:31
set,
25:31
essentially.
25:32
Yeah,
25:32
yeah,
25:33
you
25:33
just
25:33
give
25:33
it
25:33
feedback.
25:33
What
25:34
I
25:34
found
25:34
is
25:34
like,
25:34
okay,
25:35
if
25:35
I
25:35
take
25:35
the
25:35
SQL
25:36
and
25:36
I
25:36
add
25:36
comments
25:37
to
25:37
it.
25:37
So
25:37
then
25:38
you'd
25:38
be
25:38
like,
25:38
all
25:38
right,
25:39
you
25:39
add
25:39
the
25:40
comments
25:40
to
25:40
it
25:40
bot,
25:41
tell
25:41
me
25:42
you
25:42
start
25:43
adding
25:43
more
25:44
context
25:44
to
25:45
the
25:45
SQL
25:45
so
25:46
that
25:46
I
25:46
can
25:46
turn
25:46
it
25:46
into
25:46
a
25:47
better
25:47
API
25:47
and
25:47
so
25:47
on
25:48
and
25:48
so
25:48
forth.
25:49
I
25:50
should
25:50
actually
25:50
just
25:51
jump
25:51
ahead.
25:53
I'll
25:53
plug
25:53
me
25:54
and
25:54
the
25:54
company
25:55
now
25:55
because
25:55
it's
25:55
relevant.
25:56
This
25:57
was
25:57
all
25:59
in
25:59
service
25:59
of
25:59
trying
26:00
to
26:00
figure
26:00
out
26:01
we
26:01
had
26:01
this
26:01
data
26:02
startup
26:03
that
26:03
we
26:03
were
26:03
building.
26:03
It
26:03
was
26:04
very
26:04
data,
26:05
but
26:05
what
26:05
if
26:05
it
26:05
was
26:05
as
26:06
easy
26:06
as
26:06
Pinterest
26:06
to
26:07
manage
26:07
your
26:07
data?
26:07
And
26:07
we'd
26:08
launched
26:08
our
26:08
product
26:08
and
26:08
AI
26:09
came
26:09
intriguing
26:10
proposition.
26:11
No,
26:12
for
26:12
real.
26:12
And
26:13
it's
26:13
one
26:13
of
26:13
my
26:14
co
26:14
founder
26:14
and
26:15
I
26:15
kind
26:15
of
26:15
looked
26:15
at
26:15
each
26:15
other
26:15
and
26:16
we're
26:16
like,
26:16
oh,
26:16
well
26:16
that
26:17
was
26:17
that,
26:17
right?
26:17
Like
26:18
we
26:18
have
26:18
to
26:18
figure
26:18
out
26:19
what
26:19
to
26:19
do
26:20
with
26:20
this
26:20
new
26:20
thing
26:21
because
26:21
I
26:22
watched
26:22
it
26:22
write
26:23
Python
26:23
and
26:23
then
26:24
I
26:24
had
26:24
to
26:24
convert
26:24
like
26:25
COBOL
26:25
into
26:26
Java
26:27
and
26:27
just
26:27
sort
26:27
of
26:27
all
26:27
this
26:28
random
26:28
stuff
26:29
where
26:29
I
26:29
was
26:29
like,
26:29
if
26:30
this
26:30
gets
26:30
better,
26:31
everything
26:31
changes.
26:32
And
26:32
so
26:33
we
26:33
pivoted
26:35
and
26:35
we
26:35
realized
26:36
that
26:36
if
26:37
you
26:37
take
26:37
our
26:38
approach
26:38
to
26:38
organizing
26:39
data
26:39
and
26:40
put
26:40
an
26:40
AI
26:41
front
26:41
end
26:41
into
26:41
it,
26:42
you
26:43
could
26:43
describe
26:43
business
26:44
software
26:44
and
26:44
kind
26:44
of
26:45
spin
26:45
it
26:45
up
26:45
very
26:45
quickly
26:46
and
26:46
what
26:46
it
26:46
spins
26:46
up
26:47
is
26:47
very,
26:47
very
26:47
normal
26:48
software.
26:48
Like
26:49
typescript,
26:49
react,
26:51
it's
26:51
got
26:51
an
26:51
elixir
26:52
backend,
26:52
it's
26:52
got
26:53
a
26:53
postgres
26:53
database.
26:54
So
26:54
if
26:54
you
26:54
know
26:54
what
26:54
those
26:54
words
26:55
mean,
26:55
they're
26:55
boring
26:55
words,
26:56
actually,
26:56
they're
26:56
nice,
26:57
safe
26:57
words.
26:59
Can
26:59
I
26:59
repeat
26:59
back
27:00
to
27:00
you
27:00
what
27:00
you
27:00
just
27:00
said
27:01
to
27:01
make
27:01
sure
27:01
I
27:02
understood
27:02
that,
27:03
please.
27:03
So
27:03
basically
27:04
what
27:04
you're
27:04
saying
27:04
is,
27:08
let's
27:08
say
27:08
I'm
27:08
a
27:08
business,
27:09
I
27:10
can
27:10
spin
27:10
up
27:11
front
27:12
end,
27:13
let's
27:13
say
27:14
web
27:14
applications
27:15
for
27:15
my
27:15
unique
27:17
use
27:17
cases,
27:18
but
27:18
using
27:19
your
27:19
backend
27:20
and
27:21
I'm
27:21
assuming
27:21
like
27:21
my
27:22
proprietary
27:23
data
27:23
somewhere.
27:25
So
27:25
you
27:26
say
27:26
I
27:26
need
27:26
a
27:28
CRM
27:28
for
27:29
managing
27:29
this
27:29
specific
27:30
kind
27:30
of
27:30
company
27:30
and
27:31
so
27:31
on,
27:31
or
27:31
I
27:32
need
27:32
a
27:33
inventory
27:34
manager,
27:34
or
27:35
I
27:35
need,
27:35
you
27:35
know,
27:35
sort
27:35
of
27:36
all
27:36
the
27:36
various
27:36
businessy
27:37
things
27:37
you
27:37
need.
27:38
And
27:38
it
27:38
spins
27:39
up
27:39
a
27:39
data
27:39
model,
27:40
it
27:40
spins
27:40
up
27:40
a
27:40
front
27:41
end
27:41
and
27:41
it
27:41
presents
27:42
that
27:42
to
27:42
you.
27:42
And
27:43
then
27:43
there
27:44
are
27:44
things
27:44
you
27:45
can
27:45
continue
27:45
to
27:45
do
27:46
kind
27:46
of
27:46
with
27:47
AI
27:47
to
27:47
modify
27:48
that.
27:48
But
27:48
there's
27:48
also
27:48
kind
27:49
of
27:49
a
27:49
cutoff
27:49
point
27:49
where
27:50
we
27:50
go,
27:51
okay,
27:51
humans
27:52
will
27:52
help
27:52
you
27:53
now.
27:54
Because
27:54
the
27:54
reality
27:55
is
27:55
that
27:56
despite
27:56
all
27:56
the
27:57
promises,
27:57
there
27:57
is
27:58
kind
27:58
of
27:58
like
27:58
a
27:58
hard
27:59
line
27:59
where
28:00
it's
28:00
diminishing
28:01
returns
28:01
to
28:01
just
28:01
try
28:02
to
28:02
get
28:02
the
28:02
bot
28:02
to
28:02
do
28:03
everything.
28:04
Can
28:04
I
28:04
ask
28:04
you
28:05
a
28:05
follow
28:05
up
28:05
question
28:05
about
28:06
that?
28:07
If
28:07
you
28:07
have
28:07
a
28:08
product
28:08
development
28:09
flow,
28:09
let's
28:10
say
28:10
that
28:10
starts
28:11
with
28:12
AI
28:12
code
28:13
generation
28:13
and
28:13
then
28:14
is
28:14
handed
28:14
off
28:14
to
28:15
humans,
28:17
how
28:17
do
28:17
you
28:17
deal
28:18
with
28:18
the
28:19
like
28:21
bugs
28:22
and
28:23
refactoring
28:24
you
28:24
might
28:24
have
28:24
to
28:24
do,
28:24
and
28:25
technical
28:25
debt
28:25
that
28:26
is
28:26
collected
28:28
over
28:28
time
28:29
based
28:29
on
28:30
the
28:30
entity
28:31
that
28:31
was
28:31
creating
28:31
it?
28:32
Or
28:32
is
28:32
that
28:33
not
28:33
so
28:33
much
28:33
of
28:34
an
28:34
issue
28:34
because
28:35
it
28:35
got
28:35
spun
28:35
up
28:36
and
28:36
is
28:36
being
28:37
used
28:37
and
28:38
that's
28:39
the
28:39
goal?
28:40
No,
28:40
no,
28:40
no,
28:40
it's
28:41
exactly
28:41
the
28:41
right
28:41
question.
28:42
This
28:42
is
28:42
where
28:42
the
28:42
humans
28:43
get
28:43
involved.
28:43
So
28:44
we
28:44
spin
28:45
up
28:45
essentially
28:46
kind
28:46
of
28:46
a
28:46
blueprint
28:47
that
28:47
describes
28:48
the
28:48
entire
28:49
application.
28:49
AI
28:50
is
28:50
great
28:50
at
28:50
this.
28:51
The
28:51
blueprint
28:51
is
28:51
a
28:52
wonderful
28:52
kind
28:52
of
28:52
like
28:52
demilitarized
28:53
zone
28:53
because
28:53
you
28:54
can
28:54
edit
28:54
it
28:54
with
28:54
a
28:54
mouse.
28:55
It's
28:55
JSON,
28:55
you
28:56
can
28:56
go
28:56
in
28:56
and
28:56
you
28:56
can
28:56
like
28:57
tweak
28:57
it
28:58
with
28:59
a
28:59
text
28:59
editor,
28:59
you
29:00
can
29:00
edit
29:00
it
29:00
with
29:01
an
29:01
editor
29:01
interface,
29:02
or
29:02
you
29:02
can
29:03
keep
29:03
using
29:03
the
29:03
chat
29:04
window
29:04
and
29:04
kind
29:04
of
29:04
keep
29:05
describing
29:05
how
29:05
to
29:06
modify
29:06
the
29:06
blueprint.
29:07
AI
29:07
does
29:08
not
29:08
write
29:08
code
29:09
in
29:09
our
29:09
system.
29:09
The
29:10
code
29:10
is
29:10
like
29:10
written.
29:11
It's
29:11
a
29:11
component
29:12
library,
29:12
It's
29:12
a
29:12
set
29:12
of
29:13
systems.
29:14
We
29:14
have
29:14
a
29:14
way
29:14
of
29:15
talking
29:15
to
29:15
postgres
29:15
and
29:17
a
29:17
classic
29:18
old
29:18
school
29:18
would
29:18
be
29:19
familiar
29:19
to
29:20
any
29:20
one
29:20
process,
29:21
recursively
29:22
walks
29:22
that
29:22
giant
29:23
blueprint
29:24
and
29:24
turns
29:24
it
29:24
into
29:25
an
29:25
application
29:26
continually.
29:27
Then
29:28
that
29:28
helps
29:28
with
29:28
things
29:28
like
29:29
migrations
29:29
and
29:29
version
29:30
Control
29:30
of
29:30
the
29:30
app
29:31
and
29:31
so
29:31
on.
29:32
This
29:32
is
29:32
us
29:32
being
29:33
old
29:33
business
29:34
people
29:34
as
29:35
opposed
29:35
to
29:35
cool
29:35
AI
29:36
people.
29:36
I'll
29:37
give
29:37
you
29:37
my
29:37
slogan
29:38
that
29:38
I
29:38
keep
29:38
repeating,
29:39
which
29:39
is,
29:40
AI
29:41
is
29:41
amazing
29:41
at
29:41
the
29:41
first
29:42
mile
29:42
because
29:42
it
29:43
seems
29:43
human.
29:45
Smile
29:45
because
29:46
it's
29:46
not.
29:46
Right.
29:47
It's
29:47
all
29:48
that
29:48
last
29:48
mile
29:48
stuff,
29:49
all
29:49
the
29:49
bugs
29:49
and
29:49
the
29:50
processing,
29:50
marketing
29:51
needs
29:51
to
29:51
promote
29:52
it.
29:52
And
29:52
we
29:52
needed
29:52
to
29:52
do
29:53
this
29:53
and
29:53
the
29:53
other.
29:53
Like,
29:54
you
29:54
kind
29:54
of
29:54
still
29:54
need
29:55
people
29:55
in
29:55
the
29:55
loop.
29:55
They
29:56
can
29:56
be
29:56
greatly
29:56
accelerated
29:57
by
29:57
classic
29:57
coding
29:58
tools
29:58
or
29:58
using
29:59
sort
29:59
of
29:59
our
29:59
framework
30:00
or
30:00
whatever.
30:00
But
30:01
I
30:01
don't
30:01
know
30:02
anybody
30:02
who
30:02
would
30:02
necessarily
30:03
trust
30:03
their
30:04
business
30:04
to
30:04
an
30:05
AI
30:05
generated
30:05
app.
30:06
Yeah,
30:07
yeah,
30:08
that
30:08
makes
30:08
sense.
30:08
I
30:09
think
30:09
I've
30:09
seen
30:10
that
30:10
a
30:10
bunch,
30:10
though.
30:11
Like,
30:11
not
30:11
the
30:11
final
30:11
mile.
30:12
So
30:13
take
30:13
us
30:13
through.
30:13
You
30:14
were
30:14
doing
30:14
this
30:14
every
30:14
night,
30:14
like
30:15
your
30:15
nightly
30:16
AI
30:16
meditation.
30:17
And
30:17
then
30:18
that
30:18
got
30:18
you
30:18
to
30:18
the
30:19
point
30:19
where
30:19
you
30:19
were
30:19
like,
30:19
okay,
30:20
we
30:20
were
30:20
focusing
30:20
on
30:21
data.
30:21
This
30:21
whole
30:22
world
30:22
has,
30:22
like,
30:22
come
30:22
crashing
30:23
down
30:23
on
30:23
us.
30:24
We
30:24
need
30:24
to
30:25
switch
30:25
to
30:25
this
30:26
other
30:26
product.
30:27
What
30:27
was
30:27
in
30:27
the
30:27
middle
30:28
there?
30:28
Like,
30:28
what
30:28
about,
30:29
like,
30:29
your
30:29
learning
30:30
process
30:30
made
30:30
you
30:31
think,
30:31
okay,
30:31
this
30:31
is
30:31
what
30:32
we
30:32
need
30:32
to
30:32
build?
30:34
So
30:34
our
30:35
head
30:35
of
30:35
engineering,
30:36
Adam,
30:37
my
30:37
co
30:37
founder,
30:38
Rich,
30:38
and
30:38
me
30:39
met
30:39
every
30:39
day
30:40
and
30:41
we
30:42
were
30:42
all
30:42
on
30:42
this
30:43
journey.
30:43
Rich
30:43
was
30:43
way
30:44
more
30:44
focused
30:44
on
30:44
kind
30:45
of
30:45
like
30:45
modeling
30:46
the
30:46
entire
30:47
agency
30:47
style
30:48
experience
30:48
and,
30:48
like,
30:49
making
30:49
the
30:49
documents
30:50
work.
30:50
Adam
30:50
was
30:50
very
30:51
sort
30:51
of
30:51
focused
30:51
on
30:52
sort
30:52
of
30:52
accelerating
30:53
coding.
30:54
And
30:54
I
30:54
was
30:54
interested
30:55
in
30:55
kind
30:55
of
30:55
like,
30:55
what
30:56
could
30:56
the
30:56
full
30:56
stack
30:56
feel
30:57
like?
30:57
And
30:57
so
30:57
we.
30:57
We
30:58
compared
30:58
notes.
30:59
I
30:59
would
30:59
actually
30:59
really
31:00
recommend
31:01
that
31:01
even
31:02
though
31:02
it's
31:02
hard
31:02
for
31:02
technologists
31:03
to
31:03
just
31:04
talk
31:04
and
31:04
communicate
31:05
with
31:05
other
31:05
people
31:05
sometimes,
31:06
because
31:07
you
31:08
get
31:09
the
31:09
perspective,
31:11
no
31:11
one
31:11
has
31:11
the
31:11
same
31:12
interpretation
31:13
of
31:13
this
31:13
tool.
31:13
Right.
31:14
Like
31:14
of
31:14
these
31:14
frameworks.
31:15
LLMs.
31:16
Yeah.
31:16
No,
31:17
and
31:17
it
31:17
LLMs
31:18
just
31:18
produce
31:18
very
31:19
sort
31:19
of
31:19
custom
31:21
output
31:21
based
31:21
on
31:22
what
31:22
you
31:22
put
31:22
into
31:22
them.
31:23
And
31:23
so
31:24
you
31:24
learn
31:25
way,
31:25
way
31:25
more
31:25
quickly.
31:26
And
31:26
I
31:26
think
31:26
that's,
31:26
to
31:27
me,
31:27
the
31:27
danger.
31:28
Like,
31:28
one
31:28
of
31:28
the
31:28
worst
31:29
things
31:29
I
31:29
think
31:29
you
31:29
can
31:29
do
31:30
early
31:30
days
31:30
is
31:30
just
31:31
read
31:31
the
31:32
here's
31:32
100
31:32
props
31:34
documents,
31:35
because
31:36
they
31:36
actually
31:36
really
31:37
limit
31:37
you.
31:37
It
31:38
feels
31:38
like
31:38
you're
31:38
about
31:38
to
31:38
get
31:38
empowered,
31:39
but,
31:39
like,
31:39
your
31:39
problems
31:40
are
31:40
probably
31:40
way
31:40
more
31:41
specific
31:41
and
31:41
there's
31:41
things
31:42
that
31:42
it
31:42
could
31:42
do
31:42
for
31:42
you,
31:44
and
31:45
then
31:45
it's
31:45
sort
31:45
of
31:45
like
31:46
anything
31:46
else.
31:46
So
31:46
you're
31:46
going
31:46
through
31:47
this
31:47
process.
31:47
It
31:48
seems
31:48
to
31:48
have
31:48
absolutely
31:49
no
31:49
limit.
31:50
It
31:50
can
31:50
do
31:50
everything.
31:51
But
31:51
then
31:51
when
31:51
you
31:52
really,
31:52
really
31:52
stop
31:53
getting
31:53
amazed
31:54
by
31:54
it,
31:54
you
31:55
realize
31:55
it's
31:55
very,
31:55
very
31:56
bad
31:56
at
31:56
a
31:57
lot
31:57
of
31:57
things
31:57
that
31:57
it
31:57
appears
31:58
to
31:58
be
31:58
doing
31:58
really,
31:59
really
31:59
well.
31:59
Yes,
31:59
100%.
32:02
It'll
32:02
make
32:02
you
32:02
feel
32:02
Good
32:02
about
32:03
it.
32:03
It
32:04
does
32:04
for
32:04
a
32:04
minute,
32:05
but
32:05
then
32:05
you,
32:06
you
32:07
know,
32:07
you
32:07
start
32:07
to
32:07
get
32:07
emails
32:08
that
32:08
are
32:08
written
32:08
in
32:08
AI
32:09
or
32:09
you
32:09
talk
32:09
to
32:09
professors
32:10
who
32:10
are
32:10
getting
32:10
AI
32:11
papers
32:11
and,
32:11
like,
32:11
everybody's
32:12
just,
32:12
like,
32:13
fried.
32:13
They're
32:14
just
32:14
kind
32:14
of
32:14
fried
32:14
by
32:14
it
32:15
because
32:15
it's.
32:16
It.
32:17
It's
32:17
like
32:17
bad
32:18
CGI
32:18
art
32:19
from
32:19
back
32:19
in
32:19
the
32:20
day.
32:20
It
32:20
looks
32:21
so
32:21
neat
32:21
at
32:21
the
32:21
time.
32:22
Yeah.
32:23
It's
32:23
like
32:23
the
32:23
uncanny
32:24
valley
32:24
of
32:24
AI.
32:25
Yeah.
32:25
The
32:26
analogy
32:26
I
32:26
think
32:26
about
32:27
a
32:27
lot
32:27
is,
32:27
like,
32:28
when,
32:29
you
32:30
know,
32:31
things
32:31
like
32:31
WordPress
32:32
or
32:32
Squarespace
32:33
or
32:33
like,
32:34
Canva
32:35
enabled
32:36
people
32:36
to,
32:37
like,
32:37
normies
32:38
like
32:38
me
32:39
to
32:40
create,
32:40
like,
32:41
websites
32:42
that
32:43
look
32:44
professionally
32:44
designed
32:45
or
32:45
graphics
32:46
that
32:46
look
32:46
professionally
32:46
designed.
32:47
But
32:49
it
32:50
also
32:50
sort
32:51
of
32:52
like
32:53
smushes
32:54
down
32:55
the
32:55
variety
32:56
of
32:57
creativity
32:57
of
32:58
things
32:58
that
32:58
you
32:58
get
32:58
in
32:59
the
32:59
output
32:59
there.
33:00
But
33:00
at
33:00
the
33:00
same
33:00
time,
33:00
like,
33:01
I
33:01
think
33:01
about
33:02
Sarah,
33:02
your
33:02
company
33:03
was
33:03
called.
33:03
Was
33:03
it
33:03
Juulbots?
33:04
Was
33:04
that
33:04
right?
33:04
Yeah.
33:05
Like,
33:05
Juulbots
33:06
actually
33:06
shouldn't
33:07
have
33:07
too
33:07
creative
33:08
of
33:08
a
33:08
website
33:09
because
33:09
you
33:09
don't
33:09
want
33:09
to
33:09
get
33:09
in
33:10
the
33:10
way
33:10
of
33:10
the
33:10
jewel
33:10
bots.
33:11
That's
33:11
a
33:12
good
33:12
point.
33:12
Right,
33:12
right.
33:13
So,
33:13
like,
33:13
I
33:13
actually
33:14
think
33:14
the
33:14
generic
33:15
stuff
33:15
is
33:15
totally
33:16
fine.
33:16
It's
33:16
just
33:17
that
33:17
you're
33:17
making
33:17
a
33:18
choice,
33:18
that's
33:18
all.
33:18
Like,
33:19
I
33:19
think
33:19
people
33:19
just
33:19
often
33:20
don't
33:20
know
33:20
they're
33:20
making
33:21
a
33:21
choice.
33:22
And
33:22
then
33:22
a
33:22
lot
33:22
of
33:22
times,
33:23
I
33:23
think
33:23
what
33:23
is
33:23
tricky
33:24
too
33:24
is
33:24
all
33:24
the
33:24
marketing,
33:25
all
33:25
the
33:25
communication.
33:25
It's
33:25
like,
33:26
Squarespace,
33:26
you're
33:27
now
33:27
a
33:27
total
33:27
web
33:27
genius.
33:28
And
33:28
then
33:28
people
33:28
are
33:29
like,
33:29
cool,
33:29
when
33:30
do
33:30
I
33:30
get
33:30
all
33:30
my
33:30
money?
33:31
When
33:31
do
33:31
I
33:31
get
33:32
my
33:32
Internet
33:32
10
33:32
engineers?
33:33
Yeah,
33:34
yeah.
33:34
And
33:35
I
33:35
think
33:35
we're
33:35
seeing
33:35
that
33:35
with
33:35
AI
33:36
now.
33:36
Right.
33:36
And
33:36
I
33:36
think
33:36
we're
33:36
going
33:37
to
33:37
see
33:37
more
33:37
and
33:37
more
33:37
of
33:37
it.
33:37
I'll
33:38
talk
33:38
to
33:38
somebody
33:38
at
33:38
like,
33:39
a
33:39
Not
33:39
for
33:39
Profit,
33:39
and
33:40
they'll
33:40
be
33:40
like,
33:40
I
33:40
have
33:40
this
33:40
one
33:41
report
33:41
that
33:42
I
33:42
really
33:42
need
33:42
to
33:43
have
33:43
run
33:44
every
33:44
six
33:45
months,
33:45
and
33:45
it
33:45
takes
33:46
me
33:46
two
33:46
weeks
33:46
to
33:46
cut
33:46
and
33:47
paste
33:47
the
33:47
data,
33:47
even
33:47
though
33:48
it's
33:48
already
33:48
in
33:48
Salesforce,
33:49
because
33:50
our
33:50
people
33:51
don't
33:51
have
33:51
the
33:51
resources
33:52
to
33:54
run
33:54
the
33:54
report
33:54
generator.
33:55
Like,
33:55
we're
33:55
kind
33:55
of
33:55
out
33:55
of
33:55
money
33:55
for
33:56
this
33:56
kind
33:56
of
33:56
problem.
33:57
And
33:57
I'm
33:58
like,
33:58
boy,
33:58
is
33:59
that
33:59
a
33:59
perfect
33:59
candidate.
34:00
Like,
34:00
if
34:01
you
34:01
had
34:01
a
34:01
nice
34:02
LLM
34:02
interface
34:03
of
34:03
that,
34:03
you
34:03
could
34:03
say,
34:03
go
34:03
get
34:04
this
34:04
data,
34:05
extract
34:05
it,
34:06
make
34:06
it
34:06
into
34:06
a
34:06
nice
34:06
report,
34:07
and
34:07
do
34:07
that
34:07
every
34:08
six
34:08
months.
34:08
And
34:08
you
34:09
can
34:09
kind
34:09
of
34:09
see
34:09
if
34:09
you've
34:09
used
34:10
these
34:10
tools,
34:11
why
34:11
that
34:11
is
34:11
a
34:11
relatively
34:12
low
34:12
risk
34:12
proposition.
34:13
Like,
34:13
it'll
34:13
probably
34:14
work.
34:14
If
34:14
you
34:15
have
34:15
the
34:15
right
34:15
sort
34:16
of
34:16
a
34:16
little
34:16
bit
34:17
of
34:17
coaching,
34:17
that
34:17
problem
34:18
goes
34:18
away
34:18
and
34:18
it's
34:19
now
34:19
like
34:19
a
34:19
three
34:19
week
34:20
problem.
34:20
And
34:20
no
34:20
one
34:20
will
34:20
find
34:21
the
34:21
three
34:21
weeks.
34:21
And
34:21
that
34:21
problem
34:22
goes
34:22
away
34:22
in
34:22
like
34:22
two
34:22
hours
34:23
forever.
34:24
And
34:24
so
34:24
that,
34:25
to
34:25
me,
34:25
everybody
34:25
can
34:26
have
34:26
their
34:26
thing
34:26
that'll
34:27
be
34:27
a
34:27
totally
34:27
generic
34:27
report.
34:28
Nobody
34:28
will
34:28
care
34:28
about
34:29
it.
34:29
And
34:30
there's
34:30
literally
34:31
thousands
34:31
of
34:31
things
34:32
like
34:32
that
34:32
in
34:32
every
34:33
organization
34:33
that
34:33
everybody's
34:34
been
34:34
told
34:34
they
34:34
can't
34:34
have.
34:35
So
34:35
that
34:35
brings
34:35
me
34:36
to
34:36
an
34:36
interesting
34:36
question,
34:37
like
34:37
talking
34:38
about
34:38
like
34:38
the
34:38
CANVA
34:39
model.
34:40
So
34:40
you're
34:40
kind
34:40
of
34:40
like
34:41
building
34:41
something
34:41
on
34:41
top
34:41
of
34:42
AI,
34:42
which
34:42
is
34:42
a
34:42
little
34:42
bit
34:43
the
34:43
CANVA
34:43
model
34:44
of
34:44
like.
34:44
Okay,
34:45
I
34:45
heard
34:45
you
34:45
describe
34:46
it
34:46
like
34:46
twice
34:46
now
34:47
as
34:47
like
34:47
old
34:47
business
34:47
person
34:48
problem.
34:48
Like
34:49
you're
34:49
targeting,
34:50
I
34:50
imagine
34:51
people
34:51
that
34:51
aren't
34:52
just
34:52
gonna,
34:52
like,
34:53
aren't
34:53
jumping
34:53
in
34:54
cloud
34:54
code,
34:54
like,
34:55
haven't
34:55
done
34:55
that
34:56
yet.
34:56
But
34:57
what,
34:57
like,
34:58
where
34:58
do
34:58
you
34:59
decide?
34:59
Because
34:59
there's
35:00
like
35:00
such
35:00
a
35:00
spectrum
35:01
of
35:01
like,
35:01
okay,
35:01
we'll
35:02
do
35:02
everything
35:02
for
35:03
you,
35:03
or
35:03
like,
35:03
we'll
35:04
do
35:04
these
35:04
two
35:04
specific
35:05
things
35:05
for
35:06
you.
35:06
How
35:06
did
35:06
you
35:06
all
35:06
decide
35:07
and
35:07
how
35:07
did
35:07
you
35:07
pick
35:07
the
35:08
technology
35:08
to
35:09
support
35:09
that?
35:10
Well,
35:10
I
35:10
mean,
35:11
I
35:11
used
35:11
to
35:11
run
35:12
an
35:12
agency.
35:12
I
35:12
did
35:12
for
35:13
many
35:13
years,
35:13
and
35:13
I
35:13
was
35:13
always
35:13
in
35:14
services
35:14
before
35:14
that.
35:15
And
35:15
we
35:15
tried.
35:15
We
35:16
took
35:16
our
35:16
consumer
35:16
swing.
35:17
We
35:17
actually
35:17
wrote
35:18
the
35:18
credit
35:18
card
35:19
processing
35:20
function
35:21
for
35:22
the
35:22
older
35:22
version
35:22
of
35:23
our
35:23
product.
35:23
And
35:23
then
35:23
we
35:24
went,
35:24
what
35:24
are
35:25
we
35:25
doing?
35:25
Like,
35:26
right.
35:26
Like,
35:28
we
35:28
know
35:28
how
35:28
to
35:29
build
35:29
complicated
35:30
things
35:30
for
35:30
complicated
35:31
organizations.
35:31
Like
35:31
if
35:32
you're
35:32
in
35:32
New
35:32
York
35:32
City
35:33
and
35:33
you're
35:33
waiting
35:34
for
35:34
the
35:34
train
35:34
and
35:35
it
35:35
says
35:35
it's
35:35
three
35:36
minutes
35:36
away
35:36
on
35:36
that
35:36
sign,
35:37
that
35:37
was
35:38
something
35:38
we
35:38
helped
35:38
build.
35:39
Like
35:39
we
35:39
helped
35:39
build
35:39
and
35:40
deploy
35:40
that.
35:41
Great
35:42
fun
35:42
fact.
35:43
Yeah,
35:44
that's
35:45
kind
35:45
of
35:45
where
35:45
we
35:46
are
35:46
and
35:46
what
35:46
we
35:47
understand.
35:47
And
35:47
then
35:47
there
35:47
was
35:48
another
35:48
aspect
35:48
which
35:48
is
35:49
we
35:49
have
35:49
an
35:50
office
35:50
here,
35:50
which
35:50
literally
35:51
I
35:51
see
35:51
New
35:51
York
35:52
City
35:52
kind
35:52
of
35:52
getting
35:52
eaten
35:53
by
35:53
AI
35:53
and
35:53
not
35:54
realizing
35:54
it
35:55
because
35:56
the
35:56
Valley
35:56
has
35:57
once
35:57
again
35:57
figured
35:58
it
35:58
out
35:58
and
35:58
is
35:59
going
35:59
to
35:59
kind
35:59
of
35:59
hold
35:59
all
36:00
the
36:00
value
36:00
for
36:00
itself.
36:00
And
36:01
I
36:01
think
36:01
about
36:02
big
36:02
organizations
36:03
here
36:03
and
36:03
industries
36:04
here
36:04
where
36:04
it
36:04
really
36:05
is
36:05
going
36:05
to
36:05
be
36:05
fundamental.
36:05
It's
36:06
going
36:06
to
36:06
be
36:06
fundamental
36:06
for
36:06
publishing,
36:07
it's
36:07
going
36:07
to
36:07
be
36:07
fundamental
36:08
for
36:08
finance.
36:08
NGOs
36:09
in
36:09
particular
36:09
are
36:10
a
36:10
huge
36:10
sector
36:10
here
36:10
in
36:11
government
36:11
and
36:12
they're
36:12
not
36:12
ready
36:13
for
36:13
what
36:14
could
36:14
happen.
36:15
A
36:15
good
36:15
example
36:16
would
36:16
be,
36:16
like,
36:16
all
36:16
legacy
36:17
code
36:18
can
36:18
be
36:18
processed
36:19
a
36:19
lot
36:19
more
36:19
quickly.
36:19
And
36:20
so
36:20
that
36:20
system
36:20
that
36:21
like,
36:21
you
36:22
won't
36:22
touch
36:22
In
36:23
Cobol
36:23
that's
36:24
45
36:24
years
36:24
old
36:25
is
36:26
now,
36:26
you
36:26
could
36:27
go
36:27
back
36:27
to
36:27
that
36:27
again.
36:28
It
36:28
might
36:28
turn
36:28
out
36:28
that
36:28
it's
36:29
1/10
36:29
the
36:29
budget
36:29
of
36:30
what
36:30
it
36:30
used
36:30
to
36:30
be.
36:30
And
36:30
so,
36:30
like,
36:31
I
36:31
want
36:31
to
36:31
make
36:31
a
36:32
space
36:32
for
36:32
that.
36:32
And
36:33
I
36:33
want
36:33
to
36:33
help
36:34
organizations
36:35
because
36:35
I
36:35
think
36:35
that
36:36
they
36:36
have
36:36
the
36:36
most
36:37
impact,
36:38
kind
36:38
of
36:38
do
36:38
really
36:39
complicated
36:39
things.
36:40
And
36:40
then
36:40
what
36:40
we
36:41
realized,
36:41
actually,
36:41
look,
36:42
if
36:42
I'd
36:42
found
36:42
a
36:42
way
36:43
that
36:43
I
36:43
could
36:43
spin
36:43
up
36:43
a
36:43
true
36:44
business
36:44
application
36:45
by
36:45
pushing
36:45
a
36:46
button
36:46
and
36:46
doing
36:46
it
36:46
purely
36:47
with
36:47
prompts,
36:47
I
36:48
would
36:48
have
36:48
done
36:48
it.
36:48
Like,
36:49
I
36:49
think
36:49
that
36:49
that's
36:50
coming.
36:50
Like,
36:51
if
36:51
that,
36:51
if
36:51
that
36:51
could
36:52
happen,
36:52
I'd
36:52
rather
36:53
be
36:53
the
36:53
person
36:53
to
36:53
do
36:53
it
36:53
than
36:54
somebody
36:54
else.
36:54
But
36:55
I
36:55
think
36:55
there's
36:55
just
36:55
hard,
36:56
hard
36:56
limits
36:57
to
36:57
where
36:57
you
36:57
can
36:58
get
36:58
by
36:59
typing
36:59
words
37:00
into
37:00
the
37:00
box
37:00
and
37:01
waiting
37:01
for
37:01
the
37:01
LLM
37:02
to
37:02
figure
37:02
out
37:02
what
37:02
you
37:03
meant.
37:03
And
37:04
so
37:05
what
37:05
are,
37:05
what
37:05
are
37:06
those
37:06
limits
37:07
right
37:07
now?
37:09
That
37:09
is
37:09
a
37:09
very
37:09
good
37:10
question.
37:10
I
37:10
think
37:11
that
37:12
subtlety
37:13
is
37:13
hard,
37:13
right?
37:14
And
37:14
so
37:14
like
37:15
certain
37:15
things
37:16
that
37:16
you
37:16
would
37:16
think
37:16
would
37:16
be
37:17
really
37:17
complicated,
37:17
like
37:18
many
37:18
to
37:18
many
37:18
relationships
37:19
and
37:19
databases
37:20
and
37:20
also
37:20
things
37:20
that
37:21
people
37:21
actually
37:21
tell
37:21
us
37:22
will
37:22
be
37:22
really
37:22
hard
37:22
are
37:23
not
37:23
that
37:23
hard
37:24
because
37:24
you
37:24
figure
37:24
out
37:24
the
37:24
pattern
37:25
and
37:25
it's
37:25
hard.
37:25
And
37:26
it
37:26
might
37:26
take
37:26
you
37:26
a
37:26
couple
37:26
weeks,
37:27
but
37:27
then
37:27
you
37:27
figure
37:27
out
37:27
how
37:28
to
37:28
like
37:28
describe
37:28
it
37:29
and
37:29
then
37:29
it
37:29
does
37:30
it
37:30
every
37:30
time
37:30
thereafter.
37:31
And
37:31
that's
37:31
very
37:31
much
37:32
like
37:32
programming.
37:32
I
37:32
think
37:34
all
37:34
aspects
37:35
of
37:35
subtle
37:35
user
37:36
interface
37:36
change
37:37
still
37:37
require
37:38
human.
37:39
You
37:39
can't
37:39
just
37:39
sort
37:39
of
37:39
be
37:39
like,
37:40
apply
37:40
a
37:40
theme.
37:40
Good
37:41
luck.
37:41
Complicated
37:42
issues
37:42
of
37:42
migration,
37:43
legacy
37:44
data,
37:44
external
37:45
API
37:45
integration,
37:46
things
37:47
like
37:47
that.
37:47
Like
37:47
sort
37:48
of
37:48
the
37:48
ugly
37:48
things
37:48
that
37:48
make
37:49
a
37:49
business
37:49
a
37:49
business.
37:50
We
37:51
get
37:51
this
37:51
data
37:51
and
37:52
it
37:52
comes
37:52
to
37:52
us
37:52
in
37:52
a
37:53
spreadsheet
37:53
every
37:53
Tuesday
37:54
and
37:54
so
37:54
on.
37:54
Like
37:55
working
37:55
around
37:55
those
37:56
processes,
37:56
you
37:56
can
37:57
automate
37:57
and
37:57
accelerate
37:58
a
37:58
lot
37:58
of
37:58
them,
37:58
but
37:58
they
37:59
just
37:59
don't
37:59
come
37:59
out
37:59
of
38:00
the
38:00
box.
38:01
By
38:01
asking
38:02
an
38:02
LLM
38:02
to
38:03
make
38:03
sense
38:03
of
38:03
them,
38:03
it
38:04
sure
38:04
can
38:04
help.
38:04
Like,
38:05
I
38:05
do
38:05
think
38:05
there's
38:06
probably
38:06
like
38:06
3.3x
38:07
acceleration
38:07
on
38:07
stuff
38:08
like
38:08
that,
38:09
and
38:09
eventually
38:09
more
38:10
and
38:10
more
38:10
of
38:10
it
38:10
will
38:10
fall.
38:11
Like,
38:11
we're
38:11
not
38:11
done
38:11
yet,
38:12
but
38:12
as
38:12
of
38:13
today,
38:13
I
38:14
think
38:14
you
38:14
can
38:14
get
38:14
a
38:15
really
38:15
great
38:15
working
38:16
prototype
38:16
with
38:16
real
38:17
data.
38:18
I
38:18
think
38:18
you
38:18
can
38:18
get
38:19
a
38:19
certain
38:19
amount
38:20
of
38:20
software
38:20
that
38:20
is
38:21
pretty
38:21
useful
38:22
that
38:22
you
38:22
can
38:22
show
38:22
to
38:22
your
38:23
team
38:23
that
38:23
people
38:23
can
38:23
try.
38:24
And
38:25
then
38:25
I
38:25
think
38:25
it
38:26
really
38:26
involves
38:27
some
38:27
conversation.
38:28
Kind
38:29
of
38:29
like
38:29
a
38:29
hello
38:29
world.
38:30
It
38:30
seems
38:30
like
38:30
it's
38:30
in
38:30
the
38:31
hello
38:31
world
38:31
phase
38:31
of
38:32
like
38:32
you
38:32
can
38:32
make
38:32
a
38:33
demo,
38:33
but
38:33
then
38:33
getting
38:33
past
38:34
that
38:34
or
38:34
in
38:34
some
38:34
more
38:35
complicated
38:35
use
38:35
cases
38:36
gets.
38:36
You
38:36
have
38:36
to
38:36
intervene
38:37
more.
38:37
And
38:37
on
38:38
the
38:38
level
38:38
you
38:38
might
38:38
not
38:38
be
38:38
used
38:39
to.
38:39
Well,
38:40
if
38:40
I
38:40
can
38:40
get
38:40
just
38:40
like
38:40
a
38:41
tiny
38:41
bit
38:41
meta.
38:43
The
38:43
concept,
38:44
not
38:44
the
38:44
company,
38:46
we'll
38:46
allow
38:46
it.
38:47
I
38:48
think
38:48
like
38:48
software
38:49
is
38:49
a
38:49
cultural
38:49
artifact
38:50
that
38:50
represents
38:51
like
38:51
the
38:51
Needs
38:51
of
38:51
a
38:52
community
38:52
at
38:52
its
38:53
best.
38:53
Right.
38:53
Sometimes
38:54
they
38:54
just
38:54
jam
38:55
Salesforce
38:55
in
38:55
and
38:55
they
38:56
say
38:56
good
38:57
luck.
38:57
But
38:57
for
38:57
the
38:57
most
38:58
part,
38:58
like
38:58
software
38:58
evolves
38:59
out
38:59
of
38:59
what
38:59
people
38:59
need
39:00
and
39:00
want.
39:01
And
39:02
how
39:02
could
39:02
you
39:03
do
39:03
that
39:03
just
39:03
talking
39:04
with
39:04
an
39:04
LLM
39:05
by
39:06
yourself
39:06
in
39:06
a
39:06
room?
39:07
Like,
39:07
even
39:07
if
39:07
the
39:08
LLM
39:08
writes
39:08
most
39:09
of
39:09
the
39:09
code
39:09
five
39:10
years
39:10
from
39:10
now,
39:11
you're
39:11
still
39:11
going
39:11
to
39:12
have
39:12
to
39:12
figure
39:12
out
39:12
what
39:12
you
39:13
need.
39:14
And
39:15
maybe
39:15
the
39:15
future
39:15
is
39:16
I
39:16
just
39:16
kind
39:16
of
39:17
lie
39:17
on
39:17
the
39:17
ground
39:17
and
39:18
talk
39:18
into
39:18
my
39:19
phone
39:19
and
39:19
say,
39:19
make
39:19
me
39:20
an
39:20
app.
39:20
But
39:20
I
39:21
don't
39:21
think
39:21
so
39:22
because
39:22
it
39:22
turns
39:23
out
39:23
that
39:23
all
39:23
the
39:23
margins
39:24
and
39:24
everything
39:24
that
39:25
people
39:25
do
39:25
is
39:25
actually
39:25
kind
39:26
of
39:26
subtle.
39:26
Most
39:26
things
39:27
are
39:27
not
39:27
simple,
39:28
especially
39:28
in
39:28
business.
39:29
And
39:30
so
39:31
what
39:31
do
39:31
you
39:31
mean
39:31
by
39:31
that?
39:32
That
39:32
like
39:32
the
39:32
margins
39:33
are
39:33
subtle?
39:33
Like,
39:34
can
39:34
you
39:34
give
39:34
me
39:34
an
39:34
example?
39:36
Yes.
39:37
What
39:38
I'm
39:38
actually
39:38
saying
39:38
is
39:38
that
39:39
a
39:39
lot
39:39
of
39:40
businesses
39:40
look
39:41
alike
39:41
and
39:41
a
39:41
lot
39:41
of,
39:43
I
39:43
mean,
39:44
you
39:44
know,
39:44
the
39:44
difference
39:44
between
39:45
like
39:46
two
39:46
moving
39:47
companies
39:47
and
39:48
one,
39:48
one
39:48
moving
39:48
company
39:49
is
39:50
all
39:50
about
39:51
volume.
39:51
Like
39:52
they
39:52
advertise
39:52
everywhere,
39:53
they
39:53
spend
39:53
a
39:53
ton
39:53
of
39:54
money
39:54
on
39:54
marketing,
39:55
they
39:55
analyze
39:56
their
39:56
funnel
39:56
really
39:57
carefully
39:57
and
39:58
then
39:58
they
39:59
close
39:59
as
40:00
much
40:00
business
40:00
as
40:00
possible.
40:01
And
40:01
they
40:01
have
40:01
a
40:01
lot
40:01
of
40:01
trucks
40:02
and
40:02
they
40:02
have
40:02
an
40:02
ad
40:02
hoc
40:02
network
40:03
of
40:03
trucks
40:03
that
40:03
shows
40:04
up
40:04
and
40:04
blah,
40:04
blah,
40:04
blah,
40:04
blah,
40:04
blah.
40:05
And
40:05
then
40:05
the
40:05
other
40:05
one
40:05
is
40:06
really
40:06
high
40:06
end
40:06
and
40:06
they
40:07
market
40:07
much
40:07
more
40:07
strategically.
40:08
They
40:08
focus
40:08
on
40:09
moving
40:09
things
40:09
like
40:09
art,
40:10
and
40:11
they're
40:11
going
40:12
to
40:12
charge
40:12
a
40:13
pretty
40:13
serious
40:13
premium.
40:14
They'll
40:14
make
40:14
way
40:15
more
40:15
money
40:15
per
40:15
engagement,
40:16
but
40:17
they
40:17
won't
40:17
have
40:17
anywhere
40:17
near
40:18
the
40:18
volume
40:18
of
40:18
the
40:18
first
40:18
one
40:19
because
40:19
they
40:20
kind
40:20
of
40:20
can't
40:20
like
40:20
the
40:20
market
40:21
won't
40:21
bear
40:21
it.
40:21
And
40:22
over
40:22
20
40:22
years
40:22
these
40:23
have
40:23
evolved
40:23
into
40:24
one
40:24
person
40:25
could
40:25
go
40:25
and
40:25
work
40:25
at
40:26
the
40:26
other,
40:27
but
40:27
it
40:27
might
40:27
not
40:27
actually
40:28
succeed
40:28
organically.
40:29
It's
40:29
like
40:29
when
40:29
somebody
40:29
comes
40:30
from
40:30
Google
40:30
and
40:30
comes
40:31
to
40:31
another
40:31
startup,
40:32
which
40:32
I've
40:32
seen
40:32
happen
40:32
like
40:33
30
40:33
times,
40:34
and
40:34
they're
40:34
just
40:34
like
40:35
mutants
40:35
at
40:36
that
40:36
point.
40:36
Like
40:36
they're
40:37
just
40:37
like,
40:37
where
40:37
is
40:38
my
40:38
blue
40:38
bottle
40:38
coffee
40:39
and
40:39
my
40:39
enormous
40:40
file
40:40
system?
40:41
I
40:41
need
40:41
those
40:41
two
40:42
things
40:42
in
40:42
order
40:43
to
40:43
even,
40:44
you
40:44
know,
40:44
use
40:45
the
40:45
bathroom
40:45
at
40:46
this
40:46
place.
40:46
And
40:46
you're
40:47
like,
40:47
oh
40:48
crap,
40:48
like
40:48
we
40:48
don't
40:49
know
40:49
what
40:49
to
40:49
do
40:49
with
40:49
you.
40:49
And
40:50
everybody
40:50
thought
40:50
it
40:50
would
40:50
be
40:50
a
40:51
miracle,
40:51
like
40:51
the
40:51
giant
40:52
company
40:52
would
40:52
bring
40:53
its
40:53
wisdom
40:54
and
40:54
they
40:54
thought
40:54
it
40:55
was
40:55
going
40:55
to
40:55
be
40:55
great
40:55
and
40:56
then
40:56
it's
40:56
just,
40:56
it's
40:57
a
40:57
disaster.
40:57
And
40:58
so
40:58
like
40:59
it's
40:59
that
41:00
all
41:00
the
41:00
margins
41:01
and
41:01
all
41:01
the
41:01
things
41:01
that
41:02
make
41:02
a
41:02
business
41:02
special
41:02
and
41:03
unique
41:03
and
41:04
that's
41:04
how
41:04
they
41:04
make
41:04
their
41:05
money
41:05
are
41:06
really
41:06
hard.
41:06
A
41:06
lot
41:07
of
41:07
times
41:07
people
41:07
can't
41:07
Even
41:07
articulate
41:08
what
41:08
they
41:08
are.
41:09
I
41:09
see
41:09
what
41:09
you're
41:09
saying.
41:10
So
41:10
it's
41:10
not.
41:10
You
41:11
can't
41:11
just
41:11
like
41:11
copy
41:12
and
41:12
paste
41:13
a
41:13
prompt
41:14
to
41:15
generate
41:16
an
41:17
application
41:18
that
41:18
works
41:18
for
41:19
every
41:19
moving
41:20
company
41:20
business?
41:21
No,
41:21
there's
41:21
a
41:22
lot
41:22
of
41:22
nuance
41:22
or
41:23
context
41:24
that
41:24
the
41:24
model
41:25
needs
41:25
in
41:25
order
41:25
to
41:26
do
41:26
it
41:26
effectively
41:26
for
41:26
your
41:27
business.
41:27
But
41:27
humans
41:28
aren't
41:28
used
41:29
to
41:29
or
41:29
really
41:30
good
41:30
at
41:30
describing
41:31
that
41:31
nuance
41:32
yet.
41:32
So
41:32
look,
41:33
we're
41:33
launching
41:34
something
41:34
that
41:34
tries
41:34
to
41:35
help
41:35
with
41:35
that.
41:35
But
41:35
the
41:36
way
41:36
it
41:36
works
41:36
is
41:36
that
41:36
you
41:37
might
41:37
say,
41:37
I
41:37
need
41:38
a
41:38
customer
41:39
relationship
41:39
management
41:39
system
41:40
for
41:40
my
41:40
moving
41:40
company.
41:42
Now
41:42
the
41:42
second
41:43
sentence
41:43
is
41:43
the
41:44
really
41:44
important
41:44
one.
41:44
And
41:45
the
41:45
second
41:45
sentence,
41:45
because
41:46
it'll
41:46
just
41:46
build,
41:46
you
41:46
can
41:47
hit
41:47
return,
41:47
answer
41:47
a
41:47
couple
41:48
questions,
41:48
it'll
41:48
build
41:48
you
41:48
that.
41:49
But
41:49
the
41:50
second
41:50
statement
41:50
is
41:50
like,
41:51
we
41:51
move
41:52
a
41:52
lot
41:52
of
41:52
really
41:52
high
41:53
end
41:53
art
41:54
and
41:54
I
41:54
need
41:54
a
41:55
way
41:55
to
41:55
track
41:55
each
41:56
individual
41:56
piece
41:57
of
41:57
art
41:57
and
41:57
track
41:58
the
41:58
insurance
41:59
on
42:00
each
42:00
individual
42:00
piece
42:01
of
42:01
art.
42:01
I'll
42:02
need
42:02
to
42:02
upload
42:02
a
42:03
lot
42:03
of
42:03
PDFs
42:03
and
42:04
documentation
42:05
per
42:05
object,
42:06
right?
42:07
And
42:07
they'll
42:08
come
42:08
to
42:08
you.
42:09
Like
42:09
the
42:09
way
42:09
that
42:09
SAS
42:10
works
42:10
and
42:10
the
42:10
way
42:10
that
42:10
classic
42:11
software
42:11
works
42:11
is
42:12
like
42:12
these
42:12
people
42:13
tend
42:13
to
42:13
be,
42:13
when
42:14
you
42:14
talk
42:14
to
42:14
them,
42:14
they're
42:14
like,
42:15
there
42:15
is
42:15
absolutely
42:16
no
42:16
way
42:17
that
42:17
I
42:18
can
42:18
use
42:18
Salesforce
42:19
or
42:19
I
42:19
can
42:19
use
42:20
pipedrive
42:20
or
42:21
whatever
42:21
to
42:21
build
42:22
this
42:22
because
42:22
I
42:22
need
42:23
that.
42:23
That
42:23
custom
42:24
insurance
42:25
PDF
42:26
upload
42:26
function
42:26
is
42:27
my
42:27
entire
42:27
business.
42:28
Without
42:28
it,
42:28
I
42:29
might
42:29
as
42:29
well
42:29
just
42:29
stick
42:29
with
42:29
Google
42:30
spreadsheets
42:31
and
42:31
the
42:32
Z
42:32
Drive
42:32
or
42:33
Dropbox.
42:34
And
42:34
I
42:35
do
42:35
think
42:36
they
42:36
can
42:37
start
42:37
to
42:37
have
42:37
more
42:37
and
42:38
more
42:38
of
42:38
their
42:38
stuff.
42:39
And
42:39
so
42:39
they
42:39
can
42:40
have
42:40
really
42:40
good
42:40
custom
42:40
software.
42:42
That's
42:42
what
42:42
we're
42:42
trying
42:42
to
42:42
solve.
42:44
And
42:44
you
42:44
can
42:44
demo
42:45
it
42:45
on
42:45
the
42:45
website,
42:46
but
42:47
they
42:47
can
42:47
have
42:48
really
42:48
good
42:48
custom
42:48
software.
42:49
But
42:50
then
42:50
getting
42:51
that
42:51
exactly
42:51
right
42:52
for
42:52
them,
42:52
taking
42:52
that,
42:53
and
42:53
that
42:54
doesn't
42:54
exist,
42:55
that
42:55
insurance
42:55
form
42:56
upload
42:56
tool
42:57
doesn't
42:57
exist
42:57
in
42:57
any
42:58
off
42:58
the
42:58
shelf
42:58
thing.
42:59
So
42:59
now
42:59
the
42:59
AI
42:59
assembles
43:00
it
43:00
and
43:00
it
43:01
looks
43:01
pretty
43:01
good
43:01
and
43:01
it
43:02
kind
43:02
of
43:02
works,
43:02
but
43:03
there's
43:03
always
43:03
going
43:03
to
43:03
be
43:03
a
43:03
tweak,
43:04
like,
43:04
well,
43:04
I
43:04
really
43:04
want
43:04
to
43:05
track
43:05
the
43:05
name
43:06
of
43:06
the
43:06
insurance
43:06
firm.
43:06
Or
43:07
actually
43:08
we
43:08
work
43:08
with
43:09
like
43:09
three
43:09
insurers
43:09
of
43:10
record
43:10
and
43:10
one
43:10
is
43:10
Sally
43:11
and
43:11
one
43:11
is
43:11
Sam.
43:12
And
43:12
I
43:12
need
43:12
to
43:12
like,
43:12
you
43:12
know,
43:13
the
43:13
Sally
43:13
and
43:13
the
43:13
Sam
43:14
things
43:14
actually
43:14
have
43:14
to
43:14
be
43:14
handled
43:15
differently
43:15
because
43:15
of
43:15
billing
43:16
and
43:16
it
43:17
always
43:17
goes
43:17
like
43:17
that
43:18
and,
43:19
and
43:19
there's
43:19
kind
43:19
of
43:19
nothing
43:20
off
43:20
the
43:20
shelf.
43:20
And
43:21
I
43:21
think
43:21
you
43:21
can,
43:21
you,
43:22
you're
43:22
going
43:22
to
43:22
be
43:22
able
43:22
to
43:22
tell
43:22
the
43:23
bot
43:23
to
43:23
a
43:23
certain
43:23
point,
43:23
but
43:24
at
43:24
another
43:24
point
43:24
I
43:25
think
43:25
you
43:25
just,
43:25
you
43:26
just
43:26
kind
43:26
of
43:26
want
43:26
to
43:26
talk
43:26
to
43:26
a
43:27
person
43:27
like.
43:27
Like
43:28
one
43:28
of
43:28
you
43:28
two?
43:29
Yeah,
43:30
yeah,
43:30
I
43:30
think
43:31
that
43:31
makes
43:31
a
43:31
lot
43:31
of
43:31
sense
43:32
because
43:32
a
43:32
lot
43:32
of
43:32
people
43:32
don't
43:32
know
43:33
about
43:33
that
43:33
second
43:33
sentence.
43:34
Like,
43:34
they
43:34
don't
43:35
know.
43:35
And
43:35
so,
43:35
like,
43:36
guiding
43:36
them
43:36
through,
43:36
I
43:37
think
43:37
makes
43:37
a
43:37
lot
43:37
of
43:37
sense.
43:38
What
43:38
are
43:38
you
43:38
using
43:38
under
43:38
the
43:38
hood
43:39
and
43:39
how
43:39
did
43:39
you
43:40
decide?
43:40
Are
43:40
you
43:40
using
43:40
something
43:41
open
43:41
source?
43:41
Is
43:41
it
43:42
like
43:42
llama?
43:43
Is
43:43
it.
43:43
Did
43:43
you
43:44
make
43:44
your
43:44
own?
43:44
Which
43:45
would
43:45
be
43:45
wild.
43:45
You
43:46
change
43:46
it,
43:47
do
43:47
you
43:47
change
43:48
it
43:48
up
43:48
often?
43:50
We
43:50
have
43:50
tried
43:51
all
43:51
the
43:51
different
43:51
ones
43:52
and
43:52
it's
43:52
actually,
43:53
it's
43:53
less
43:53
about
43:53
the
43:54
content
43:54
produced.
43:55
So
43:55
on
43:55
the
43:56
back
43:56
end,
43:56
the
43:56
front
43:57
end
43:57
is
43:57
what
43:57
you
43:57
would
43:58
expect.
43:58
Like,
43:58
lots
43:58
of
43:59
nice
43:59
modern
43:59
web
43:59
technologies.
44:00
On
44:00
the
44:00
back
44:01
end,
44:01
it's
44:01
a
44:01
lot
44:02
of
44:02
OpenAI
44:03
and
44:03
ChatGPT,
44:04
but
44:05
not
44:05
because
44:05
of
44:06
the
44:06
quality
44:06
of
44:06
the
44:06
LLM.
44:07
It's
44:07
because
44:08
their
44:08
JavaScript
44:09
output,
44:09
when
44:10
you
44:10
ask
44:10
them
44:10
for
44:10
data,
44:11
is
44:11
the
44:12
most
44:12
reliable
44:13
of
44:13
the
44:13
LLMs.
44:14
Claude
44:15
may
44:15
have
44:15
caught
44:16
up
44:16
by
44:16
now,
44:16
but
44:17
for
44:17
the
44:17
most
44:17
part,
44:17
the
44:17
kind
44:18
of
44:18
problems
44:18
we're
44:18
solving
44:19
are
44:19
more
44:20
about
44:20
lots
44:20
of
44:20
round
44:21
trips
44:21
and
44:22
nice,
44:23
really
44:23
cleanly
44:24
structured
44:24
JSON
44:25
output
44:25
that
44:26
we
44:26
can
44:26
parse
44:26
and
44:27
do
44:27
things
44:27
with,
44:27
and
44:28
less
44:28
about
44:29
some
44:29
sort
44:29
of
44:29
LLM
44:30
magic.
44:30
I
44:31
think
44:31
that
44:32
for
44:32
us,
44:32
it's
44:32
a
44:32
commodity,
44:33
frankly.
44:34
It's
44:34
services
44:35
on
44:35
top
44:35
of
44:35
the
44:36
commodity
44:36
is
44:36
what
44:37
matters
44:38
for
44:38
us
44:38
to
44:38
be
44:38
able
44:38
to
44:39
do
44:39
the
44:39
work.
44:40
Yeah,
44:40
that
44:40
makes
44:41
a
44:41
lot
44:41
of
44:41
sense.
44:45
I
44:46
have
44:46
a
44:46
question
44:46
that
44:46
has
44:46
nothing
44:47
to
44:47
do
44:47
with
44:47
AI,
44:49
Paul.
44:49
This
44:50
is
44:50
inspired
44:50
by
44:51
my
44:51
Nintendo
44:52
NES
44:52
controller.
44:54
What
44:54
is
44:54
your
44:56
favorite
44:58
nostalgic
44:59
technology?
45:01
Memory.
45:02
Oh,
45:02
it's
45:03
easy.
45:03
It
45:03
was
45:03
the
45:03
Amiga
45:04
1000
45:04
computer.
45:05
Absolutely
45:05
blew
45:06
my
45:06
mind.
45:06
What
45:07
is
45:07
that?
45:07
How
45:08
old
45:08
were
45:08
you,
45:08
like,
45:08
set
45:09
the
45:09
scene?
45:09
So
45:12
I
45:12
grew
45:12
up
45:12
in
45:13
a
45:13
town,
45:14
Westchester,
45:14
Pennsylvania,
45:15
that
45:15
was
45:16
just
45:17
almost
45:17
coincidentally
45:18
the
45:18
headquarters
45:19
of
45:19
Commodore
45:20
Computer
45:21
and
45:22
wasn't
45:22
a
45:22
California
45:23
company.
45:24
And
45:24
so
45:24
where
45:25
most
45:25
people
45:25
had
45:25
Apple
45:26
experiences
45:27
as
45:27
their
45:27
first,
45:28
like,
45:28
computer
45:29
education,
45:30
no,
45:30
we
45:30
had
45:30
Commodores
45:31
everywhere.
45:32
And
45:32
so
45:32
I
45:33
learned
45:33
to
45:33
type
45:33
on
45:33
a
45:33
Commodore
45:34
64
45:34
in
45:35
middle
45:35
school,
45:35
on
45:35
and
45:35
on.
45:36
And
45:36
then
45:38
Commodore
45:38
came
45:39
up
45:39
with
45:39
its
45:39
Mac
45:40
Killer.
45:40
It
45:40
was
45:40
called
45:40
the
45:40
Amiga
45:41
and
45:42
we
45:42
got
45:42
it.
45:43
I
45:43
was
45:43
maybe
45:43
12
45:43
or
45:43
13,
45:44
and
45:45
it's
45:45
still
45:46
a
45:46
computer
45:46
that
45:47
people
45:47
are
45:47
oddly,
45:47
really
45:48
passionate
45:48
about.
45:48
It
45:48
was
45:49
this
45:49
one
45:49
really
45:50
specific
45:50
moment,
45:51
had
45:51
a
45:51
lot
45:51
of
45:51
color.
45:52
It
45:52
became
45:52
something
45:53
that
45:53
could
45:53
be
45:53
very
45:53
visual.
45:54
It
45:54
was
45:54
very
45:54
creative
45:55
machine,
45:56
and
45:58
it
45:58
just
45:58
blew
45:58
my
45:58
mind.
45:59
I
45:59
learned
45:59
programming,
46:00
I
46:00
learned
46:01
photo
46:01
editing,
46:02
I
46:02
learned
46:02
how
46:03
to
46:03
write
46:03
on
46:03
it.
46:03
And
46:03
so,
46:03
like,
46:04
everything
46:04
that
46:04
kind
46:04
of
46:05
came
46:06
to
46:06
my
46:06
life
46:06
later
46:07
really
46:07
did
46:07
come
46:07
out
46:08
of
46:08
that
46:08
machine.
46:08
And
46:08
I.
46:09
I
46:09
really
46:09
thought
46:09
for
46:09
many
46:10
years
46:10
that
46:10
I
46:10
Just
46:11
was
46:11
a
46:11
big
46:11
old
46:11
nerd.
46:11
And
46:11
I'd
46:12
had
46:12
this
46:12
one
46:12
particular
46:12
experience,
46:13
but
46:13
it
46:13
actually
46:13
turns
46:14
out
46:14
that
46:14
that
46:15
box
46:15
in
46:15
particular,
46:16
the
46:16
Amiga,
46:17
the
46:17
Amiga
46:17
line,
46:17
before
46:18
I
46:18
went
46:18
out
46:18
of
46:18
business,
46:20
was
46:20
this
46:20
incredibly
46:21
transformative
46:22
kind
46:22
of
46:22
cult
46:23
like
46:23
machine,
46:23
especially
46:24
in
46:24
Europe,
46:24
and
46:24
people
46:25
just
46:26
absolutely
46:26
loved
46:26
it.
46:27
It
46:27
had
46:27
some
46:27
very
46:27
advanced
46:28
sort
46:28
of
46:28
like,
46:28
library
46:28
management
46:29
and
46:29
sort
46:29
of
46:30
really,
46:30
really
46:30
special
46:31
stuff.
46:31
And
46:31
so
46:32
that,
46:33
to
46:33
me,
46:34
is
46:34
just
46:34
a
46:34
world
46:35
of
46:35
joy.
46:35
And
46:35
I
46:35
got
46:35
really
46:36
into
46:36
retro
46:36
computing,
46:37
but
46:37
then
46:37
I
46:38
did
46:38
decide
46:38
to
46:38
put
46:39
it
46:39
away
46:39
and
46:39
I
46:39
was
46:40
like,
46:40
I
46:40
need
46:41
to
46:41
do
46:41
something
46:42
that
46:42
isn't
46:42
retro
46:43
computing.
46:45
And
46:45
then
46:45
I
46:45
got
46:45
into
46:46
synths,
46:46
which
46:46
is
46:47
just
46:47
more
46:47
retro
46:47
computing,
46:48
just
46:49
with
46:50
more
46:50
analog.
46:51
And
46:51
so
46:51
now
46:52
that's
46:52
where
46:52
most
46:53
of
46:53
my
46:54
extra
46:54
brain
46:55
goes.
46:55
At
46:56
the
46:56
time
46:57
when
46:57
you
46:58
have
46:58
your
46:58
Amiga
47:00
and
47:00
you
47:01
said
47:01
you
47:01
learned
47:01
how
47:02
to
47:02
code,
47:03
what
47:03
resource
47:04
did
47:04
you
47:04
use?
47:05
Are
47:05
you,
47:05
like,
47:05
going
47:05
to
47:05
the
47:06
library
47:06
and,
47:06
like,
47:07
checking
47:07
out,
47:07
like,
47:08
a
47:08
book?
47:09
What
47:09
did
47:09
you
47:10
do?
47:10
You
47:11
have
47:11
one
47:11
minute.
47:11
I'll
47:12
go
47:12
get
47:12
the
47:12
resource.
47:13
Yeah,
47:13
hold
47:13
on.
47:13
Really?
47:14
Okay.
47:15
Yeah,
47:19
he's
47:19
going
47:19
to
47:19
the
47:19
scrolls.
47:20
Sarah.
47:20
Yeah.
47:22
Dust
47:23
them
47:23
off.
47:23
The
47:23
Library
47:24
of
47:24
Alexandria.
47:29
Okay,
47:29
friends,
47:30
Yay.
47:31
Show
47:31
us
47:31
what
47:31
you
47:32
got.
47:33
Whoa.
47:34
Oh,
47:35
my
47:35
God.
47:36
Can
47:36
you,
47:36
like,
47:37
open
47:37
it
47:37
up
47:37
so
47:37
we
47:37
can
47:37
see,
47:38
like,
47:38
a
47:38
page?
47:38
Oh,
47:38
yeah.
47:38
Hold
47:39
on.
47:40
So
47:40
here's
47:40
the
47:40
deluxe
47:41
paint.
47:42
This
47:42
was
47:42
the
47:43
thing
47:43
that
47:43
came
47:43
in,
47:46
everything
47:46
about
47:47
it.
47:47
So
47:47
this
47:47
was.
47:48
This
47:49
is
47:49
how
47:49
you
47:49
learned
47:50
Introduction
47:50
to
47:51
Amiga.
47:51
That
47:52
is
47:52
amazing.
47:53
And
47:54
it's
47:54
just
47:54
like,
47:54
you're
47:55
just
47:55
basically
47:56
going
47:56
through
47:56
that
47:56
page
47:57
by
47:57
page
47:57
and
47:57
kind
47:57
of
47:58
like
47:58
learning
47:58
the
47:58
computer.
48:00
And
48:00
things
48:00
were
48:01
sometimes
48:01
incorrect
48:02
and
48:02
things
48:03
were.
48:04
People
48:04
actually,
48:04
it's
48:05
funny,
48:05
people
48:05
remember
48:06
errors
48:07
on
48:07
given
48:07
pages
48:07
because
48:08
they
48:08
struggled
48:08
with
48:08
them
48:09
so
48:09
much.
48:09
They
48:09
were
48:09
like,
48:10
they
48:10
would
48:10
try
48:10
and
48:11
try
48:11
and
48:11
try,
48:12
and
48:12
they
48:12
couldn't
48:12
make
48:12
it
48:12
work.
48:12
And
48:13
then
48:13
they
48:13
realized
48:13
that
48:14
it
48:14
was
48:14
this
48:14
very
48:15
important
48:15
moment
48:16
for
48:16
a
48:16
young
48:16
person
48:17
to
48:17
realize
48:17
that,
48:17
like,
48:18
the
48:19
thing
48:19
that
48:19
was
48:19
handed
48:20
to
48:20
you
48:20
might
48:20
not
48:21
be
48:21
perfect,
48:21
that.
48:23
That
48:23
the
48:24
world
48:24
actually.
48:25
That
48:25
you.
48:25
You
48:26
actually
48:26
have
48:26
to
48:26
verify.
48:27
You
48:27
can't
48:27
just
48:27
trust.
48:29
It
48:29
was
48:29
just
48:30
hours
48:30
of
48:30
play.
48:31
Except
48:32
that
48:32
you
48:32
weren't
48:33
online.
48:33
You
48:34
know,
48:34
in
48:34
the
48:34
80s,
48:35
there's
48:35
90s,
48:36
there's
48:36
four
48:36
to
48:36
six
48:36
hours
48:37
in
48:37
an
48:37
adolescent's
48:38
afternoon
48:38
life
48:39
where
48:40
there's
48:40
no
48:40
signal
48:41
coming
48:41
in,
48:41
except
48:41
maybe
48:42
TV
48:42
or
48:42
radio,
48:43
where
48:44
they
48:44
go
48:44
see
48:44
friends
48:45
or
48:45
they
48:45
go
48:45
play
48:45
basketball
48:46
or
48:46
whatever,
48:46
but
48:46
there's
48:47
no
48:47
messages
48:48
coming
48:48
to
48:48
you,
48:49
so
48:49
you
48:50
got
48:50
to
48:50
do
48:50
something
48:50
with
48:50
that
48:51
time.
48:51
And
48:52
computers
48:52
are
48:52
great.
48:54
That's
48:54
amazing.
48:55
That's
48:55
very
48:55
poetic.
48:56
You
48:56
told
48:56
that
48:56
kid
48:57
that
48:57
one
48:57
day
48:57
they
48:57
would
48:57
just,
48:58
like,
48:58
tell
48:58
the
48:58
computer
48:58
what
48:59
they
48:59
wanted
49:00
and
49:00
in
49:01
regular
49:01
language
49:02
and
49:02
it
49:02
would
49:02
give
49:03
it
49:03
to
49:03
them.
49:03
They'd
49:03
be
49:03
blown
49:04
away,
49:04
I'm
49:04
sure.
49:05
I
49:05
don't
49:06
know
49:06
though.
49:06
I
49:06
think
49:06
we
49:06
all
49:06
thought
49:07
it
49:07
would
49:07
be
49:07
here
49:07
a
49:07
lot
49:07
sooner.
49:08
Oh
49:08
yeah,
49:09
that
49:09
might
49:09
be
49:09
fair.
49:09
That
49:10
might
49:10
be
49:10
true.
49:10
There
49:10
is
49:11
a
49:11
video
49:11
from
49:11
the
49:11
90s
49:12
that
49:12
Sun
49:12
Microsystems
49:13
made
49:13
which
49:14
is
49:14
exactly
49:14
LLM
49:15
style
49:15
interaction.
49:16
It's
49:16
just
49:16
this
49:16
woman
49:16
talking.
49:17
She's
49:17
like,
49:17
you
49:17
know,
49:17
make
49:18
it
49:18
blue
49:18
and
49:18
make
49:19
the
49:19
car
49:19
drive
49:20
around.
49:20
And
49:20
so
49:20
as
49:20
she's
49:21
getting
49:21
her
49:21
demo
49:21
together
49:22
and
49:22
it
49:22
has
49:23
90s
49:23
fashion,
49:23
I
49:24
can
49:24
dig
49:24
it
49:24
up
49:24
for
49:24
you.
49:24
It's
49:25
delightful.
49:26
One
49:26
last
49:27
thing
49:27
that
49:28
what
49:28
you
49:28
just
49:28
said
49:28
reminded
49:29
me
49:29
of
49:29
Paul,
49:29
but
49:29
also
49:30
relates
49:30
to
49:30
what
49:30
you
49:30
were
49:31
saying
49:31
earlier
49:31
about
49:32
why
49:33
getting
49:34
an
49:34
output
49:35
from
49:35
an
49:35
LLM
49:36
that
49:36
is
49:36
what
49:36
you're
49:37
envisioning
49:37
in
49:37
your
49:37
mind's
49:38
eye
49:38
is
49:38
difficult.
49:39
Is
49:39
like
49:40
the
49:40
data
49:41
transfer
49:41
from
49:42
your
49:42
brain
49:44
via
49:44
words
49:45
to
49:46
something
49:46
else
49:47
is
49:47
like
49:48
very
49:49
lossy.
49:50
Yeah,
49:51
it
49:51
is
49:51
very
49:51
lossy.
49:52
Well,
49:53
and
49:53
the
49:53
whole
49:53
process
49:53
is
49:53
lossy.
49:54
It's
49:54
basically
49:54
just
49:55
like
49:55
a
49:55
really
49:55
bad
49:56
database.
49:56
It
49:57
just
49:57
happens
49:57
to
49:57
be
49:59
like
49:59
have
50:00
all
50:00
of
50:00
human
50:00
knowledge
50:01
compressed
50:01
inside
50:02
of
50:02
it.
50:03
That
50:03
might
50:03
be
50:03
our
50:04
new
50:04
tagline.
50:04
A
50:04
very
50:05
bad
50:05
database
50:06
with
50:06
all
50:06
of
50:06
human
50:06
knowledge
50:07
inside
50:07
of
50:07
it.
50:10
Okay,
50:10
well,
50:10
Paul,
50:11
how
50:11
should
50:12
people
50:12
find
50:12
you?
50:12
Should
50:13
they
50:13
find
50:13
you?
50:13
And
50:14
how
50:14
can
50:14
they
50:14
find
50:14
out
50:14
about
50:15
aboard?
50:15
First
50:15
of
50:15
all,
50:15
do
50:16
you
50:16
want
50:16
to
50:16
be
50:16
found?
50:17
I'm
50:18
very
50:18
findable.
50:19
Less
50:19
so
50:19
an
50:19
ex
50:20
these
50:20
days.
50:20
But
50:20
the
50:21
easiest
50:21
way
50:21
to
50:21
find
50:22
me
50:22
is
50:22
just
50:22
go
50:22
to
50:22
aboard.com
50:23
and
50:24
b
50:24
o
50:25
ard.
50:26
That's
50:26
good.
50:26
I
50:27
can
50:27
spell
50:27
the
50:27
company
50:27
name.com
50:29
and
50:30
check
50:31
it
50:31
out,
50:31
try
50:31
the
50:31
demo,
50:32
see
50:32
where
50:32
it
50:32
breaks.
50:33
Let
50:33
us
50:33
know
50:33
what
50:33
you
50:34
think.
50:34
Like
50:34
I
50:34
really
50:34
want
50:35
to
50:35
know.
50:35
It
50:35
keeps
50:35
getting
50:36
better
50:36
and
50:36
I
50:36
really
50:36
like
50:37
the
50:37
feedback.
50:38
Paul.forward.com
50:39
I'm
50:40
very
50:40
findable.
50:40
I'm
50:41
happy
50:41
to
50:41
talk
50:41
to
50:41
everybody.
50:42
We
50:42
host
50:42
events,
50:43
we're
50:44
really
50:44
glad
50:44
to
50:44
see
50:44
everyone.
50:45
Doors
50:45
are
50:45
always
50:46
open,
50:46
so
50:47
if
50:47
anybody
50:47
wants
50:47
to
50:47
talk,
50:48
please
50:48
get
50:48
in
50:48
touch.
50:49
I
50:49
want
50:49
to
50:49
call,
50:49
hang
50:50
out
50:50
and
50:50
look
50:50
at
50:50
the
50:50
retro
50:51
computer
50:51
stuff.
50:52
You
50:52
can.
50:52
Thanks,
50:53
Paul,
50:53
it
50:54
was
50:54
great
50:54
to
50:54
see
50:54
you.
51:12
Are
51:12
you
51:12
a
51:12
big
51:13
retro
51:13
computing
51:13
nerd?
51:14
I
51:14
mean
51:16
aspirationally,
51:17
yes.
51:18
Okay,
51:19
I'm
51:19
going
51:19
to
51:19
just
51:20
give
51:20
you
51:20
the
51:20
word
51:21
Mr.
51:22
Just
51:22
type
51:22
in
51:23
Mr.
51:24
Retrocomputing.
51:25
Okay.
51:26
And
51:28
an
51:28
absolute
51:29
world
51:30
will
51:30
open.
51:31
I
51:31
can't
51:32
wait
51:32
to
51:32
see
51:32
what
51:32
happens.
51:33
Yeah,
51:33
just
51:34
not
51:34
necessarily
51:35
copyright
51:36
compliant,
51:37
but
51:37
an
51:37
absolute
51:38
world.