00:00:00
welcome to episode
00:00:01
124 of the all in podcast my
00:00:04
understanding is there's going to be a
00:00:06
bunch of global fan meetups for episode
00:00:09
125 if you go to Twitter and you search
00:00:11
for
00:00:12
all in fan meetups you might be able to
00:00:14
find the link but just to be clear we're
00:00:15
not they're not official all in this
00:00:17
they're fans it's self-organized which
00:00:20
is pretty mind-blowing but we can't
00:00:21
vouch for any particular organization
00:00:24
right nobody knows what's going to
00:00:26
happen at these things you can get
00:00:27
robbed it could be a setup I don't know
00:00:29
but I retweeted it anyway because there
00:00:32
are 31 cities where you lunatics are
00:00:35
getting together to celebrate the
00:00:37
world's number one business technology
00:00:39
podcast it is pretty crazy you know what
00:00:42
this reminds me of is in the early 90s
00:00:44
when Rush Limbaugh became a phenomenon
00:00:46
there used to be these things called
00:00:48
Rush rooms where like restaurants and
00:00:50
bars would literally broadcast rush over
00:00:54
their speakers during I don't know like
00:00:56
for the morning through lunch broadcast
00:00:57
and people would go to these restrooms
00:01:00
and listen together what was it like sex
00:01:01
when you were about 16 17 years old at
00:01:04
the time what was it like when you
00:01:05
hosted this it was a phenomenon but I
00:01:07
mean it's kind of crazy we've got like a
00:01:08
phenomenon going here where people are I
00:01:10
love it organizing you've said
00:01:12
phenomenon three times instead of
00:01:14
phenomenon he said it's phenomenon
00:01:16
phenomenal why is Saxon a good moochima
00:01:19
what's going on there's a specific
00:01:21
secret toe tap that you do under the
00:01:23
bathroom stalls when you go to a
00:01:24
rushroom
00:01:26
which we're already off the rails I
00:01:29
think you're getting confused about a
00:01:31
different event you went to
00:01:33
[Music]
00:01:37
let your Rain Man
00:01:38
[Music]
00:01:45
[Music]
00:01:49
there's a lot of actual news in the
00:01:51
world and generative AI is taking over
00:01:55
the dialogue and it's moving at a pace
00:01:57
that
00:01:59
none of us have ever seen in the
00:02:01
technology industry I think we'd all
00:02:02
agree the number of companies releasing
00:02:04
product
00:02:06
and the compounding effect of this
00:02:08
technology is phenomenal I think we
00:02:11
would all agree
00:02:12
a product came out this week called Auto
00:02:15
GPT
00:02:17
and people are losing their mind over it
00:02:20
basically what this does is it lets
00:02:23
different gpts
00:02:25
talk to each other and so you can have
00:02:29
agents working in the background and
00:02:30
we've talked about this on previous
00:02:31
podcasts but they could be talking to
00:02:35
each other essentially
00:02:36
and then completing tasks without much
00:02:39
intervention so if let's say you had a
00:02:41
sales team and you said to the sales
00:02:44
team hey look for leads that have these
00:02:48
characteristics for our sales software
00:02:49
put them into our database find out if
00:02:52
they're already in the database alert A
00:02:54
salesperson to it compose a message
00:02:56
based on that person's profile on
00:02:57
LinkedIn or Twitter or wherever and then
00:03:00
compose an email send it to them if they
00:03:02
reply offer them to do a demo and then
00:03:04
put that demo on the calendar of the
00:03:06
salesperson thus eliminating a bunch of
00:03:07
jobs and you could run these what would
00:03:09
essentially be cron jobs in the
00:03:11
background
00:03:12
forever and they could interact with
00:03:15
other llms in real time sex just gave
00:03:18
but one example here but when you see
00:03:19
this happening give us your perspective
00:03:21
on what this Tipping Point means let me
00:03:24
take a shot at explaining it in a
00:03:26
slightly different way not that your
00:03:28
explanation was wrong but I just think
00:03:30
that maybe explain it in terms of
00:03:32
something more tangible sure so I had a
00:03:35
friend who's a developer has been
00:03:36
playing with auto GPT by the way so you
00:03:40
can see it's on GitHub it's kind of an
00:03:41
open source project it was sort of a
00:03:43
hobby project it looks like that
00:03:44
somebody put up there it's been out for
00:03:46
about two weeks it's already got 45 000
00:03:49
stars on GitHub which is a huge number
00:03:51
explain what GitHub is for the audience
00:03:53
is this a code repository and you can
00:03:55
create you know repos of code for open
00:03:57
source projects that's where all the
00:03:58
developers check in their code so you
00:04:01
know for open source projects like this
00:04:02
anyone can go see it and play with it
00:04:03
it's like PornHub but for developers it
00:04:07
would be more like amateur or PornHub
00:04:09
because you're contributing your scenes
00:04:11
as it were your code yes but yes
00:04:12
continue this thing has a ton of of
00:04:15
stars and apparently just last night I
00:04:18
got another 10 000 Stars overnight this
00:04:19
thing is like exploding in terms of
00:04:21
popularity but anyway what you do is you
00:04:24
give it an assignment and what Auto GPT
00:04:27
can do that's different is it can string
00:04:30
together prompts so if you go to chat
00:04:32
GPT you prompt it one at a time and what
00:04:34
the human does is you get your answer
00:04:36
and then you think of your next prompt
00:04:37
and then you kind of go from there and
00:04:39
you end up in a long conversation that
00:04:40
gets you to where you want to go
00:04:42
so the question is what if the AI could
00:04:45
basically prompt itself then you've got
00:04:48
the basis for autonomy and that's what
00:04:50
this project is designed to do so what
00:04:53
you'll do is what my friend did is he
00:04:54
said okay you're an event planner
00:04:57
Ai and what I would like you to do is
00:05:01
plan a trip for me
00:05:02
for a wine tasting in heelsburg this
00:05:06
weekend and I want you to find like the
00:05:08
best place I should go and it's got to
00:05:10
be kid-friendly not everyone's going to
00:05:11
drink we're gonna have kids there and
00:05:12
I'd like to be able to have other people
00:05:14
there and so I'd like you to plan this
00:05:16
for me and so what Auto GPT did is it
00:05:20
broke that down into a task list and
00:05:22
every time I completed a task it would
00:05:25
add a new task to the bottom of that
00:05:27
list and so the output of this is that
00:05:30
it searched a bunch of different wine
00:05:32
tasting venues it found a venue that had
00:05:34
a bocce ball lawn area for kids it came
00:05:38
up with a schedule
00:05:39
it created a budget it created a
00:05:42
checklist for an event planner
00:05:44
it did all these things my friend says
00:05:46
he's actually in a book The Venue this
00:05:47
weekend and use it so we're going beyond
00:05:50
the ability just for a human to just
00:05:52
prompt the AI we're now
00:05:54
the AI can take on complicated tasks and
00:05:58
again it can recursively update its task
00:06:00
list based on what it learns from its
00:06:03
own previous prompt so what you're
00:06:04
seeing now is the basis for a personal
00:06:07
digital assistant this is really where
00:06:09
it's all headed is that you can just
00:06:11
tell the AI to do something for you
00:06:12
pretty complicated and it will be able
00:06:15
to do it it will be able to create its
00:06:16
own task list and get the job done
00:06:19
in quite complicated jobs so that's why
00:06:22
everyone's losing their [ __ ] over this
00:06:23
freeberg your thoughts on automating
00:06:26
these tasks and having them run and and
00:06:28
add tasks to the list this does seem
00:06:31
like a sort of seminal moment in time
00:06:34
that this is actually working
00:06:36
I think we've been seeing seminal
00:06:39
moments
00:06:41
over the last couple of weeks and months
00:06:43
kind of continuously every time we chat
00:06:47
about stuff or every day there's new
00:06:49
releases that are Paradigm shifting and
00:06:52
kind of reveal new applications and and
00:06:55
perhaps Concepts structurally that we
00:06:58
didn't really
00:06:59
have a good grasp of before some
00:07:01
demonstration came across chat GPT was
00:07:03
kind of the seat of that and then all of
00:07:05
this Evolution sense has really I think
00:07:08
changed the landscape for really how we
00:07:10
think about our interaction with digital
00:07:12
world and where the digital world can go
00:07:15
and how it can interact with the
00:07:16
physical world it's it's just really
00:07:17
profound
00:07:19
one of the interesting aspects that I
00:07:21
think I saw with some of the
00:07:23
applications of Auto GPT were these
00:07:25
almost like
00:07:28
autonomous characters in in like a game
00:07:31
simulation that could interact with each
00:07:33
other or these autonomous characters
00:07:34
that would speak back and forth to one
00:07:36
another
00:07:37
where each
00:07:38
instance has its own kind of predefined
00:07:41
role and then it explores some set of
00:07:44
Discovery or application or prompt back
00:07:46
and forth with the other agent and that
00:07:49
the kind of recursive outcomes with this
00:07:51
agent to agent interaction model and
00:07:53
perhaps multi-agent interaction model
00:07:56
again reveals an entirely new paradigm
00:07:58
for you know how things can be done
00:08:00
simulation wise
00:08:02
you know Discovery wise engagement wise
00:08:04
where One agent you know each agent can
00:08:06
be a different character in a room and
00:08:08
you can almost see how a team might
00:08:10
resolve to create a new product
00:08:11
collaboratively by telling each of those
00:08:14
agents to have a different character
00:08:16
background or different set of data or a
00:08:18
different set of experiences or
00:08:19
different set of personality traits and
00:08:21
the evolution of those that multi-agent
00:08:23
system outputs you know something that's
00:08:25
very novel that perhaps any of the
00:08:27
agents operating independently we're not
00:08:29
able to kind of reveal themselves so
00:08:31
again like another kind of dimension of
00:08:32
interaction
00:08:34
with these with these models and it
00:08:37
again like every week it's a whole other
00:08:39
layer to The Onion
00:08:41
it's super exciting and compelling and
00:08:43
the rate of change and the pace of kind
00:08:45
of you know New Paths being
00:08:48
being defined here really I think makes
00:08:51
it difficult to catch up and
00:08:53
particularly it highlights why it's
00:08:55
going to be so difficult I think for
00:08:56
Regulators to come in and try and set a
00:08:59
set of standards and a set of rules at
00:09:02
this stage because we don't even know
00:09:03
what we have here yet and it's going to
00:09:05
be very hard to kind of put the genie
00:09:06
back in the Box yeah and you're also
00:09:09
referring I think to the Stanford and
00:09:12
Google paper that was published this
00:09:14
week they did a research paper where
00:09:16
they created essentially The Sims if you
00:09:18
remember that video game put a bunch of
00:09:20
and what you might consider NPCs
00:09:22
non-playable characters you know the
00:09:24
merchant or the whoever in a um in a
00:09:28
video game and they said
00:09:30
each of these agents should talk to each
00:09:32
other put them in a simulation one of
00:09:33
them decided to have a birthday party
00:09:34
they decided to invite other people and
00:09:37
then they have memories and so then over
00:09:38
time they would generate responses like
00:09:41
I can't go to your birthday party but
00:09:43
happy birthday and then they would
00:09:45
follow up with each player and seemingly
00:09:48
emergent behaviors came out of this sort
00:09:51
of simulation which of course now has
00:09:53
everybody thinking well of course we as
00:09:55
humans and this is simulation there are
00:09:57
living in a simulation we've all just
00:09:59
been put into this is what we're
00:10:01
experiencing right now
00:10:03
how impressive this technology is or is
00:10:07
it
00:10:08
oh wow human cognition maybe we thought
00:10:11
was incredibly special but we can
00:10:13
actually simulate a significant portion
00:10:15
of what we do as humans so we're kind of
00:10:17
taking the Shine off of
00:10:19
Consciousness I'm not sure it's that but
00:10:21
I would make two comments I think this
00:10:23
is a really important week
00:10:25
because it starts to show
00:10:28
how fast the recursion is
00:10:31
with AI so in other Technologies and in
00:10:35
other breakthroughs
00:10:37
the recursive iterations took years
00:10:39
right if you think about how long did we
00:10:41
wait for from iPhone 1 to iPhone 2 it
00:10:44
was a year
00:10:45
right we waited two years for the App
00:10:48
Store
00:10:49
everything was measured in years maybe
00:10:51
things when they were really really
00:10:53
aggressive and really disruptive were
00:10:56
measured in months
00:10:57
except now these incredibly Innovative
00:11:00
breakthroughs are being measured in days
00:11:02
and weeks
00:11:03
that's incredibly profound
00:11:06
and I think it has some really important
00:11:09
implications to like the three big
00:11:11
actors in this play right so it has I
00:11:14
think huge implications to these
00:11:15
companies it's not clear to me how you
00:11:18
start a company anymore
00:11:20
I don't understand why
00:11:23
you would have a 40 or 50 person company
00:11:27
to try to get to an MVP I think you can
00:11:30
do that with three or four people
00:11:33
and that has huge implications then to
00:11:36
the second actor in this play which are
00:11:38
the investors in Venture capitalists
00:11:40
that typically fund this stuff because
00:11:42
all of our Capital allocation models
00:11:45
were always around writing 10 and 15 and
00:11:47
20 million dollar checks and 100 million
00:11:49
dollar checks then 500 million dollar
00:11:51
checks into these businesses that
00:11:53
absorbs tons of money
00:11:55
but the reality is like you know you're
00:11:57
looking at things
00:11:58
like mid-journey and others that can
00:12:00
scale to enormous size with very little
00:12:03
Capital many of which can now be
00:12:05
bootstrapped
00:12:07
so it takes really really small amounts
00:12:10
of money
00:12:11
and so I think that's a huge implication
00:12:13
so for me personally I am looking at
00:12:15
company formation
00:12:17
being done in a totally different way
00:12:19
and our Capital allocation model is
00:12:21
totally wrong size look fund four for me
00:12:23
was one billion dollars does that make
00:12:26
sense
00:12:27
nope for the next three or four years no
00:12:29
the right number may actually be 50
00:12:31
million dollars invested over the next
00:12:33
four years I think the VC job is
00:12:35
changing I think company startups are
00:12:36
changing I want to remind you guys of
00:12:38
one quick thing as a tangent
00:12:41
I had this meeting with Andre carpathy I
00:12:42
talked about this on the Pod where I
00:12:44
said I challenged him I said listen the
00:12:46
real goal should be to go and disrupt
00:12:49
existing businesses using these tools
00:12:51
cutting out all the sales and marketing
00:12:53
right and just delivering something and
00:12:55
I use the example of stripe
00:12:57
disrupting stripe by going to Market
00:12:59
with an equivalent product with
00:13:00
one-tenth the number of employees at one
00:13:03
tenth the cost
00:13:04
what's incredible is that this Auto GPT
00:13:06
is the answer to that exact problem
00:13:09
why because now if you are a young
00:13:12
industrious entrepreneur
00:13:14
if you look at any bloated organization
00:13:17
that's building Enterprise class
00:13:18
software
00:13:19
you can string together a bunch of
00:13:22
agents that will Auto construct
00:13:24
everything you need
00:13:26
to build a much much cheaper product
00:13:28
that then you can deploy for other
00:13:31
agents to consume so you don't even need
00:13:33
a sales team anymore this is what I mean
00:13:35
by this crazy recursion that's possible
00:13:37
yeah so I'm really curious to see how
00:13:40
this actually affects like all of this
00:13:42
all of these you know continuation
00:13:44
companies I mean it's a continuation of
00:13:47
and then the last thing I just want to
00:13:48
say is related to my tweet I think this
00:13:50
is exactly the moment where we now have
00:13:53
to have a real conversation about
00:13:54
regulation and I think it has to happen
00:13:56
otherwise it's going to be a [ __ ] show
00:13:57
let's put a pin in that for a second but
00:13:59
I want to get Sax's response to some of
00:14:01
this so sax we saw this before it used
00:14:04
to take two or three million dollars to
00:14:05
commercialize a web-based software
00:14:07
product app then it went down to 500k
00:14:09
then 250. I don't know if you saw this
00:14:12
story but if you remember the hit game
00:14:14
on your iPhone Flappy Birds Flappy Birds
00:14:18
uh you know was a phenomenon at you know
00:14:21
hundreds of millions of people played
00:14:23
this game over some period of time
00:14:25
somebody made it by talking to chat gpt4
00:14:29
in mid-journey in an hour
00:14:31
so the perfect example and listen it's a
00:14:33
game so it's something silly but I was
00:14:35
talking to two developers this weekend
00:14:37
and one of them was an okay developer
00:14:40
and the other one was an actual 10x
00:14:41
developer who's built you know very
00:14:43
significant companies and they were
00:14:45
coding together last week and because of
00:14:48
how fast chat GPT and other services
00:14:50
were writing code for them
00:14:52
he looked over at her and said you know
00:14:55
you're basically a 10x developer now my
00:14:59
superpower is gone so where does this
00:15:02
lead you to believe company formation is
00:15:05
going to go is this going to be
00:15:08
you know massively deflationary
00:15:09
companies like stripe are going to have
00:15:11
a hundred competitors in a very short
00:15:13
period of time or are we just going to
00:15:14
go down the long tail of ideas and solve
00:15:18
everything with software how is this
00:15:20
going to play out in the in the startup
00:15:22
space David sacks
00:15:24
well I think it's true that
00:15:27
developers and especially Junior
00:15:29
developers get a lot more leverage on
00:15:32
their time and so it is going to be
00:15:34
easier for small teams to get to an MVP
00:15:37
which is something they always should
00:15:38
have done anyway with their seed round
00:15:40
you shouldn't have needed you know 50
00:15:43
developers to build your V1 it should be
00:15:45
you know that's the founders really
00:15:48
so that that I think is already
00:15:51
happening and that Trend will continue I
00:15:53
think we're still a ways away from
00:15:57
stores being able to replace entire
00:15:59
teams of people I just you know I think
00:16:01
right now
00:16:02
to find a ways months years decade well
00:16:07
it's in the years I think for sure we
00:16:08
don't know how many years and the reason
00:16:10
I say that is it's just very hard
00:16:13
to replace
00:16:14
you know 100 of what any of these
00:16:17
particular job functions do 100 of what
00:16:20
a sales rep does 100 of what a marketing
00:16:22
rep does or even what a coder does so
00:16:25
right now I think we're still at the
00:16:26
phase of this where it's a tool that
00:16:28
gives a human leverage and I think we're
00:16:32
still a ways away from the you know
00:16:34
human being completely out of the loop I
00:16:36
think right now I see it mostly as a
00:16:38
Force for good
00:16:40
as opposed to something that's creating
00:16:42
okay a ton of dislocation Friedberg your
00:16:45
thoughts if we follow the trend line you
00:16:48
know to make that video game that you
00:16:49
shared took probably a few hundred human
00:16:52
years then a few dozen human years then
00:16:55
you know with other tool kits coming out
00:16:57
maybe a few human months and now this
00:17:00
person did it in one human day using
00:17:03
this tooling so if you think about the
00:17:06
implication for that I mentioned this
00:17:07
probably last year I really do believe
00:17:10
that at some point the whole concept of
00:17:12
Publishers and Publishing maybe goes
00:17:14
away where you know much like we saw so
00:17:17
much of the content on the internet
00:17:18
today being user generated you know most
00:17:20
of the content is made by individuals
00:17:21
posted on YouTube or Twitter that's most
00:17:24
of what we consume nowadays or Instagram
00:17:25
or Tick Tock in terms of video content
00:17:29
we could see the same
00:17:31
in terms of software itself where you no
00:17:34
longer need a software startup or a
00:17:36
software company to render or generate a
00:17:39
set of tools for a particular user but
00:17:41
that the user may be able to Define to
00:17:44
their agent their AI agent the set of
00:17:47
tools that they would individually like
00:17:48
to use or to create for them to do
00:17:50
something interesting and so the idea of
00:17:52
buying or subscribing to software or
00:17:55
even buying or subscribing to a video
00:17:57
game or to a movie or to some other form
00:17:59
of content
00:18:01
starts to diminish as The Leverage goes
00:18:04
up with these tools the accessibility
00:18:06
goes up you no longer need a computer
00:18:07
engineering degree or computer science
00:18:09
degree to be able to harness them or use
00:18:11
them and individuals may be able to
00:18:13
speak in simple and plain English that
00:18:15
they would like a book or a movie that
00:18:17
does that looks and feels like the
00:18:19
following or a video game that feels
00:18:21
like the following and so when I open up
00:18:23
my iPhone maybe it's not a screen with
00:18:25
dozens of video games but it's one
00:18:27
interface and the interface says what do
00:18:29
you feel like playing today and then I
00:18:30
can very clearly and succinctly State
00:18:32
what I feel like playing and it can
00:18:33
render that game and render the code
00:18:35
render the engine render the graphics
00:18:37
and everything on the Fly for me and I
00:18:39
can use that and so you know I kind of
00:18:41
think about this as being a bit of a
00:18:42
leveling up that the idea that all
00:18:44
technology again starts Central and
00:18:46
moves to kind of the edge of the network
00:18:47
over time that may be what's going on
00:18:50
with computer programming itself now
00:18:52
where the toolkit to actually use
00:18:55
computers to generate stuff for us is no
00:18:58
longer a toolkit that's harnessed and
00:19:00
controlled and utilized by a set of
00:19:02
centralized Publishers but it becomes
00:19:04
distributed and used at the edge of the
00:19:06
network by users like anyone and then
00:19:08
the edge of the Network Technology can
00:19:10
render the software for you and it
00:19:12
really creates a profound change in the
00:19:14
entire business landscape of software
00:19:17
and the internet
00:19:18
and I think it's uh you know it's it's
00:19:20
really like we're just starting to kind
00:19:22
of see have our heads unravel around
00:19:24
this notion and we're sort of trying to
00:19:26
link it to the old Paradigm which is all
00:19:28
startups are going to get cheaper
00:19:29
smaller teams but it may be that you
00:19:30
don't even need startups for a lot of
00:19:32
stuff anymore you don't even need teams
00:19:33
and you don't even need companies to
00:19:35
generate and render software to do stuff
00:19:37
for you anymore when we look at this it
00:19:39
it's kind of a pattern of
00:19:43
augmentation as we've been talking about
00:19:45
here we're augmenting human intelligence
00:19:48
then replacing this replication or this
00:19:52
automation I guess might be a nice way
00:19:54
to say it so it's augmentation then
00:19:56
automation
00:19:57
and then perhaps deprecation where do
00:20:00
you sit on this it seems like sax feels
00:20:03
it's going to take years
00:20:04
and Freeburg thinks hey maybe startups
00:20:06
and content are over where do you sit on
00:20:08
this augmentation automation deprecation
00:20:11
Journey we're on I think that humans
00:20:13
have judgment and I think it's going to
00:20:14
take decades for agents to replace good
00:20:17
judgment I think that's where we have
00:20:20
some
00:20:21
defensible ground and I'm going to say
00:20:23
something controversial I don't think
00:20:25
developers anymore have good judgment
00:20:28
developers get to the answer or they
00:20:30
don't get to the answer and that's what
00:20:31
agents have done because the the 10x
00:20:33
engineer had better judgment than the 1X
00:20:36
engineer
00:20:37
but by making everybody a 10x engineer
00:20:39
you're taking judgment away
00:20:41
you're taking code paths that are now
00:20:44
obvious and making it available to
00:20:46
everybody it's effectively like what you
00:20:48
did in chess an AI created a solver so
00:20:51
everybody understood the most efficient
00:20:53
path in every single spot to do the eat
00:20:56
most EV positive thing the most expected
00:20:58
value positive thing coding is very
00:21:00
similar that way you can reduce it and
00:21:02
view it very very reductively so there
00:21:05
is no differentiation in code and so I
00:21:07
think Freeburg is right so for example
00:21:09
let's say you're going to start a
00:21:11
company today
00:21:12
why do you even care
00:21:14
what database you use
00:21:16
why do you even care
00:21:19
which Cloud you're built on to free
00:21:21
Brook's Point why do any of these things
00:21:23
matter they don't matter they were
00:21:25
decisions that used to matter when
00:21:27
people had a job to do and you paid them
00:21:30
for their judgment oh well we think gcp
00:21:32
is better for this specific workload and
00:21:34
we think that this database architecture
00:21:36
is better for that specific workload and
00:21:38
we're going to run this on AWS but that
00:21:40
on azure
00:21:41
and do you think an agent cares if you
00:21:44
tell an agent find me the cheapest way
00:21:46
to execute this thing
00:21:48
and if it ever gets not you know cheaper
00:21:50
to go someplace else do that for me as
00:21:52
well and you know ETL all the data and
00:21:54
put it in the other thing and I don't
00:21:56
really care so you're saying it will it
00:21:58
will swap out stripe for add Yen or it
00:22:01
doesn't for Amazon web services it's
00:22:03
going to be ruthless it's going to be
00:22:05
ruthless and I think that the point of
00:22:06
that that and that's the exact perfect
00:22:08
word Jason AI is ruthless because it's
00:22:11
emotionless it was not taken to a steak
00:22:14
dinner
00:22:14
it was not brought to a basketball game
00:22:17
it was not sold into a CEO it's an agent
00:22:21
that looked at a bunch of API endpoints
00:22:24
figured out how to write code to it to
00:22:27
get done the job at hand that was tasked
00:22:29
to it within a budget right the other
00:22:31
thing that's important is these agents
00:22:32
execute within budgets so another good
00:22:35
example was and this is a much simpler
00:22:38
one but a guy said I would like seven
00:22:43
days worth of meals here are my
00:22:46
constraints from a dietary perspective
00:22:48
here are also my budgetary constraints
00:22:50
and then what this agent did was figured
00:22:53
out how to go and use the instacart
00:22:55
plug-in at the time and then these other
00:22:57
things and execute within the budget
00:22:59
how is that different when you're a
00:23:01
person that raises five hundred thousand
00:23:03
dollars and says I need a full stack
00:23:05
solution that does X Y and Z for two
00:23:07
hundred thousand dollars it's the exact
00:23:09
same problem
00:23:10
so I think it's just a matter of time
00:23:13
until we start to cannibalize these
00:23:15
extremely expensive fossified large
00:23:18
organizations that have relied on a very
00:23:20
complicated go to market in sales and
00:23:22
marketing motion I don't think you need
00:23:24
it anymore in a world of of agents and
00:23:25
auto gpts and I think
00:23:28
that to me is quite interesting because
00:23:30
a it creates an obvious set of public
00:23:33
company shorts
00:23:34
and then B
00:23:36
you actually want to arm the rebels and
00:23:39
arming the rebels to use the Tobi
00:23:40
lootkey analogy here would mean to seed
00:23:43
hundreds of one-person teams hundreds
00:23:47
and just say go and build this entire
00:23:49
stack all over again using a bunch of
00:23:51
Agents
00:23:52
yeah and I think recursively you'll get
00:23:54
to that answer in in less than a year
00:23:57
interestingly
00:23:59
when you talk about the emotion of
00:24:00
making these decisions if you look at
00:24:02
Hollywood I just interviewed on my other
00:24:04
podcast the founder of you have another
00:24:07
podcast I do it's called startups thank
00:24:10
you episode you've been on her four
00:24:12
times
00:24:13
don't give them an excuse to plug it
00:24:15
listen I'm not going to this week in
00:24:17
startups available on Spotify and iTunes
00:24:19
and youtube.com this weekend Runway is
00:24:22
the name of this company I interviewed
00:24:23
and what's fascinating about this is he
00:24:26
told me on everything everywhere all at
00:24:29
once the award-winning film
00:24:31
they had seven visual effects people on
00:24:33
it and they were using his software
00:24:35
the late night shows like Colbert and
00:24:37
stuff like that are using it they are
00:24:39
ruthless in terms of creating crazy
00:24:41
visual effects now without and you can
00:24:44
do text prompt to get video output and
00:24:48
it is quite reasonable what's coming out
00:24:50
of it but you can also train it on
00:24:52
existing data sets so they're going to
00:24:54
be able to take something sax like The
00:24:56
Simpsons
00:24:58
or South Park or Star Wars or Marvel
00:25:00
take the entire Corpus of the comic
00:25:03
books and the movies and the TV shows
00:25:05
and then have people type in have Iron
00:25:07
Man do this have Luke Skywalker do that
00:25:10
and it's going to Output stuff and I
00:25:12
said hey when would this reach
00:25:15
the the level that the Mandalorian TV
00:25:17
show is and he said within two years now
00:25:19
he's talking his own book but it's quite
00:25:21
possible about that all these visual
00:25:23
effects people from industrial Light
00:25:25
Magic on down
00:25:27
are going to be replaced with director
00:25:29
sacks who are currently using this
00:25:31
technology to do
00:25:33
what do they call the images like that
00:25:35
go with the script storyboards
00:25:36
storyboards thank you they're doing
00:25:38
storyboards in this right now right the
00:25:40
difference between the storyboards acts
00:25:42
and the output is closing in the next 30
00:25:45
months I would say right I mean maybe
00:25:48
you could speak to a little bit about
00:25:49
the pace here because that is the
00:25:51
perfect ruthless example of Ruthless AI
00:25:52
I mean you could have the entire team at
00:25:55
industrial Light Magics or Pixar be
00:25:58
unnecessary this decade well I mean you
00:26:01
see a bunch of the pieces already there
00:26:02
so you have stable diffusion you have
00:26:04
the ability to type in the image that
00:26:06
you want and it spits out you know a
00:26:08
version of it or 10 different versions
00:26:10
of it and you can pick which one you
00:26:11
want to go with you have the ability to
00:26:13
create characters you have the ability
00:26:14
to create voices you have the ability to
00:26:18
replicate a celebrity voice the only
00:26:20
thing that's not there yet as far as I
00:26:23
know is the ability to take static
00:26:24
images and stream them together into a
00:26:26
motion picture but that seems like it's
00:26:28
coming really soon so yeah in theory you
00:26:31
should be able to train the model where
00:26:33
you just give it screenplay and it
00:26:35
outputs essentially an animated movie
00:26:37
and then you should be able to fine tune
00:26:39
it by choosing the voices that you want
00:26:41
and the characters that you want and you
00:26:44
know and that kind of stuff so yeah I
00:26:46
think we're close to it now I think that
00:26:48
the question though is you know every
00:26:50
nine let's call it of reliability is a
00:26:54
big advancement so yeah it might be easy
00:26:56
to get to 90 percent within two years
00:26:59
but it might take another two years to
00:27:01
go from 90 to 99 and then it might take
00:27:03
another two years to get to 99.9 and so
00:27:06
on and so to actually get to the point
00:27:08
where you're at this stage where you can
00:27:10
release a theatrical quality movie I'm
00:27:12
sure it will take a lot longer than two
00:27:14
years well but look at this sex I'm just
00:27:15
going to show you one image this is the
00:27:17
input was aerial Drone footage of a
00:27:19
mountain range and this is what it came
00:27:21
up with now if you were watching TV in
00:27:23
the 80s or 90s on a non-hd TV this would
00:27:26
look indistinguishable from anything
00:27:28
you've seen and so this is at a pace
00:27:32
that's kind of crazy there's also
00:27:33
opportunity here right Friedberg I mean
00:27:35
if we were to look at something like The
00:27:37
Simpsons which has gone on for 30 years
00:27:39
if young people watching The Simpsons
00:27:42
could create their own scenarios or with
00:27:45
auto GPT imagine you told The Simpsons
00:27:50
stable diffusion instance read what's
00:27:53
happening in the news have Bart Simpson
00:27:55
respond to it have the South Park
00:27:57
characters parody whatever happened in
00:28:00
the news today you could have automated
00:28:02
real-time episodes of South Park just
00:28:04
being published onto some website before
00:28:07
you move on did you see the the Wonder
00:28:09
Studio demo we can pull this one up it's
00:28:12
really cool yeah please this is a
00:28:14
startup that's using this type of
00:28:16
technology and the way it works is
00:28:19
new film a live action scene with a
00:28:23
regular actor but then you can just drag
00:28:25
and drop and animate a character onto it
00:28:27
and it then converts that scene into a
00:28:32
movie with that character like Planet of
00:28:34
the Apes or Lord of the Rings right yeah
00:28:36
exactly you see the person who kept
00:28:38
winning all the Oscars so there it goes
00:28:40
after the robot has replaced the human
00:28:42
wow you can imagine like every piece of
00:28:44
this just eventually gets swapped out
00:28:46
with AI right like you should be able to
00:28:48
tell the AI
00:28:50
give me a picture of a human leaving a
00:28:56
building
00:28:57
like a Victorian era building in New
00:29:00
York and certainly can give you a static
00:29:02
image of that so it's not that far to
00:29:04
then give you a video of that right
00:29:07
and so yeah I think we're we're pretty
00:29:09
close for let's call it hobbyists or
00:29:11
amateurs to be able to create pretty
00:29:13
nice looking movies using these types of
00:29:16
tools but again I think there's a jump
00:29:19
to get to the point where you're just
00:29:20
all together replacing one of the things
00:29:24
I'll say on this is we still keep trying
00:29:25
to relate it back to the way media
00:29:28
narrative has been explored and written
00:29:31
by humans in the past very kind of
00:29:34
linear storytelling you know it's a
00:29:36
two-hour movie 30 minute TV segment
00:29:38
eight minute YouTube clip 30 second
00:29:41
Instagram clip whatever
00:29:42
but one of the
00:29:44
enabling capabilities with this set of
00:29:48
tools is that these stories
00:29:51
the way that they're rendered and the
00:29:52
way that they're explored by individuals
00:29:54
can be fairly dynamic
00:29:58
you could watch a movie with the same
00:30:00
story all four of us could watch a movie
00:30:02
with the same story but from totally
00:30:05
different Vantage points and some of us
00:30:06
could watch it in an 18 minute version
00:30:08
or a two-hour version or a you know
00:30:10
three season episode episodic version
00:30:13
where the the way that this opens up the
00:30:16
potential for creators and all so so now
00:30:18
I'm kind of saying before I was saying
00:30:20
hey individuals can make their own
00:30:21
movies and videos that's going to be
00:30:23
incredible there's a separate I think
00:30:25
creative output here
00:30:27
which is the leveling up that happens
00:30:30
with creators that maybe wasn't possible
00:30:33
to them before so perhaps a Creator
00:30:35
writes a short book a short story and
00:30:38
then that short story gets rendered into
00:30:40
a system that can allow each one of us
00:30:42
to explore it and enjoy it in different
00:30:43
ways and I as the Creator can define
00:30:46
those different Vantage points I as the
00:30:48
Creator can say here's a little bit of
00:30:50
this personality this character trait
00:30:52
and so what I can now do as a Creator is
00:30:55
stuff that I never imagined I could do
00:30:57
before think about old school
00:30:58
photographers doing black and white
00:30:59
photography with pinhole cameras and
00:31:02
then they come across Adobe Photoshop
00:31:03
what they can do with Adobe Photoshop
00:31:05
was stuff that they could never
00:31:06
conceptualize of in those old days I
00:31:09
think what's going to happen for
00:31:10
creators going forward and this is going
00:31:12
back to that point that we had last week
00:31:13
or two weeks ago about the guy that was
00:31:15
like hey I'm out of a job I actually
00:31:17
think that the opportunity for creating
00:31:19
new stuff in new ways is so profoundly
00:31:22
expanding that individuals can now write
00:31:25
entire universes that can then be
00:31:27
enjoyed by millions of people from
00:31:29
completely different lengths and
00:31:30
viewpoints and and models that can be
00:31:33
interactive they can be static they can
00:31:35
be dynamic
00:31:36
and that the person
00:31:37
personalized but the tooling that you as
00:31:39
a Creator now has you could choose which
00:31:42
characters you wanted to find you could
00:31:43
choose which content you want to write
00:31:46
you could choose which content you want
00:31:48
the AI to fill in for you and say hey
00:31:50
create 50 other characters in the
00:31:52
village and then when the viewer reads
00:31:54
the book or watches the movie Let Them
00:31:56
explore or have a different interaction
00:31:57
with a set of of those villagers uh in
00:32:00
that Village or you could say hey here's
00:32:01
the one character everyone has to meet
00:32:03
here's what I want them to say and you
00:32:05
can Define the dialogue and so the way
00:32:07
the creators can start to kind of
00:32:08
harness their creative chops and create
00:32:10
new kinds of
00:32:12
modalities for content and for
00:32:14
exploration I think is going to be so
00:32:16
beautiful and incredible I mean
00:32:18
Freeburg yeah you can choose the limits
00:32:20
of how much you want the individual to
00:32:23
enjoy from your content versus how
00:32:25
narrowly you want to Define it and my
00:32:27
guess is that the creators that are
00:32:28
going to win are going to be the ones
00:32:30
that are going to create more dynamic
00:32:31
range in the creative output and then
00:32:34
individuals are going to kind of be
00:32:35
stuck they're gonna be more into that
00:32:37
than they will with the static everyone
00:32:39
watches the same thing over and over so
00:32:40
there will be a whole new world of
00:32:42
creators that you know maybe have a
00:32:43
different set of tools that then just
00:32:46
just realizing a lot better to build on
00:32:48
what you're saying for a burmes I think
00:32:49
it's incredibly insightful just think
00:32:50
about the controversy around two aspects
00:32:53
of a franchise like James Bond number
00:32:55
one who's your favorite Bond we grew up
00:32:57
with Roger Moore We lean towards that
00:32:59
then we discover Sean Connery and then
00:33:00
all of a sudden you see you know the
00:33:02
latest one he's just extraordinary
00:33:04
and and Daniel Craig you're like you
00:33:06
know what that's the one that I love
00:33:07
most but what if you could take any of
00:33:09
the films you could say let me get you
00:33:10
know give me the spy who loved me but
00:33:11
put Daniel Craig in it Etc and that
00:33:13
would be available to you and then think
00:33:15
about the next controversy which is oh
00:33:16
my God does Daniel does James Bond need
00:33:18
to be a white guy from the UK of course
00:33:20
not you can you place it around the
00:33:22
world and each region could get their
00:33:24
own celebrity their number one celebrity
00:33:26
to play the lead and controversy over
00:33:29
you know the old story The Epic of
00:33:31
Gilgamesh right so like that story was
00:33:33
retold in dozens of different languages
00:33:35
and it was told through the oral
00:33:37
tradition it was like You Know spoken by
00:33:38
bards around a fire pit and whatnot and
00:33:41
all of those stories were told with
00:33:43
different characters and different names
00:33:44
and different experiences some of them
00:33:46
were 10 minutes long some of them were
00:33:48
multi-hour sagas explained through the
00:33:50
story but ultimately the morality of the
00:33:53
story the storyline the intentionality
00:33:55
of the original creator of that story
00:33:57
yes came through the the Bible is
00:33:58
another good example of this where much
00:34:00
of the underlying morality and ethics in
00:34:02
the Bible comes through in different
00:34:03
stories read by different people in
00:34:05
different languages everything that that
00:34:06
may be where we go like my kids want to
00:34:09
have a 10 minute bedtime story well let
00:34:10
me give them Peter Pan at 10 minutes I
00:34:12
want to do you know a chapter or a night
00:34:14
for my older daughter for a week long of
00:34:16
Peter Pan now I can do that and so the
00:34:19
way that I can kind of consume content
00:34:21
becomes different so I guess what I'm
00:34:23
saying is there's two aspects to the way
00:34:25
that I think the entire content the the
00:34:27
realm of content can be Rewritten
00:34:29
through AI the first is like individual
00:34:31
personalized creation of content where I
00:34:33
as a user can render content that of my
00:34:36
liking and my interest the second is
00:34:38
that I can engage with content that is
00:34:40
being created that is so much more
00:34:42
multi-dimensional than anything we
00:34:43
conceive of today we're current
00:34:45
centralized content creators now have a
00:34:47
whole set of tools now from a business
00:34:48
model perspective I don't think that
00:34:50
Publishers are really the play anymore
00:34:52
but I do think the platforms are going
00:34:53
to be the play and the platform tooling
00:34:55
that enables the individuals to do this
00:34:57
stuff and the platform tooling that
00:34:58
enables the content creators to do this
00:35:00
stuff are definitely entirely new
00:35:02
Industries and models that can create
00:35:04
multi-hundred billion dollar outcomes
00:35:06
let me hand this off to sax because
00:35:08
there has been the dream for everybody
00:35:10
especially in the Bay area of a hero
00:35:13
coming and saving Gotham City
00:35:16
and this has finally been realized David
00:35:20
sacks I did my own little Twitter AI
00:35:24
hashtag and I said to Twitter AI if only
00:35:28
please generate a picture of David Sax's
00:35:30
Batman crouched down on the bridge the
00:35:33
amount of creativity sacks that came
00:35:35
from this and this is
00:35:38
something that you know if we were
00:35:40
talking about just five years ago this
00:35:42
would be like a ten thousand dollar
00:35:43
image you could create a birthday these
00:35:46
were not professional quote unquote
00:35:47
artists these were individuals
00:35:49
individuals that were able to harness a
00:35:51
set of platform tools to generate this
00:35:53
incredible new content and I think it
00:35:55
speaks to the opportunity ahead and by
00:35:57
the way we're in inning one right so
00:35:59
you see yourself as Batman do you ever
00:36:01
think you should take your enormous
00:36:03
wealth and resources and put it towards
00:36:05
building a cave under your mansion that
00:36:08
lets you out underneath the Golden Gate
00:36:10
Bridge and you could go fight crime so
00:36:11
good do you want to go fight this crime
00:36:14
in Gotham
00:36:16
and I think San Francisco has a lot of
00:36:18
gotham-like qualities I think the
00:36:19
villains are more real than the heroes
00:36:21
unfortunately we don't have a lot of
00:36:23
Heroes but yeah we got a lot of jokers
00:36:25
Jokers yeah that's a whole separate
00:36:28
topic I'm sure a whole separate topic
00:36:30
we'll get to it at some point today you
00:36:31
guys are talking about all this stupid
00:36:33
[ __ ] like there are trillions of
00:36:35
dollars of software companies that could
00:36:36
get disrupted and you're talking about
00:36:37
making [ __ ] children's books and fat
00:36:39
pictures of socks it's so dumb no
00:36:41
special conversations
00:36:44
great job
00:36:46
cares about entertainment anymore
00:36:47
because it's totally obvious okay so one
00:36:49
of the biggest industries where the
00:36:50
money is why don't you teach people
00:36:52
where there's going to be actual
00:36:54
economic destruction
00:36:55
amazing economic destruction and
00:36:57
opportunity you spend all this time on
00:36:59
the most stupidest [ __ ] topics listen
00:37:01
it's an illustrative example no it's an
00:37:04
elitist example that you know it's
00:37:06
[ __ ] circle jerk yourself Batman's
00:37:08
not nobody nobody cares about movies
00:37:10
well let's bring nobody
00:37:12
tweet over everybody I mean I think I
00:37:14
think U.S box office is something like
00:37:16
20 billion a year I remember when like
00:37:18
they now got to like 100 billion a year
00:37:19
payment volume and now it's like
00:37:21
hundreds of billions so yeah and stripe
00:37:23
are going to process two trillion
00:37:25
dollars almost why don't you talk about
00:37:26
that disruption you ninny Market size of
00:37:28
U.S media and entertainment industry 717
00:37:30
billion okay it's not insignificant
00:37:33
video games are nearly half a trillion a
00:37:35
year yeah I mean this is the number
00:37:37
insignificant but let's pull up chamat's
00:37:39
tweet of course the dictator wants to
00:37:41
dictate here all this incredible
00:37:43
Innovation is being made and
00:37:45
a new Hero has been born chamath
00:37:48
polyhapatia a tweet that went viral over
00:37:51
1.2 million views already
00:37:53
I'll read your Tweet for the audience if
00:37:55
you invent a novel drug you need the
00:37:57
government to vet and approve it FDA
00:37:59
before you can commercialize it if you
00:38:00
invent a new mode of air travel you need
00:38:03
the government to vet and improve it FAA
00:38:04
I'm just going to edit this down a
00:38:06
little bit if you create new security
00:38:07
you need the government to vet it and
00:38:09
approve at SEC more generally when you
00:38:10
create things with broad societal impact
00:38:13
positive and negative the government
00:38:14
creates a layer to review and approve it
00:38:16
AI will need such an oversight body the
00:38:19
FDA approval process seems the most
00:38:21
credible and adaptable into a framework
00:38:23
to understand how
00:38:25
a model behaves and its counter factual
00:38:28
our political leaders need to get in
00:38:31
front of this sooner rather than later
00:38:32
and create some oversight before the
00:38:34
eventual big avoidable mistakes happen
00:38:36
and Genies are let out of the bottle
00:38:38
Tremont you really want the government
00:38:39
to come in and then when people build
00:38:43
these tools they have to submit them to
00:38:45
the government to approve them that's
00:38:46
what you're saying here and you want
00:38:48
that to start now
00:38:49
here's the alternative the alternative
00:38:51
is going to be the debacle that we know
00:38:53
as section 230. so if you try to write a
00:38:57
brittle piece of legislation or try to
00:39:00
use
00:39:01
old legislation to deal with something
00:39:04
new
00:39:05
it's not going to do a good job because
00:39:07
technology advances way too quickly
00:39:10
and so if you look at the section 230
00:39:12
example where have we left ourselves the
00:39:14
politicians have a complete inability to
00:39:17
pass a new framework to deal with social
00:39:19
media to deal with misinformation and so
00:39:21
now we're all kind of guessing what
00:39:24
a bunch of age 70 and 80 year old
00:39:27
Supreme Court Justices will do in trying
00:39:30
to rewrite technology law when they have
00:39:33
to apply on Section 230. so the point of
00:39:35
that tweet was to lay the Alternatives
00:39:38
there is no world in which this will be
00:39:41
unregulated
00:39:43
and so I think the question to ask
00:39:45
ourselves is do we want a chance for a
00:39:48
new body
00:39:49
so the FDA is a perfect example why even
00:39:52
though the FDA commissioner is appointed
00:39:54
by the president this is a quasi
00:39:56
organization it still arms length away
00:39:59
it has subject matter experts that they
00:40:02
hire and they have many Pathways to
00:40:05
approval some Pathways take days some
00:40:09
pathways are months and years some
00:40:11
pathways are for breakthrough Innovation
00:40:13
some pathways are for devices so they
00:40:16
have a broad spectrum of ways of of
00:40:18
arbitrating
00:40:19
what can be commercialized and what
00:40:21
cannot otherwise my prediction is we
00:40:24
will have a very brittle law that will
00:40:27
not work it'll be like the Commerce
00:40:29
department and the FTC
00:40:31
trying to gerrymander some old piece of
00:40:33
legislation and then what will happen is
00:40:36
it'll get escalated to the Supreme Court
00:40:38
and I think they are the last group of
00:40:40
people
00:40:41
who should be deciding on this
00:40:44
incredibly important topic for society
00:40:45
so
00:40:47
what I have been advocating our leaders
00:40:50
and I will continue to do so is
00:40:52
don't try to Ram this into an existing
00:40:54
body it is so important it is worth
00:40:57
creating a new organization like the FDA
00:41:01
and having a framework that allows you
00:41:03
to look at a model
00:41:04
and look at the counter factual judge
00:41:08
how good how important how disruptive it
00:41:10
is
00:41:11
and then release it in the wild
00:41:13
appropriately otherwise I think you'll
00:41:14
have these chaos GPT things
00:41:17
scale infinitely because again as
00:41:19
Friedberg said in Sac said you're
00:41:21
talking about one person that can create
00:41:22
this chaos multiply that by every person
00:41:26
that is an anarchist or every person
00:41:27
that just wants to sow seeds of chaos
00:41:30
and I think it's going to be all
00:41:31
avoidable I think regulating what
00:41:33
software people can write is a near
00:41:35
impossible task number one I think you
00:41:38
can probably put rules and restrictions
00:41:39
around Commerce right that's certainly
00:41:41
feasible uh in terms of how people can
00:41:44
monetize but in terms of writing and
00:41:46
utilizing software it's going to be as
00:41:49
challenged as trying to monitor
00:41:52
and demand
00:41:53
oversight and regulation around how
00:41:55
people write and use
00:41:57
tools for uh for genome and biology
00:42:00
exploration certainly if you want to
00:42:02
take a product to Market and sell a drug
00:42:04
to people that can influence their body
00:42:06
you have to go get that approved but in
00:42:08
terms of you know doing your work in a
00:42:10
lab it's very difficult I think the
00:42:12
other challenge here is
00:42:15
software can be written anywhere
00:42:17
it can be executed anywhere and so if
00:42:21
the US does try to regulate or does try
00:42:25
to put the brakes on the development of
00:42:27
tools where the U.S can have kind of a
00:42:29
great economic benefit and a great
00:42:31
economic interest
00:42:32
there will be advances made elsewhere
00:42:35
without a doubt and those markets and
00:42:37
those
00:42:39
those places will benefit in an
00:42:42
extraordinarily out outpaced way as we
00:42:44
just mentioned there's such
00:42:46
extraordinary kind of economic gain to
00:42:48
be realized here that if we're not
00:42:51
if the United States is not leading the
00:42:54
world we are going to be following we
00:42:56
are going to get disrupted we are going
00:42:57
to lose an incredible amount of value
00:42:58
and talent and so any attempt at
00:43:01
regulation or slowing down or telling
00:43:04
people that they cannot do things when
00:43:06
they can easily hop on a plane and go do
00:43:07
it elsewhere I think is is fraught with
00:43:10
Peril so you don't agree with regulation
00:43:12
sax are you on board with the chamoth
00:43:14
plan or you're on board with the Free
00:43:15
Bird well I'll say I think I think just
00:43:17
like with computer hacking it's illegal
00:43:19
to break into someone else's computer it
00:43:20
is illegal to steal someone's personal
00:43:22
information there are laws that are
00:43:24
absolutely simple and obvious and
00:43:28
you know no-nonsense laws those a lot of
00:43:31
legal to get rid of a hundred thousand
00:43:33
jobs by making a piece of software
00:43:34
though that's right and so I think
00:43:36
trying to intentionalize how we do
00:43:39
things versus intentionalizing
00:43:41
um the things that we want to prohibit
00:43:43
happening as an outcome we can certainly
00:43:45
try and prohibit the things that we want
00:43:46
to happen as an outcome and pass laws
00:43:48
and Institute governing bodies with
00:43:51
authority to oversee those laws
00:43:53
with respect to things like stealing
00:43:55
data but you can jump on a plane and go
00:43:57
do it in Mexico Canada or whatever
00:43:59
region you get to Saks where do you
00:44:01
stand on this debate yeah I'm saying
00:44:02
like there are ways to protect people
00:44:03
there's ways to protect Society about
00:44:05
passing laws that that make it illegal
00:44:07
to do things as the output is the
00:44:08
outcome what law do you pass on chaos
00:44:10
GPT explain chaos GPT Give an example
00:44:13
please yeah do you want to talk about it
00:44:14
real quick it's a recursive agent that
00:44:16
basically is trying to destroy itself
00:44:19
try to destroy Humanity yeah but I guess
00:44:22
by first becoming all-powerful and
00:44:23
destroying humanity and then destroying
00:44:25
itself yeah it's a tongue-in-cheek Auto
00:44:28
GPT
00:44:30
it's not a tongue-in-cheek auto GPT the
00:44:33
guy that created it you know put it out
00:44:34
there and said like he's trying to show
00:44:36
everyone to your point what
00:44:38
intentionality could arise here which is
00:44:39
negative intentionality I think it's
00:44:41
very naive for anybody
00:44:44
to think that this is not equivalent to
00:44:47
something that could cause harm to you
00:44:49
so for example if the prompt is hey here
00:44:52
is a security leak that we figured out
00:44:54
in Windows and so why don't you exploit
00:44:57
it so look a hacker now has to be very
00:44:59
technical
00:45:00
today with with these Auto gpts a hacker
00:45:03
does not need to be technical at all
00:45:04
exploit the zero day
00:45:07
exploit in Windows hack into this plane
00:45:09
and bring it down oh okay the GPT will
00:45:12
do it so who's going to tell you that
00:45:14
those things are not allowed who's going
00:45:15
to actually vet that that wasn't allowed
00:45:18
to be released in the wild so for
00:45:19
example if you worked with Amazon and
00:45:21
Google and Microsoft and said you're
00:45:23
going to have to run these things in a
00:45:25
sandbox and we're going to have to
00:45:27
observe the output before we allow it to
00:45:29
run on actual bare metal in the wild
00:45:32
again that seems like a reasonable thing
00:45:34
and it's super naive for people to think
00:45:36
it's a free market so we should just be
00:45:38
able to do what we want this will end
00:45:40
badly quickly and when the first plane
00:45:43
goes down and when the first [ __ ]
00:45:44
thing gets blown up all of you guys will
00:45:46
be like oh sorry Saks a pretty
00:45:48
compelling example here by chamoth
00:45:50
somebody puts out into the wild chaos
00:45:52
GPT you can go do a Google search for it
00:45:53
and says hey what are the
00:45:55
vulnerabilities to the electrical grid
00:45:58
compile those and automate a series of
00:46:01
attacks and write some code to probe
00:46:04
those until we and success in this
00:46:06
Mission you get a hundred points and
00:46:08
stars every time you Jason do this such
00:46:10
a it's such a beautiful example but it's
00:46:12
even more nefarious it is
00:46:14
hey this is an enemy that's trying to
00:46:17
hack our system so you need to hack
00:46:19
theirs and bring it down you know like
00:46:21
you can easily trick these gpts right
00:46:24
yes they have no judgment they have no
00:46:25
judgment and as you said they're
00:46:27
ruthless in in getting to the outcome
00:46:29
right
00:46:31
so why why do we think all of a sudden
00:46:33
this is not going to happen I mean it's
00:46:34
literally the science fiction example
00:46:36
you say Hey listen make sure no humans
00:46:38
get cancer and like okay well The
00:46:39
Logical way to make sure no humans get
00:46:40
cancer is to kill all the humans can you
00:46:43
just address the point so what do you
00:46:44
think you're regulating are you
00:46:45
regulating the code here's what I'm
00:46:47
saying to write if you look at the FDA
00:46:49
no you're allowed to make any chemical
00:46:51
drug you want but if you want to
00:46:52
commercialize it you need to run a
00:46:55
series of trials with highly qualified
00:46:58
measurable data and you submit it to
00:47:00
like-minded experts that are trained as
00:47:02
you are to evaluate the viability of
00:47:05
that and but hold on there are Pathways
00:47:08
that allow you to get that done in days
00:47:09
under emergency use and then there are
00:47:12
Pathways that can take years depending
00:47:14
on how gargantuan the task is at hand
00:47:16
and all I'm suggesting
00:47:18
is having some amount of oversight
00:47:21
is not bad in this specific example I
00:47:25
get what you're saying but I'm asking
00:47:26
tactically how what are you overseeing
00:47:28
you're overseeing cat GPT you're
00:47:30
overseas the the model you're doing
00:47:32
exactly chips okay look I used to run
00:47:35
the Facebook platform we used to create
00:47:37
sandboxes if you submit code to us
00:47:40
you would we would run it as a Sandbox
00:47:42
we would observe it we would figure out
00:47:43
what it was trying to do and we would
00:47:45
tell you this is allowed to run in the
00:47:46
wild there's a version of that that
00:47:48
Apple does when you submit an app for
00:47:50
review and approval Google does it as
00:47:53
well in this case all the bare metal
00:47:55
providers all the people that provide
00:47:57
gpus will be forced by the government in
00:48:00
my opinion to implement something and
00:48:03
all I'm suggesting is
00:48:04
that it should be a new kind of body
00:48:07
that essentially observes that has phds
00:48:10
that has people who are trained in this
00:48:12
stuff to develop the kind of testing and
00:48:15
the output that you need
00:48:16
to figure out whether it should even be
00:48:18
allowed to run in the Wild on bare metal
00:48:20
sorry but you're saying that the mod the
00:48:22
model sorry I'm just trying to
00:48:23
understand Shaman's points you're saying
00:48:24
that the models need to be reviewed by
00:48:26
this body and those models if they're
00:48:28
run on a third-party set of servers
00:48:36
you cannot run an app on your computer
00:48:38
you know that right it needs to be
00:48:40
connected to the internet right like if
00:48:41
you wanted to run an auto GPT it
00:48:43
actually crawls the internet it actually
00:48:45
touches other apis it tries to then
00:48:47
basically send a push request sees what
00:48:50
it gets back parses the Json figures out
00:48:52
what it needs to do all of that is
00:48:55
allowed because it's hosted by somebody
00:48:56
right that code is running not locally
00:49:00
so the host becomes sure if you want to
00:49:03
run it locally you can do whatever you
00:49:04
want to do but evil agents are going to
00:49:06
do that right so if I'm an evil agent
00:49:08
I'm not going to go use AWS to run my
00:49:10
evil agent I'm going to set up a bunch
00:49:12
of servers and connect it to the
00:49:13
internet how
00:49:15
I could use vpns the internet is open
00:49:17
there's openings
00:49:20
I think that what you're going to see is
00:49:23
that if you for example try to VPN and
00:49:25
run it out of like Tajikistan back to
00:49:27
the United States
00:49:29
it's not going to take years for us to
00:49:31
figure out that we need to IP block
00:49:32
Rando [ __ ] coming in push and pull
00:49:34
requests from all kinds of ips that we
00:49:36
don't trust anymore because we don't now
00:49:38
trust the regulatory oversight that they
00:49:40
have for code that's running from those
00:49:42
IPS that are not us domestications let
00:49:44
me steal man tremont's position for a
00:49:46
second Jason hold on I I think the
00:49:47
ultimate if what chamoth is saying is
00:49:50
the point of view of Congress and if
00:49:52
tomoth has this point of view then there
00:49:54
will certainly be people in Congress
00:49:55
that will adopt this point of view the
00:49:58
only way to ultimately do that degree of
00:50:00
Regulation and restriction is going to
00:50:02
be to restrict the open internet it is
00:50:04
going to be to have monitoring and
00:50:05
firewalls and safety protocols across
00:50:07
the open internet because you can have a
00:50:09
set of models running on any set of
00:50:10
servers sitting in any physical location
00:50:12
and as long as they can move data
00:50:14
packets around they're going to be able
00:50:16
to get up to their nefarious activities
00:50:18
let me still man that for you Freeburg I
00:50:21
think yes you're correct the internet
00:50:23
has existed in a very open way but there
00:50:25
are organizations and there are places
00:50:27
like the national highway traffic safety
00:50:29
administration if I were to steal
00:50:31
Management's position if you want to
00:50:33
manufacture a car and you want to make
00:50:36
one in your backyard and put it on your
00:50:37
track in on your land up in Napa
00:50:40
somewhere and you don't want to have
00:50:41
brakes on the car and you don't want to
00:50:43
have you know a speed limiter or airbags
00:50:46
or seat belts and you want to drive on
00:50:47
the hood of the car you can do that but
00:50:49
once you want it to go on the open road
00:50:50
the open internet you need to get you
00:50:53
need to submit it for some safety
00:50:54
standards like nht sa like Tesla has to
00:50:58
or Ford has to so sax where do you sit
00:51:00
on this or is let's assume that people
00:51:04
are going to do very bad things with
00:51:06
very powerful models that are becoming
00:51:09
available Amazon today said there'll be
00:51:10
Switzerland they're going to put a bunch
00:51:11
of llms and other models available on
00:51:13
AWS Bloomberg's llm Facebooks Google
00:51:17
bard and of course at gbt open Ai and
00:51:20
Bing all this stuff's available to have
00:51:21
access to that do you need to have some
00:51:24
regulation of who has access to those at
00:51:26
scale powerful tools should there be
00:51:28
some FDA or nhtsa I don't think we know
00:51:32
how to regulate it yet I think that's
00:51:34
too early and I think the harms that
00:51:36
we're speculating about we're making the
00:51:38
AI more powerful than it is and I
00:51:40
believe it will be that powerful but I
00:51:42
think that it's premature to be talking
00:51:43
about regulating something that doesn't
00:51:45
really exist yet take the chaos GPT
00:51:46
scenario
00:51:48
the way that would play out would be
00:51:51
you've got some future incarnation of
00:51:53
Auto GPT
00:51:54
and somebody says okay Auto GPT I want
00:51:57
you to be you know wmdai and figure out
00:52:01
how to cause like a mass destruction
00:52:03
event you know and then it creates like
00:52:05
a planning checklist and that kind of
00:52:06
stuff
00:52:07
so that's basically the the type of
00:52:10
scenario we're we're talking about we're
00:52:12
not anywhere close to that yet I mean
00:52:14
the chaos GPT is kind of a joke it
00:52:17
doesn't produce it doesn't produce a
00:52:20
checklist I can give an example that
00:52:21
would actually
00:52:22
be completely plausible one of the first
00:52:25
things on the chaos gpt's checklist was
00:52:27
to stay within the boundaries of the law
00:52:29
because it didn't want to get prosecuted
00:52:30
got it so the person who did that had
00:52:32
some sort of good intent but I can give
00:52:34
you an example right now
00:52:36
that could be done by chat GPT and auto
00:52:38
GPT that could take down large swaths of
00:52:40
society and cause massive destruction
00:52:42
I'm almost reticent to say it here say
00:52:44
it uh well I'll say it and then maybe
00:52:45
we'll have to delete this but if
00:52:47
somebody created this and they said uh
00:52:49
figure out a way to compromise as many
00:52:52
powerful peoples and as many systems
00:52:54
passwords then go in there and delete
00:52:56
all their files and turn off as many
00:52:58
systems as you can
00:53:00
Chachi PT and auto GPT could very easily
00:53:02
create phishing accounts create billions
00:53:05
of websites to create billions of logins
00:53:07
have people log into them get their
00:53:09
passwords log into whatever they do and
00:53:11
then delete everything in their account
00:53:12
which would cause chaos and it could be
00:53:15
done today I don't think we've done
00:53:17
today simpler than this how about how
00:53:18
about you fishing website yeah pieces of
00:53:21
it can be created today but you're
00:53:22
you're accelerating the progress yeah
00:53:25
but you can automate 30 days yeah
00:53:28
exactly and but I think I'm accelerating
00:53:30
it in weeks why don't you just spoof the
00:53:33
bank accounts and just steal the money
00:53:34
like that's even simpler like people
00:53:36
will do the stuff because they're trying
00:53:37
to do it today holy cow now they just
00:53:39
have a more efficient way to solve
00:53:41
somebody think about bank accounts geez
00:53:42
so number one this is a tool and if
00:53:44
people use a tool in nefarious ways you
00:53:46
prosecute them number two the platforms
00:53:49
that are commercializing these tools do
00:53:51
have trust and safety teams now in the
00:53:54
past trust and safety has been a
00:53:56
euphemism for censorship which it
00:53:58
shouldn't be but you know open AI has a
00:54:01
safety team and they try to detect when
00:54:03
people are using their Tech in a
00:54:05
nefarious way and they try to prevent it
00:54:08
well no not on censorship but I think
00:54:11
that they're probably a million people
00:54:13
are using chapters I think they're
00:54:15
policing it are you willing to abdicate
00:54:17
your or societal responsibility to to
00:54:20
open AI to do the trust sensation what
00:54:22
I'm what I'm saying is I'd like to see
00:54:24
how far we get in terms of the system
00:54:27
yes you're saying you want to see the
00:54:29
mistakes you want to see where the
00:54:30
mistakes are and how bad had the
00:54:32
mistakes are I'm saying it's still very
00:54:33
early to be imposing regulation we don't
00:54:35
even know what to regulate so I think we
00:54:37
have to keep tracking this to develop
00:54:39
some understanding of how it might be
00:54:41
misused how the industry is going to
00:54:43
develop safety guard rails okay and then
00:54:47
you can talk about regulation look you
00:54:49
create some new FDA right now okay first
00:54:51
of all we know what would happen look at
00:54:53
the drug process as soon as the FDA got
00:54:55
involved it slowed down massively now it
00:54:57
takes years many years to get a drug
00:54:59
approved appropriately so yes but at
00:55:02
least with a drug we know what the gold
00:55:04
standard is you run a double-blind study
00:55:06
to see whether it causes harm or whether
00:55:09
it's beneficial we don't know what that
00:55:11
standard is for AI yet we have no idea
00:55:14
what's going to happen in a double blind
00:55:16
study in AI what no you'd have somebody
00:55:19
review the code you have two instances
00:55:24
Auto GPT
00:55:27
it's benign I mean
00:55:28
my friend used it to book a wine tasting
00:55:32
so who's going to review that code and
00:55:34
then speculate say oh well 99.9 of cases
00:55:38
it's perfectly benevolent and fine and
00:55:42
innocuous you know I can fantasize about
00:55:44
some cases someone might do hold on how
00:55:46
are you supposed to resolve that very
00:55:48
simple there are two types of Regulation
00:55:49
that can occur in any industry you can
00:55:51
do with the movie industry did which is
00:55:53
they self-regulate and they came up with
00:55:55
their own rating system or you can do
00:55:56
what happens with the FDA and what
00:55:58
happens with cars which is an external
00:56:01
government-based body I think now is the
00:56:03
time for self-regulation so that we
00:56:05
avoid the massive heavy hand of
00:56:08
government having to come in here but
00:56:09
these tools can be used today to create
00:56:11
massive Farm they're moving at a pace we
00:56:13
just said in the first half of the show
00:56:15
that none of us have ever seen every 48
00:56:17
Hours something drops that is
00:56:19
mind-blowing that's never happened
00:56:20
before and you can
00:56:22
take these tools and in the one example
00:56:25
that shmoth and I came up with the top
00:56:27
of our head in 30 seconds you could
00:56:30
create phishing sites compromise
00:56:31
people's bank accounts take all the
00:56:33
money out Delete all the files and cause
00:56:34
chaos on a scale that has never been
00:56:37
possible by a series of Russian hackers
00:56:40
or Chinese hackers working in a boiler
00:56:42
room this can scale and that is the the
00:56:45
fundamental difference here and I didn't
00:56:46
think I would be sitting here Steel Man
00:56:48
in tremont's argument I think humans
00:56:49
have a horrible ability to compound I
00:56:52
think people do not understand compound
00:56:53
interest and this is a perfect example
00:56:55
where when you start to compound
00:56:56
technology at the rate of 24 hours or 48
00:56:59
hours which we've never really had to
00:57:01
acknowledge most people's brains break
00:57:03
and they don't understand what six
00:57:04
months from now looks like and six
00:57:06
months from now when you're compounding
00:57:08
at 48 or 72 hours is like 10 to 12 years
00:57:12
in other Technology Solutions this is
00:57:15
compounding this is this is different
00:57:16
because of the compounding I agree with
00:57:18
that the pace Revolution is very fast we
00:57:20
are on a bullet train to something and
00:57:22
we don't know exactly what it is and
00:57:24
that's disconcerting however let me tell
00:57:26
you what would happen if we create a new
00:57:27
regulatory body like the FDA to regulate
00:57:29
this they would have no idea how to
00:57:32
arbitrate whether a technology should be
00:57:34
approved or not development will
00:57:36
basically slow to a crawl to slight drug
00:57:37
development there is no double-blind
00:57:39
standard I agreement can we do what
00:57:42
self-regulation can we do there is no
00:57:43
double blind standard in AI that
00:57:46
everyone can agree on right now to know
00:57:48
whether something should be approved and
00:57:49
what's going to happen is the thing
00:57:51
that's made software development so
00:57:52
magical and allowed all this Innovation
00:57:55
over the last 25 years is permissionless
00:57:57
innovation any developer
00:58:00
any Dropout from a university can go
00:58:03
create their own project which turns
00:58:05
into a company and that is what has
00:58:07
driven all the Innovation and progress
00:58:09
in our economy over the last 25 years so
00:58:11
you're going to replace permissionless
00:58:13
Innovation with going to Washington to
00:58:14
go through some approval process and it
00:58:16
will be the politically connected it'll
00:58:18
be the big donors who get their projects
00:58:21
approved and the next Mark Zuckerberg
00:58:23
who's trying to do his little project in
00:58:24
a dorm room somewhere will not know how
00:58:27
to do that well not know how to compete
00:58:28
and that highly political process I
00:58:31
think you're mixing a bunch of things
00:58:32
together so first of all permissionless
00:58:34
Innovation happens today in biotech as
00:58:37
well it's just that it's what Jason said
00:58:39
when you want to put it on the rails of
00:58:41
society and make it available to
00:58:43
everybody you you actually have to go
00:58:45
and do something substantive
00:58:47
in the negotiation of these drug
00:58:49
approvals it's not some standardized
00:58:51
thing you actually sit with the FDA and
00:58:52
you have to decide what are our
00:58:54
endpoints what is the mechanism of
00:58:55
action and how will we measure the
00:58:57
efficacy of this thing
00:58:59
the idea that you can't do this today in
00:59:01
AI is laughable yes you can and I think
00:59:03
that smart people so for example if you
00:59:05
pit deep Minds team versus open ai's
00:59:08
Team to both agree that a model is good
00:59:10
and correct I bet you they would find
00:59:13
a systematic way to test that it's fine
00:59:15
I just want to point out okay so
00:59:17
basically in order to do what you're
00:59:19
saying okay this entrepreneur
00:59:21
who just dropped out of college to do
00:59:23
their project they're gonna have to
00:59:24
learn how to go sit with Regulators have
00:59:26
a conversation with them go through some
00:59:28
complicated approval process and you're
00:59:30
trying to say that that won't turn into
00:59:32
a game of political connections of
00:59:34
course it will of course it will of
00:59:36
course
00:59:37
which is self-regulation yeah well let's
00:59:40
get to that hold on a second and let's
00:59:42
look at the drug approval process if you
00:59:45
want to create a drug company you need
00:59:46
to raise hundreds of millions of dollars
00:59:48
it's incredibly expensive it's
00:59:50
incredibly Capital intensive there is no
00:59:53
drug company that is two guys in their
00:59:55
garage like many of the biggest
00:59:57
companies like many of the biggest
00:59:59
companies in Silicon Valley started that
01:00:01
is because you're talking about taking a
01:00:04
chemical or biological compound and
01:00:06
injecting into some hundreds or
01:00:08
thousands of people who are both
01:00:11
racially
01:00:12
gender-based age-based highly stratified
01:00:15
all around the world or at a minimum all
01:00:17
around the country you're not talking
01:00:19
about that here David I think that you
01:00:21
could have a much simpler and cheaper
01:00:23
way where you have a version of the
01:00:26
internet that's running in a huge
01:00:27
sandbox someplace that's closed off from
01:00:29
the rest of the internet and another
01:00:30
version of the internet that's closed
01:00:32
off from everything else as well and you
01:00:34
can run on a parallel path
01:00:35
as it is with this agent and you can
01:00:38
easily in my opinion actually figure out
01:00:40
whether this agent is good or bad and
01:00:42
you can probably do it in weeks so I
01:00:45
actually think the approvals are
01:00:46
actually not that complicated and the
01:00:48
reason to do it here is because I get
01:00:51
that it may cause a little bit more
01:00:52
friction for some of these Mom and Pops
01:00:56
but if you think about what's the
01:00:59
societal and consequences of letting the
01:01:02
worst case outcomes happen the AGI type
01:01:05
outcomes happen I think those are so bad
01:01:09
they're worth slowing some folks down
01:01:11
and I think like just because you want
01:01:12
to you know buy groceries for a hundred
01:01:15
dollars you should be able to do it I
01:01:16
get it but if people don't realize and
01:01:19
connect the dots between that and
01:01:20
bringing airplanes down then that's
01:01:22
because they don't understand what this
01:01:23
is capable of I'm not saying we're never
01:01:25
going to need regulation what I'm saying
01:01:27
is it's way too early we don't even know
01:01:29
what we're accolading we don't know what
01:01:31
the standard would be and what we will
01:01:33
do by racing to create a new FDA is
01:01:35
destroying American innovation in the
01:01:37
sector and other countries will not slow
01:01:39
down they will beat us to the puncher
01:01:40
got it I think there's a middle ground
01:01:43
here
01:01:44
of self-regulation and thoughtfulness on
01:01:46
the part of the people who are providing
01:01:48
these tools at scale to give just but
01:01:50
one example here and this tweet is from
01:01:52
five minutes ago so to look at the pace
01:01:54
of this five minutes ago this tweet came
01:01:57
out a developer who is an AI developer
01:02:00
says AI agents continue to amaze my gpt4
01:02:03
coding assistant learned how to build
01:02:04
apps with authenticated users that can
01:02:06
build and design a web app create a back
01:02:08
end handle off logins upload code to
01:02:12
GitHub and deploy
01:02:14
he literally while we were talking is
01:02:17
deploying websites now if this website
01:02:19
was a phishing app or the one that
01:02:22
shamop is talking about he could make a
01:02:25
gazillion different versions of banking
01:02:27
of America Wells Fargo et cetera then
01:02:29
find everybody on the internet's email
01:02:31
then start sending different spoofing
01:02:33
emails determine which spoofing emails
01:02:35
work iterate on those and create a
01:02:37
Global Financial collapse now this
01:02:38
sounds insane but it's happening right
01:02:40
now people get hacked every day at one
01:02:42
two three percent
01:02:44
sax fraud is occurring right now in the
01:02:47
low single digit percentages identity
01:02:49
theft is happening in the low single
01:02:50
identity percentages this technology is
01:02:53
moving so fast that bad actors could 10x
01:02:56
that relatively easy and so if 10 of us
01:02:59
want to be hacked and have our credit
01:03:00
card attacked this could create chaos I
01:03:03
think self-regulation is the solution
01:03:05
I'm the one who brought up
01:03:06
self-regulation what I said no I brought
01:03:07
it up first I brought it up first I get
01:03:09
credit no go ahead no it's not about
01:03:11
credit I'm no self-regulations
01:03:15
you talk for eight minutes so if you
01:03:17
have a point to make you should have got
01:03:18
in the eight minutes oh my God you guys
01:03:20
kept interrupting me go ahead what I
01:03:22
said is that there are trust and safety
01:03:25
teams at these big AI companies these
01:03:27
big foundation model companies like open
01:03:30
AI like I said
01:03:32
in the past trust and safety has been a
01:03:34
euphemism for censorship and that's why
01:03:35
people don't trust it but I think it
01:03:37
would be appropriate for these platform
01:03:40
companies to apply some guard rails on
01:03:42
how their tools can be used and based on
01:03:45
everything I know they're doing that so
01:03:47
this guy just released websites to the
01:03:49
Austin web with chat gp4 and he's going
01:03:51
to have it do it automated you're
01:03:52
basically postulating capabilities that
01:03:55
don't yet exist I just tweeted the guy
01:03:57
is doing it he's got a video of himself
01:03:58
doing it on the web what do you think
01:03:59
that's a far cry from basically running
01:04:02
like some fishing
01:04:04
Expedition that's going to bring down
01:04:06
the entire banking system uh literally a
01:04:08
fishing a fishing site and a
01:04:10
ns are the same thing go ahead Freeburg
01:04:12
I think that that guy
01:04:14
is doing something illegal if he's
01:04:16
hacking into computers uh into people's
01:04:19
emails and bank accounts that's illegal
01:04:21
you're not allowed to do that
01:04:23
and so that action breaks the law that
01:04:27
person can be prosecuted for doing that
01:04:29
the tooling that one might use to do
01:04:31
that
01:04:32
can be used in a lot of different ways
01:04:34
just like you could use Microsoft Word
01:04:37
to forge letters just like you could use
01:04:40
Microsoft Excel to create fraudulent
01:04:43
financial statements I think that the
01:04:44
application of a platform technology
01:04:48
needs to be distinguished from
01:04:50
the technology itself and while we all
01:04:53
feel extraordinarily fearful because the
01:04:55
unbelievable leverage that these AI
01:04:57
tools provide again I'll remind you that
01:05:00
the chat gpt4 or this gpt4 model
01:05:05
by some estimates is call it a few
01:05:06
terabytes you could store it on a hard
01:05:08
drive or you could store it on your
01:05:09
iPhone and you could then go run it on
01:05:11
any set of servers that you could go set
01:05:13
up physically anywhere so you know it's
01:05:16
a little bit naive to say we can go
01:05:18
ahead and you know regulate platforms
01:05:20
and we can go regulate the tools
01:05:21
certainly we should continue to enforce
01:05:24
and protect ourselves against nefarious
01:05:27
actors using you know new Tools in
01:05:29
inappropriate illegal ways
01:05:31
you know I I also think that there's a
01:05:34
moment here
01:05:35
that we should all kind of
01:05:38
observe just how quickly we want to shut
01:05:40
things down when
01:05:43
you know they take away what feels like
01:05:46
the the control that we all have
01:05:49
from one day to the next and you know
01:05:52
that the the real kind of sense of fear
01:05:56
that seems to be quite contagious for a
01:05:58
large number of people that have
01:06:00
significant assets or significant things
01:06:03
to lose
01:06:04
is that uh you know tooling that's
01:06:06
that's you know creating entirely newly
01:06:09
disruptive systems and models for
01:06:11
business and and economics
01:06:13
an opportunity for so many
01:06:15
needs to be regulated away to minimize
01:06:18
you know what we claim to be some
01:06:20
potential downside when we already have
01:06:22
laws that protect us on the other side
01:06:25
so you know I just kind of
01:06:27
want to also consider that this set of
01:06:30
tools creates extraordinary opportunity
01:06:32
we gave one sort of simple example about
01:06:34
the opportunity for creators but we
01:06:36
talked about how new business models new
01:06:38
businesses can be started with one or
01:06:40
two people
01:06:41
you know entirely new tools can be built
01:06:43
with a handful of people entirely new
01:06:45
businesses this is an incredible
01:06:47
Economic Opportunity and again if the
01:06:50
U.S tries to regulate it or the U.S
01:06:51
tries to come in and stop the
01:06:53
application of models in general or
01:06:54
regulate models in general you're
01:06:56
certainly going to see those models of
01:06:57
continue to evolve and continue to be
01:06:59
utilized in very powerful ways that are
01:07:02
going to be advantageous
01:07:03
to places outside the U.S there's over
01:07:05
180 countries on Earth they're not all
01:07:07
going to regulate together it's been
01:07:09
hard enough to get any sort of
01:07:11
coordination around Financial systems to
01:07:13
get coordination around climate change
01:07:15
to get coordination around anything on a
01:07:17
global basis to try and get coordination
01:07:19
around the software models that are
01:07:21
being developed I think is is pretty
01:07:23
naive you don't want to have a global
01:07:25
organization I think you need to have a
01:07:26
domestic organization that protects U.S
01:07:28
and I think Europe will have their own
01:07:30
they again FDA versus Emma Canada has
01:07:34
its own Japan has its own China has its
01:07:36
own and they they have a lot of overlap
01:07:39
and a lot of commonality in in the
01:07:40
guardrails they use and I think that's
01:07:42
what's going to happen here this will be
01:07:43
beneficial only for political insiders
01:07:45
who will basically be able to get their
01:07:47
projects and their apps approved with a
01:07:49
huge dead weight loss for the system
01:07:50
because Innovation will completely slow
01:07:51
down but to many build on freeburg's
01:07:53
point which is that
01:07:56
we have to remember that AI won't just
01:07:58
be used by nefarious actors it'll be
01:08:00
used by positive actors so there will be
01:08:03
new tools that law enforcement will be
01:08:05
able to use and if somebody's creating
01:08:07
phishing sites at scale they're going to
01:08:09
be probably pretty easy for you know law
01:08:11
enforcement AIS to detect so let's not
01:08:14
forget that there'll be co-pilots
01:08:15
written for our law enforcement
01:08:18
authorities they'll be able to use that
01:08:19
to basically detect and fight crime and
01:08:21
a really good example of this is in the
01:08:23
crypto space we saw this article over
01:08:24
the past week that chain analysis has
01:08:27
figured out how to basically track you
01:08:30
know illicit Bitcoin transactions and
01:08:32
there's now a huge number of
01:08:33
prosecutions that are happening of
01:08:36
illegal use of Bitcoin and if you go
01:08:38
back to when Bitcoin first took off
01:08:40
there was a lot of conversations around
01:08:43
Silk Road and the only thing that
01:08:44
Bitcoin was good for was basically
01:08:46
illegal transactions blackmailing drug
01:08:49
trafficking and therefore we have to
01:08:51
stop Bitcoin remember that was the main
01:08:53
argument and the counter argument was
01:08:56
that well no Bitcoin like any technology
01:08:59
can be used for good or bad however
01:09:00
there will be technologies that spring
01:09:03
up
01:09:03
to combat those nefarious or illicit use
01:09:06
cases and sure enough you had a company
01:09:08
like chain analysis come along and now
01:09:10
it's been used by law enforcement to
01:09:12
basically crack down on the illicit use
01:09:14
of Bitcoin and if anything it's cleaned
01:09:16
up the Bitcoin Community tremendously
01:09:18
and I think it's dispelled this idea
01:09:20
that the only thing you'd use Bitcoin
01:09:22
for is Black Market transactions quite
01:09:25
the contrary I think you'd be really
01:09:26
stupid now to use Bitcoin in that way
01:09:29
it's actually turned Bitcoin into
01:09:30
something of a Honeypot now because if
01:09:33
you used it for nefarious transactions
01:09:35
your transactions recording the
01:09:37
blockchain forever or just waiting for
01:09:38
chain analysis to find it so again using
01:09:42
Bitcoin to do something illegal be
01:09:43
really stupid I think in a similar way
01:09:45
you're going to see self-regulation by
01:09:47
these major AI platform companies
01:09:49
combined with new tools that are used
01:09:51
new AI tools does spring up to help
01:09:54
combat the nefarious uses and until we
01:09:57
let those forces play out
01:09:59
I'm not saying regulate never I'm just
01:10:01
saying we need to let those forces play
01:10:03
out before we leap to creating some new
01:10:06
regulatory body that doesn't even
01:10:08
understand what its mandated mission's
01:10:09
supposed to be the Bitcoin story is
01:10:11
hilarious by the way
01:10:12
oh my God Wall Street Journal story it's
01:10:15
unbelievable pretty epic it took years
01:10:17
but basically this guy was buying blow
01:10:19
on suck Road and he deposited his
01:10:23
Bitcoin and then when he withdrew it he
01:10:25
there was a bug that gave him twice as
01:10:27
many Bitcoins so he kept creating more
01:10:29
accounts putting more money into Silk
01:10:31
Road and getting more Bitcoin out
01:10:33
and then years later the authorities
01:10:35
figured this out again with you know
01:10:37
chain analysis type things look at James
01:10:39
zong over there same song he accused uh
01:10:43
had a Lamborghini a Tesla a lake house
01:10:45
uh and was living his best life
01:10:48
apparently when the feds uh knocked on
01:10:51
his door and found the digital keys to
01:10:54
his crypto Fortune
01:10:55
in a popcorn tin in his bathroom and in
01:10:58
a safe in his basement
01:11:01
floor so there you have it well the
01:11:04
reason the reason I posted this was I
01:11:05
was like what if this claim that you
01:11:09
could have all these Anonymous
01:11:10
transactions actually fold
01:11:13
the entire Market because it looks like
01:11:17
that this anonymity has effectively been
01:11:20
reverse engineered and there's no
01:11:22
anonymity at all and so what Bitcoin is
01:11:25
quickly becoming is like the most
01:11:27
singular Honeypot of transactional
01:11:30
information that's complete and
01:11:32
available in public and I think what
01:11:35
this article talks about is how
01:11:36
companies like chain analysis and others
01:11:38
have worked now for years almost a
01:11:41
decade with law enforcement to be able
01:11:44
to map all of it and so now every time
01:11:48
money goes from one Bitcoin wallet to
01:11:49
another they effectively know the sender
01:11:52
and the recipient and I just want to
01:11:53
make a one quick correction here it
01:11:55
wasn't actually exactly popcorn it was
01:11:58
Cheetos spicy flavored popcorn
01:12:01
and there's the tin of it
01:12:03
where he had a motherboard of a computer
01:12:05
that held is there a chance that that
01:12:07
this project was actually introduced by
01:12:10
the government I mean there's been
01:12:12
reports
01:12:13
or network that the CIA had their hands
01:12:17
all over to our Tor if you don't know it
01:12:19
which is an anonymous like multi-relay
01:12:22
peer-to-peer web browsing system and
01:12:25
people believe it's a CIA Honeypot an
01:12:29
intentional trap for criminals to get
01:12:32
themselves uh caught up in
01:12:35
all right as we wrap here what an
01:12:37
amazing discussion my Lord I didn't I
01:12:38
never thought I would be I want to say
01:12:40
one thing yes we saw that
01:12:44
someone was arrested for the murder of
01:12:46
Bob Lee that's what I was about to
01:12:48
answer this morning yeah which turns out
01:12:50
that the report of the sfpd's arrest is
01:12:53
that it's uh someone that he knew that
01:12:55
also works in the tech industry someone
01:12:56
that possibly know right so also
01:12:58
breaking news yes possibly
01:13:00
but I I want to say two things one
01:13:03
obviously based on this arrest and the
01:13:05
storyline it's quite different than what
01:13:07
we all assumed it to be which was some
01:13:09
sort of homeless robbery type moment
01:13:12
that has become all too commonplace in
01:13:14
SF it's a commentary for me on two
01:13:17
things one is how quick we all were to
01:13:20
kind of Judge and assume that you know a
01:13:23
homeless robber type person would do
01:13:25
this in SF which I think speaks to the
01:13:28
condition in SF right now
01:13:30
also speaks to our conditioning that
01:13:32
that we all kind of lacked or didn't
01:13:34
even want to engage in a conversation
01:13:36
that maybe this person was murdered by
01:13:37
someone that they knew
01:13:40
because we wanted to kind of very
01:13:42
quickly fill our own narrative about how
01:13:43
bad SF is
01:13:45
and that's just something that I really
01:13:46
felt when I read this this morning I was
01:13:47
like man like I didn't even consider the
01:13:49
possibility that this guy was murdered
01:13:52
by someone that he knew because I am so
01:13:54
enthralled right now by this narrative
01:13:56
that SF is so bad and it must be another
01:13:58
data point that validates my point of
01:13:59
view on SF so you know I kind of want to
01:14:02
just acknowledge that and acknowledge
01:14:03
that we all kind of do that right now
01:14:05
but I do think it also does in fact
01:14:06
unfortunately speak to how bad things
01:14:08
are in SF because we all are we've all
01:14:11
have these experiences of feeling like
01:14:12
we're in danger and under Threat all the
01:14:14
time we're walking around in SF uh in so
01:14:16
many parts of San Francisco I should say
01:14:18
where things feel like they've gotten
01:14:19
really bad I think both things can be
01:14:22
true that we can kind of feel biased and
01:14:25
fill our own narrative
01:14:27
by kind of latching on to our assumption
01:14:30
about what something tells us
01:14:32
but but it also tells us quite a lot
01:14:33
about what is going on in ourselves so I
01:14:35
I just I just wanted to make sure In
01:14:37
fairness and I think it's fine for you
01:14:38
to make that point I am extremely
01:14:41
Vigilant on this program to always say
01:14:42
when something is breaking news with
01:14:44
whole judgment whether it's the Trump
01:14:45
case or Jesse Smollett or anything in
01:14:47
between January 6th let's wait until we
01:14:49
get all the facts and in fact quote from
01:14:52
sacks
01:14:53
we don't know exactly what happened yet
01:14:56
correct literally sax started with that
01:14:59
we do that every [ __ ] time on this
01:15:01
program we know when there's breaking
01:15:04
news to withhold judgment but you can
01:15:06
also know
01:15:07
two things can be true a tolerance for
01:15:10
ambiguity is necessary but I'm saying I
01:15:12
didn't even do that as soon as I heard
01:15:14
this I was like I was like an assumption
01:15:16
but David that is a fine assumption to
01:15:19
make that's a fine essential technical
01:15:21
assumption listen you make that
01:15:23
assumption for your own protection we
01:15:25
got all these reporters who are
01:15:27
basically propagandists trying to claim
01:15:28
that crime is down in San Francisco
01:15:29
they're all basically seeking comment
01:15:31
from me this morning sending emails or
01:15:33
trying to dunk God on us because we
01:15:35
basically talked about the bubbly case
01:15:38
in that way listen we said that we
01:15:41
didn't know what happened but if we were
01:15:43
to bet at least what I said is I bet
01:15:46
this case it looks like a lot like the
01:15:47
Brianna cup for case that was logical
01:15:49
that's not conditioning or biased that's
01:15:51
logic and you need to look at what else
01:15:54
happened that week okay so just the same
01:15:57
week that bubbly was killed let me give
01:15:59
you three other examples of things that
01:16:01
happened in Gotham City AKA San
01:16:03
Francisco so number one former fire
01:16:07
commissioner Don carmignani was beaten
01:16:09
within an inch of his life by a group of
01:16:11
homeless addicts in the marina and one
01:16:14
of them was interviewed in terms of why
01:16:17
it happened and basically Don came down
01:16:19
from his mother's house and told them to
01:16:21
move off his mother's front porch
01:16:23
because they were obstructing her
01:16:24
ability to get in and out of her
01:16:25
apartment they interpreted that as
01:16:27
disrespect and they beat him with a tire
01:16:29
iron or a metal pipe and one of the
01:16:32
hoodlums who was involved in this
01:16:34
apparently admitted this yeah play the
01:16:36
video somebody over the head like that
01:16:38
and attack him as he was really
01:16:42
disrespectful he
01:16:49
so he was being disrespectful and then
01:16:52
but is that enough to beat him up yeah
01:16:56
sometimes oh my Lord I mean so this is
01:16:59
case number one and apparently in the
01:17:01
reporting on that person who was just
01:17:04
interviewed he's been in the marina kind
01:17:05
of terrorizing people maybe not
01:17:07
physically but verbally
01:17:09
so you have you know bands of homeless
01:17:12
people encamped in front of people's
01:17:14
houses Don carmignani gets beaten within
01:17:17
an inch of his life you then had the
01:17:19
case of the Whole Foods Store on Market
01:17:21
Street shut down in San Francisco and
01:17:24
this was not a case of shoplifting like
01:17:26
some of the other store closings we've
01:17:27
seen they said they were closing the
01:17:29
store because they could not protect
01:17:31
their employees the bathrooms were
01:17:33
filled with needles and pipes that were
01:17:36
drug paraphernalia you had drug acts
01:17:38
going in there using it they were
01:17:39
engaging in altercations with store
01:17:41
employees and Whole Foods felt like they
01:17:43
had to close the store because again
01:17:44
they could not protect their employees
01:17:46
third example Board of Supervisors had
01:17:50
to disband their own meeting because
01:17:52
their internet connection got vandalized
01:17:55
the fiber for the cable connection to
01:17:58
provide their internet got vandalized so
01:18:00
they had to basically disband their
01:18:01
meeting Aaron Prescott was the one who
01:18:03
announced this and you saw in the
01:18:04
response to this yeah my retweetingham
01:18:07
went viral there were lots of people
01:18:09
said yeah I've got a small business and
01:18:11
the fiber or the copper wire whatever
01:18:13
was vandalized and in a lot of cases I
01:18:16
think it's basically drug addicts
01:18:17
stealing whatever they can they steal
01:18:18
ten dollars of copper wire sell that to
01:18:20
get a hit and it causes forty thousand
01:18:23
dollars of property damage here's the
01:18:25
insincerity sex literally the proper
01:18:28
response when there's violence in San
01:18:30
Francisco is hey we need to make this
01:18:32
place less violent is there a chance
01:18:33
that it could be people who know each
01:18:35
other of course that's inherent in any
01:18:37
crime that occurs that there'll be time
01:18:39
to investigate it but literally the
01:18:41
Press is now using this as a moment to
01:18:44
say there's no crime in San Francisco
01:18:45
where that people are acting like I just
01:18:48
have the New York Times email me during
01:18:49
the podcast
01:18:51
Heather knight from The Chronicle San
01:18:53
Francisco Chronicle in light of the Bob
01:18:55
Lee killing appearing to be an
01:18:56
interpersonal dispute she still doesn't
01:18:58
know right we don't have all the facts
01:19:00
with another tech leader do you think
01:19:01
the Tech Community jumped to conclusions
01:19:03
why are so many Tech leaders painting
01:19:05
San Francisco as a dystopian hellscape
01:19:07
with the reality with the reality is
01:19:10
more nuanced I think it's a little typo
01:19:11
there yeah yes it's like of course the
01:19:14
reality is nuanced of course it's a
01:19:16
hellscape walk down the street Heather
01:19:20
can I give you a theory please
01:19:23
I think it was most evident in the way
01:19:26
that Elon
01:19:28
dismantled and manhandled the BBC
01:19:30
reporter oh my God that was brutal this
01:19:33
is a small microcosm of what I think
01:19:35
media is so I used to think that media
01:19:39
had an agenda
01:19:42
I actually now think that they don't
01:19:44
particularly have an agenda
01:19:46
other than to be relevant because they
01:19:49
see waning relevance and so I think what
01:19:53
happens is whenever there are a bunch of
01:19:55
articles that tilt a pendulum into a
01:19:58
narrative
01:19:59
they all of a sudden become very focused
01:20:02
on refuting that narrative
01:20:05
and even if it means they have to lie
01:20:08
they'll do it
01:20:09
right so you know I think for months and
01:20:11
months I think people have seen that the
01:20:13
quality of the discourse on Twitter
01:20:15
became better and better
01:20:16
Elon is doing a lot with Bots and all of
01:20:18
this stuff cleaning it up
01:20:20
and this guy had to try to establish the
01:20:23
counter narrative and was willing to lie
01:20:26
in order to do it then he was dismantled
01:20:28
here you guys I don't have a bone to
01:20:30
pick so much at San Francisco I think
01:20:32
I've been relatively silent on this
01:20:33
topic but you guys as residents and
01:20:36
former residents I think have a vested
01:20:37
interest in the quality of that City and
01:20:39
you guys have been very vocal but I
01:20:41
think that you're not the only ones
01:20:42
Michelle tandler you know sheldonberger
01:20:44
there's a bunch of smart thoughtful
01:20:46
people who've been beating this drum
01:20:49
Gary tan
01:20:51
and so now I think reporters don't want
01:20:54
to write the N plus first article saying
01:20:56
that San Francisco is a hellscape
01:20:58
so they have to take the other side
01:21:00
and so now they're going to go and pick
01:21:02
up the counter narrative and they'll
01:21:03
probably dismantle the truth and kind of
01:21:06
redirect it in order to do it so I think
01:21:09
that what you're seeing is
01:21:10
they'll initially tell a story but well
01:21:13
then there's too much of the truth
01:21:14
they'll go to the other side because
01:21:15
that's the only way to get clicks and be
01:21:17
seen
01:21:18
so I think that that's what you guys are
01:21:19
a part of right now they are in the
01:21:21
business of protecting the narrative but
01:21:22
I I do think there's a huge ideological
01:21:24
component to the narrative both in the
01:21:26
Elon case where they're trying to claim
01:21:28
that there was a huge rise in hate
01:21:30
speech on Twitter the reason they're
01:21:31
saying that is because they want Twitter
01:21:33
to engage in more censorship that's the
01:21:36
ideological agenda here the agenda is
01:21:39
this radical agenda of decarceration
01:21:41
they actually believe that more and more
01:21:43
people should be led out of prison and
01:21:45
so therefore they have an incentive to
01:21:48
deny the existence of crime in San
01:21:52
Francisco and the rise in crime in San
01:21:53
Francisco if you pull most people in San
01:21:56
Francisco large majorities of San
01:21:57
Francisco believe that crime is on the
01:21:59
rise because they can see it they hear
01:22:00
it and what I would say is look I think
01:22:02
there's a pyramid of activity a pyramid
01:22:06
of criminal or anti-social behavior in
01:22:09
San Francisco that we can all see the
01:22:11
base level is you've got a level of
01:22:13
chaos on the streets where you have
01:22:15
open-air drug markets people doing drugs
01:22:18
sometimes you'll see you know a person
01:22:20
doing something disgusting you know like
01:22:22
people defecating on the streets or even
01:22:24
worse then there's like a level up where
01:22:26
they're chasing after you or you know
01:22:28
harassing you people have experienced
01:22:29
that I've experienced that then there's
01:22:31
a level up where there's petty crime
01:22:33
your car gets broken into or something
01:22:36
like that then there's the level where
01:22:38
you get mugged and then finally the top
01:22:40
of the pyramid is that there's a murder
01:22:42
and it's true that most of the time the
01:22:45
issues don't go all the way to the top
01:22:46
of the pyramid where someone is murdered
01:22:48
okay but that doesn't mean there's not a
01:22:51
vast pyramid underneath that of
01:22:53
basically quality of life issues and I
01:22:56
think this term quality of life was
01:22:58
originally used as some sort of way to
01:23:02
minimize the behavior that was going on
01:23:04
saying that they weren't really crimes
01:23:06
we shouldn't worry about them but if
01:23:08
anything what we've seen in San
01:23:09
Francisco is that when you ignore
01:23:11
quality of life crimes you will actually
01:23:13
see
01:23:14
a huge diminishment in what it's like to
01:23:17
live in these cities like quality of
01:23:19
life is real and that's the issue and I
01:23:21
think what they're trying to do now is
01:23:23
that say that because Baba Lee wasn't
01:23:25
the case that we thought it was that
01:23:27
that whole pyramid doesn't exist that
01:23:30
pyramid exists we can all experience oh
01:23:32
my God I mean and that's the insincerity
01:23:34
of this it is insincere and the
01:23:36
existence of that pyramid that we can
01:23:37
see and hear and feel and experience
01:23:40
every day is why we're willing to make a
01:23:42
bet we called it a bet that the Bob Lee
01:23:45
case was like the Brianna kupfer case
01:23:46
and in that with the disclaimer with a
01:23:49
disclaimer and we always do a disclaimer
01:23:51
here and just to George Hammond from the
01:23:53
financial times who emailed me here's
01:23:55
what he asked me there's a lot of public
01:23:56
attention lately on weather San
01:23:57
Francisco status has one of the top
01:23:59
Business and Technology hubs in the US
01:24:00
is at risk in the aftermath of the
01:24:02
pandemic duh obviously it is I wondered
01:24:05
if you had a moment to chat about that
01:24:06
and whether there's a danger that
01:24:08
negative perceptions about the city will
01:24:10
damage its reputation for Founders and
01:24:11
capital locations in the future so
01:24:13
essentially the enemy says the obviously
01:24:15
a lot of potential for hysteria in this
01:24:17
conversation which I'm Keen to avoid and
01:24:19
it's like have you walked down the
01:24:21
street and I asked him have you walked
01:24:23
down the street in San Francisco Jason
01:24:25
the best response is send him the thing
01:24:27
that Sac sent which is the amount of
01:24:28
available office space in San Francisco
01:24:33
companies are voting with their feet so
01:24:35
it's already if the quality of life
01:24:37
wasn't so poor they'd stay this is the
01:24:39
essence of gaslighting is what they do
01:24:41
is the people who've actually created
01:24:43
the situation in San Francisco with
01:24:44
their policies their policies of
01:24:47
defunding the police making it harder
01:24:49
for the police to do their job
01:24:50
decriminalizing theft under 950 allowing
01:24:53
open-air drug markets the people who
01:24:55
have now created that Matrix of policies
01:24:57
have created the situation what they
01:24:59
then turn around and do is say no the
01:25:00
people are creating the problem are the
01:25:02
ones who are observing this that's all
01:25:04
we're doing is observing and complaining
01:25:06
about it and what they try to do is say
01:25:08
well no you're you're running down San
01:25:10
Francisco we're not the ones creating
01:25:11
the problem we're observing it and just
01:25:12
this week another data point is is that
01:25:15
the mayor's office said that they were
01:25:17
short more than 500 police officers in
01:25:19
San Francisco yeah nobody who who's
01:25:21
going to become a police officer here
01:25:22
are you crazy well and there's another
01:25:24
article just this week about how there's
01:25:27
a lot of speculation rumors are swirling
01:25:29
of an unofficial strike in an informal
01:25:31
strike by police officers who are
01:25:33
normally on the force who are tired of
01:25:36
risking life and limb and then you know
01:25:38
they basically risk getting at a
01:25:40
physical altercation with a homeless
01:25:42
person they bring them in and then
01:25:44
they're just released again so there's a
01:25:46
lot of quiet quitting that's going on in
01:25:47
the job it's like this learned
01:25:49
helplessness because why take a risk and
01:25:51
then the police commission doesn't have
01:25:53
your back it seems like the only time
01:25:55
you have prosecutorial Zeal by a lot of
01:25:57
these prosecutors is when they can go
01:25:58
after a cop not one of these repeat
01:26:01
offenders and you just saw that by the
01:26:02
way in L.A oh look motherboard and New
01:26:04
York Times just emailed and dm'd me and
01:26:06
then and then did you guys say that
01:26:08
instead of solving these issues the
01:26:09
Board of Supervisors was dealing with
01:26:12
a wild parrot what was it
01:26:17
the meeting that was disbanded they had
01:26:19
or yeah they had scheduled a meeting
01:26:21
to vote on whether the wild parrots are
01:26:25
the official animal of the city of San
01:26:27
Francisco
01:26:28
so that was the uh the scheduled meeting
01:26:30
that got uh disbanded also connect may I
01:26:33
just clarify what is talking about with
01:26:35
the Elon interview a BBC reporter
01:26:36
interviewed Elon and said there is much
01:26:39
more race and hate and hate speech in
01:26:43
the feeds on Twitter and he said can you
01:26:45
give me an example and he said well I
01:26:47
don't have an example but people are
01:26:48
saying this he said which people are
01:26:49
saying it and the BBC reporter said well
01:26:51
just different groups of people are
01:26:53
saying it and you know I've certainly
01:26:54
seen you said okay you saw it and for
01:26:56
you he goes no I stopped looking at for
01:26:57
you he said so give me one example of
01:27:00
hate speech that you've seen in your
01:27:02
feed
01:27:02
now we without speaking about any inside
01:27:05
information which I do not have much of
01:27:07
they've been pretty deliberate of
01:27:09
removing hate speech from places like
01:27:10
for you and you know it's a very
01:27:12
complicated issue when you have an open
01:27:13
platform but the the you people may say
01:27:17
a word but it doesn't reach a lot of
01:27:19
people so if you were to say something
01:27:20
really nasty it doesn't take a genius to
01:27:22
block that and not have it reach a bunch
01:27:24
of people this reporter kept insisting
01:27:26
to Elon that this was on the rise with
01:27:28
no factual basis for it that other
01:27:30
people said it and then he said but I
01:27:32
don't look at the feed he said so you're
01:27:33
telling me that there's more hate speech
01:27:35
that you've seen But you just admitted
01:27:37
to me that you haven't looked at the for
01:27:38
you feed in three months and it was just
01:27:40
like this completely weird thing
01:27:44
and this is the thing if you're a
01:27:46
journalist just cut it down the middle
01:27:49
come with prepared with facts listen and
01:27:51
stop taking a position either way I want
01:27:54
to connect one dot please which is that
01:27:56
he filled in his own narrative even
01:28:00
though the data wasn't necessarily there
01:28:02
in the same way that you know we kind of
01:28:05
filled in our narrative about San
01:28:06
Francisco with the Bob Lee
01:28:09
you know Murder being another example we
01:28:12
put a disclaimer on it
01:28:14
we did it hold on a second we so we knew
01:28:17
we didn't know and furthermore we're
01:28:20
taking great pains this week to correct
01:28:22
the record and explain what we now know
01:28:24
yeah
01:28:26
to be intellectually honest
01:28:29
this is just intellectual honesty
01:28:31
honestly you're you're you're going soft
01:28:34
here Freeburg you're getting gas lit by
01:28:35
all these people
01:28:37
I think the guy the guy totally the guy
01:28:40
totally had zero data
01:28:43
put the report on data and evidence so
01:28:45
he certainly you know I think probably
01:28:47
with Don carmignani it's the same story
01:28:50
yeah this is that Don Don happened to
01:28:53
survive guys I love I love you but I
01:28:55
gotta go goodbye here's what Maxwell
01:28:57
from motherboard have fun there's been a
01:29:00
lot of discussion about the future of
01:29:02
San Francisco and the death has quickly
01:29:03
become politicized
01:29:05
has that caused any division or
01:29:07
disagreement from what you've seen or
01:29:09
has that not been the case the Press is
01:29:12
gleeful right now like oh my God you
01:29:16
know it's just like the right was
01:29:17
gleeful with Jesse Smollett having
01:29:19
gotten himself beaten up or you know
01:29:21
setting up his own all right everybody
01:29:23
four the Sultan of science currently
01:29:27
conducting
01:29:29
experiments on a beach to see uh exactly
01:29:32
how burned he can get with his SPF 200
01:29:35
under an umbrella wearing a sun shirt
01:29:37
and pants
01:29:39
Freeburg on the beach wears the same
01:29:41
outfit astronauts wear when they do
01:29:43
space walks hey stable diffusion make me
01:29:46
an image of David Friedberg wearing a
01:29:49
full body bathing suit covered in SPF
01:29:51
200 under three umbrellas on a sunny
01:29:55
Beach
01:29:57
oh my God for the dictator
01:30:01
polyhapatia creating regulations and the
01:30:04
regular oh the regulator you can call me
01:30:06
the regular the regulator see you
01:30:08
tonight when we'll eat our orchalant
01:30:10
what's left of them the final four or
01:30:11
five ortalons in existence otherwise I'm
01:30:15
putting you on the b list today I will
01:30:17
be there I will be there I promise I
01:30:18
promise I promise can't wait to be there
01:30:19
and
01:30:21
the Rain Man himself he didn't even get
01:30:24
to putting Ron
01:30:27
versus
01:30:30
I think you should ask Auto GPT how you
01:30:34
can eat more endangered animals have a
01:30:36
plan for you yes and then have it go
01:30:39
kill those animals
01:30:41
on the dark web to go kill the remaining
01:30:44
rhinos and bring them to chamat's house
01:30:45
for poker night
01:30:48
I don't think rhinos would taste good
01:30:50
wasn't that the plot of a movie but it
01:30:52
was a oh did you guys see is cocaine
01:30:54
bear out yet no it was a Matthew
01:30:55
Broderick Marlon Brando movie right
01:30:57
where they're doing the takeoff on The
01:30:58
Godfather was the Freshman yeah yeah
01:31:00
yeah yeah yeah it's like a conspiracy to
01:31:02
eat uh endangered animals yes the
01:31:06
freshman
01:31:06
the president came out in 1990 yeah that
01:31:10
Marlon Brando did it with um Matthew
01:31:12
Broderick and like Bruno Kirby they
01:31:15
actually they that was the whole thing
01:31:16
there's no Kirby that's a deeper
01:31:18
actually uh
01:31:19
they were eating endangered animals what
01:31:22
do you what do you think heat too is
01:31:23
that going to be good sex I know Heat's
01:31:25
one of your favorite films me too it's
01:31:27
awesome is there a sequel coming they're
01:31:28
gonna do he too and the novel's already
01:31:30
come out Adam Driver so I saw the novel
01:31:32
yeah he's amazing
01:31:34
one of those movies where when it comes
01:31:36
on you just can't stop watching yeah
01:31:39
best bank robbery slash shootout in
01:31:42
movie history you know that is literally
01:31:45
the best film ever like it's up there
01:31:47
with like the Joker with Reservoir Dogs
01:31:49
the the the Joker in that Batman movie
01:31:52
where he robs the bank like I mean what
01:31:54
I love you guys all right love you
01:31:55
besties and four blah blah blah blah
01:31:58
blah this has been the all-in podcast
01:31:59
124. if you want to go to the fan
01:32:00
meetups and hang out with others
01:32:07
we'll let your winners ride
01:32:12
[Music]
01:32:17
and they've just gone crazy
01:32:19
[Music]
01:32:33
foreign
01:32:35
[Music]
01:32:56
[Music]
