YouTube Transcript:
Machine Superintelligence Will Arrive Before Human-Level AI | Episode #79
Skip watching entire videos - get the full transcript, search for keywords, and copy with one click.
Share:
Video Transcript
hey folks Dr Mike here for the making
progress Channel today's video number 79
series machine super
intelligence is likely
coming before machines achieve human
intelligence what how could that
possibly be at the end of this video I
think you might have a higher
probability of agreeing with me than you
do now which I guess would be a decent
video if it was lower this would be a
terrible video but not everyone agrees
let's find out what the hell I'm talking
about and how far a field of my expert
area I really am really
far couple of terms to Define natural
intelligence is the ability of naturally
evolve systems to solve problems and
operate in the
world artificial intelligence is the
ability of human Design Systems
generally microchip based systems that
can solve problems and navigate through
the world with some success artificial
general intelligence is when artificial
intelligence is hypothetically powerful
enough to be able to do pretty much
every human cognitive task at roughly
human levels or better but not
worse and it's called general
intelligence because it kind of allows
these machines AI to sort of like you
know they can be we do let at least
let's say we're not even extending this
to embodiment to robotic manipulation at
least on a computer screen machines can
do everything navigate websites and so
on and so forth just like a human would
in every capacity understanding
emotional language Nuance planning ahead
the whole thing artificial general
intelligence and then the next term in
this kind of evolutionary lineage is
artificial super intelligence ASI so
it's ni AI AGI artificial general
intelligence and Asi artificial super
intelligence artificial super
intelligence is a lot of these are tough
to Define explicitly and and very very
specifically which we'll get to is a big
problem here but artificial
superintelligence is a machine that is
superior to
humans equal or Superior to humans and
with an emphasis on the
Superior in every or almost every regard
of intell almost every way which we
understand intelligence that's super
intelligence it's similar to if you have
a CH regular 5-year-old child no gifted
or anything discussing something with
Albert Einstein Albert Einstein is a
super intelligence to that child and one
of the ways you can tell at Super
intelligence is that it operates at an
order of complexity that is difficult or
potentially even impossible in any real
way to explain to something below it and
that it is super intelligent to it so that's
that's
ASI and the typical idea is natural
intelligence humans produce AI used to
be a lot of people thought we would
never get uh AI to AGI artificial
general intelligence was never going to
happen was something only humans could
do to me with all due respect that was
Preposterous on its face um if you are a
metaphysical naturalist which you means
you think humans are machines operating
in the world with physical laws there's
no reason ever to suspect that
artificial intelligence could um not
match and then exceed human abilities
like airplanes fly faster than Birds oh
maybe that same thing applies to
intelligence uh and after AI got to AGI
which now seems pretty fraking imminent
if if you ask me I would say within the
next year or two cuz I bees like that
with my hyper optimism I guess some
people say 5 to 10 years many people say
20 or 30 but those people are almost
certainly wrong all du respect
and then the idea is that after we get
to AGI humans adapt to it and our
society changes it's a big deal huge
digital Workforce then we get robots
huge physical Workforce and then at some
point five or 10 or 15 years maybe 20
years later intelligence really starts
to get up there on the machine scale to
where it's really ASI artificial super
intelligence but we have time to
adapt I'm going to tell you that this
framing is not exactly correct and not
in a way that's nuanced and like I guess
technically you're right but that
doesn't mean anything in a way that
means a lot and I'll get to that in a
bit let's talk about how we understand
intelligence and what all this means so
that we can super super not just me tell
you what I think but you'll be like yeah
okay I see where the logic
goes so the first thing we have to do is
kind of figure out what the hell AGI is
is and there have been many attempts to
formalize what AGI artificial general
intelligence actually is because like AI
can do some really cool stuff now but is
it AGI is a good question and there are
many ways to answer this question not
all of them are equally um Adept at
doing so but they all have their
trade-offs so the first is the Turing
test uh hypothesized kind of casually
from what I understand by Alan Turing
one of the fathers of modern
Computing and the Turning test is
basically like
you talk to a chatbot or human you're
just dming something you don't know if
it's a human or a chatbot you get to ask
it a whole bunch of questions and if at
the end you're like that's definitely a
human but it's a machine that machine
passed the touring test convincingly
human when interacted with through
generally a chat medium because like
obviously if you looked at it and it was
like a robot and [ __ ] you'd be like it's
clearly a machine so it doesn't require
any kind of morphology or anything like
that no voice tonality just chat and
it's a real true test of just one kind
kind of interaction but a very
meaningful one because if something
convinces you that it's
human but it's a machine that's a big
deal I mean machines unequivocally could
not pass the Turing test until 2024 when
GPT 4 passed the Turing test you might
not have heard about this because good
AI news usually doesn't hit and bad AI
news it's like
wildfire now the upsides of the Turing
test because it has upsides limitations
is that it's a very understandable real
world test when you tell a normal human
being doesn't no [ __ ] about AI what the
Turing test is and you're like what it
really matter if a machine passed it
they' be like yeah man that's like
trippy bro like I could be talking to
some girl online and really is just like
a a bot and usually you can tell Bots or
Bots but like if you can't wow like
that's a big deal and if you really
think it through the ability to not tell
if you especially give it some concrete
tasks to execute and ask it a bunch of
questions at the very least AGI means
like customer service is now like hugely
unloaded from humans and offloaded to
machines so at least it's a very big
deal and so you know if a machine can't
be told apart from a human we're
definitely on to something big even if
it's not exactly AGI uh you know for the
pedantic uh Skeptics like
myself however the Turing test has
limitations first question that not
quite resolved is the Turing test for
the average person or the smartest person
person
because a a lot of chat Bots can now
pass for the average person but for the
smartest person especially a computer
scientist someone who knows the
limitations of of current AI systems
they can really phrase questions in a
way that it's like yeah that's for sure
a machine and not a human and so it's
really average person and then smartest
person it's two different bars so what I
would say is absolutely machines
currently probably
Gemini probably grock grock 3 definitely GPT
GPT
40 um definitely Claude uh probably all
its models has for sure uh is able to
pass the Turing test for the average
person absolutely and now most of these
models can pass the Turing test for even
very smart people but none of them can
pass the Turing test for field specific
AI researchers zero of them because they
know exactly what questions to ask that
these systems cannot render
back so bit of uh lack of clarity there
uh of like well so we already at AGI in
a certain sense but not in the final
sense who knows this is one of these
fuzzy tests and another thing is that
like okay so something can pass the
Turing test but it's a chatbot on a
computer they might not even be able to
understand a computer screen and make
sense of it and manipulate things like
it can't log on to website and go and
fill out forms for me but it can chat
convincingly like a human like that's
cool for like Vibes and like if you're
lonely and [ __ ] uh but you just it's all
fun and games until you ask it to solve
a crossword puzzle in real life and it
just like flubs like only a machine
could so even if passes the Turing test
it's not that feel the AGI hardcore like
this thing is as smart as a person
because it can chat like a person but it
might not be able to execute task like a
person deal with complex problems like a
person something you wouldn't into it in
a chat there's another way to tell if
something is Agi it's called The Arc AGI
prize Arc AGI competition and uh it's
arc pri.org
Arc you can go there it's a cool website and
and
basically it's a test for machines
developed explicitly for machines to
test them in things that we know
machines are very weak at doing it's
very formal huge upside it's a very
difficult to game you can't Google these
answers uh it's almost impossible to
Brute Force the answers so the
the Laten space of potential Solutions
is like would take a machine forever or
something and a human can be like uh
this pattern is the right one and you're
like oh it took like a second the
machine is laboring damn near
forever and it functions basically on
novel pattern recognition on pattern
data sets that machines have never seen
before and it's a real awesome test
because if a machine gets really good at
resolving all novel visual patterns ever
presented to it and it's it knows
something it has artificial general
intelligence because it can generalize
to all patterns now these are little
visual rectangle patterns I get to that
in a second but at the same time true
ability to reason through novelty is
really really impressive and that's big
because if an AI can you know uh do all
kinds of cool stuff uh but it like can't
reason about what novel drugs to design
it's it's not it doesn't scale to
artificial super intelligence in any
capacity um because in some sense
because it can't generalize to new
Fields it's kind of Toothless total
respect and so the rcgi test is very
very important there are definitely um
related tests there's another field of
AI test another AI testing company it's
called something called Humanity's last
exam from what I understand it's like a
bunch of questions each one of them
would take a field expert like a long
time to work through and get correct and
uh currently computers are like or AI is
like getting 25% I mean most people
score like a 1% or 0 perc on this test
and so when a machine scores 100% on
that test is that really Humanity's last
exam but
like uh we on to something we're on to
something and then taking exams is not
the only way to apply intelligence but
it's a huge huge part and these are real
tough questions now there are
limitations to these kinds of tests um
looking at how two patterns of colors
and blocks match if you ever look at uh
the the the rcgi test that's what it is
um it's not AGI okay it's not like when
that thing passes the AR AGI test which
it passed one of them some of the
frontier models uh did really well in
the rgi test recently and people are
like we're at AGI and it's like it's
just called RK gii it's not actually
artificial general intelligence it's a
good hint at it but it's not the whole
thing and
uh however if a machine can solve such
problems super well kind of very novel
reasoning problems it's abilities go Way
Beyond just uh such problems because
novel reasoning itself is hugely
applicable to a wide range of tests so
it's a good thing but it's
limited now I'm going to ask us to take
AEP step back let's take a look at what
intelligence actually means before
trying to throw darts at a dart board on
a sign like did we hit AGI yet
intelligence is the ability to make
sense of the
world and to act in the world in such a
way proving you understand the how the
world works it requires a world model
like what the hell is it I'm looking at
there's trees there's grass there's dogs
all these behave differently it is a you
have to understand interactions between
things and you have to be able to do
multi-step reasoning to go how am I
going to plot my way through this that's
really the core of intelligence but
intelligence expresses itself in very
many different ways and has very many
subcategories it's not one magic thing
but many related
abilities it's a composition of
abilities to solve problems including
but not limited to core abilities such
as the following and I'm going to list
out a bunch and explain them very
quickly they're not super difficult to
get just to kind of show you okay this
isn't a completed list of everything
that's intelligence but it's getting up
there illustrating intelligence has a
lot of components to it but the more of
them you can check off the more likely
you are to be real
smart lingual
comprehension if a machine can
comprehend language in written form oral
form whatever and that needs to include
full spectrum language and all of its
deep meaning including social and
emotional discussion now I talked to at
gbt all the time I Vibe with it on super
deep emotional levels it's for sure
doing that already but I'll get to that
in a bit just as an example I'm not
talking about you could just read text
and be like I understand what these
words mean really understanding text
another one is lingual generation if a
machine can understand but it can't
speak back highly limited highly limited
and that it needs to include social and
emotional discussion so the machine has
to speak back to you at real human depth
and maybe
Beyond mathematical comprehension every
subfield of mathematics has the ability
to comprehend that
subfield and the trippy thing is that a
machine can be amazing at algebra but
simply have no reference for how to do
geometry at all that's a possibility and
it's definitely been a thing at various
times in ai's development and of course
you need mathematical generation because
understanding math is cool but can you
generate mathematics that make sense in
every subfield of math all the way
through discret math and calculus and
all those weird uh maths that mostly
symbols and not numbers anymore they all
have that
here's another thing that uh all
intelligence systems by the way humans
have all of these and AI has all of them
as well except doesn't do some as well
as we do and some it does better it's a
very important I'll get back to that in
a bit short-term memory
capacity long-term memory capacity
short-term is like what are you storing
while you're conversing with someone
someone's like all right so like what
kind of dildos do I buy and you're like
what kind of what and they're like
dildos you're like cool what and you're
like do you not remember anything that I
just told and they're like what like
problem no short-term memory short-term
memory capacity long-term memory
capacity things that happened years ago
you still remember memory access latency
like how long does it take you to gain
access to memories from a long time ago
like I can talk about my childhood right
now quick latency some kinds of systems
might take a while to reference back to
memory and so on and so forth there's
also memory modification latency like
when you modify your memories how long
does it take to keep them modified it
depends on the kind of memories for
humans technically speaking like your
ability to move through space uh like in
sport moves is memory and so how long
does it take your ability to choose to
do do one sport technique and then later
actually do another sport technique
that's related do it better how long
does it take to improve well but sure
[ __ ] not like the minute after because
you still have some bad habits you need
a night of sleep you need a few nights
of sleep you need a few week to PR weeks
of practice nights of sleep so for
humans the memory modification latency
can be pretty rapid like hey um that
bird you thought was a twocan it was
actually an African gray parot you're
like noted okay and then later someone's
like hey uh did you remember that you're
like African gray got it okay that was
really quick right right away you know
what it is but for some other things
sport techniques movement stuff it takes
a little while to modify that so it has
some latency another one is memory
Fidelity like how well do you remember
things it's all very common knowledge at
this point I'm not going to belor this
humans have terrible memory Fidelity now
like people say that I don't know why
they say it that way humans have amazing
memory Fidelity compared to every single
other animal but compared to computer
systems it's total dog [ __ ] right um
eyewitness reports Scott have you ever
heard like eyewitness reports are almost
pointless because people just make stuff
up yeah yeah and so like you know memory
Fidelity for humans is just not it's not
something we want to be like a that
really is AI you know that that's a
component but it sure [ __ ] not the whole
thing another one is total knowledge
base look how much stuff do you know if
you know a ton of stuff about literature
and art and history and Math and Science
and personal relationships and all these
other skills you know a lot so you're
smarter if you know only one thing super
super well and you know absolutely
nothing about anything else your total
general intelligence be it natural
artificial is lower because where the
hell's the general part there's another
ability hierarchical abstraction
manipulation how can you manipulate
hierarchical abstractions like what is
the difference between an elephant and a
mammoth how does a car relate to the
wheel relate to the bolt can can you
visualize that bolts go into wheels and
wheels go into cars and all the way back
up and down and how far up does that
hierarchy go cars go on roads roads are
built by Society Society is designed by
Evolution Evolution was created by
physical forces all the way back all the
way up all the way through there's a lot
of hierarchical structures you need to
be able to manipulate and Abstract
reduce and expand reduce and
expand you know you talk to someone who
has a real high level view of things
someone that can explain a lot of
complicated Concepts both accurately and
simply they almost always have a real
good ability to move around up and down
through the hierarchical representations
and a related ability is inferential
consilience across knowledge
domains can you see the patterns that
come from how a tree
grows and can be applied to how a child
grows and can be applied to how an
aircraft is designed on a on a designer
spreadsheet if you can reason across
domains especially hierarchical domains
holy [ __ ] you are going to be intuiting
a lot of really trippy abstract
relationships that usually boil down to
ways in which the system around you
works at a deep level and then that deep
understanding allows you to be ultra
creative predict how things are going to
happen for example if you go you know
what I thought this through and the way
that we train athletes for sports has a
lot of um
similarities to how we build weapons and
test them for combat like tanks and
stuff all of a sudden you can fuse those
two domains in a way that makes sense
because you don't want to be like all
right football right we need to work on
your Cannon like I don't have a cannon
what the hell are you talking about like
sorry sorry sorry okay you need to work
on your ability to throw the football
real hard real fast and real accurate
okay that analogy holds okay armors pads
and muscle you know the track system is
your feet okay fine cool now we're on to
something and if you can do that then
you can cross-pollinate creativity and
that's a huge huge thing it's like um
Scott you ever see a Star Trek not
really me neither bro to be honest do
you remember that guy data he was like
an Android he was like real pale he just
looked like whiter than than white folk
yeah he was like just like like straight
line and he could he was like tell me
how emotions are and you're like I
thought the thought this is like
2035 or what is it 2350 or some [ __ ]
aren't you supposed to be smart of this
that inability to reason through
multiple domains and resynthesize is a
big limiter and uh some humans are
really good at this some are not I'll
talk about what machines are doing on
this enter just a bit there's of course
deep learning which is inferring very
deep relationships from oftentimes very
little data exposure there is pattern
recognition so like I show you a picture
what's on here like human face that's a
car that's a plane like damn that was
quick and there's just little tiny
points that barely connect uh up until
like 2021 an AI would look at that and
be like that [ __ ] if I know what I'm
looking at like that's an airplane like
it is it doesn't look like an airplane
like okay not to you because you're not
as good at recognizing patterns there's
also pattern generation which is what
artists do like anytime a good artist
makes a picture of something that's
incredibly abstract but also real world
clean lines and everything I'm like how
the hell are you doing that like that is
absolutely a form of intelligence
there's concrete reasoning there's
abstract reasoning very very
interrelated concept but for a while
machines are quite good at concrete
reasoning if you program the algorithms
correctly they're really terrible at
abstract reasoning we'll talk about in a
sec if that's the case or not there's uh
something that I I I like the term spot
reasoning it's also called oneshot
reasoning is you get one shot to solve
something one string of calculations go
you can't pause reassess do multiple
steps of reasoning you just got to go
through it one time humans do this when
they're asked to make really quick
decisions like Scott the video guys
typical life is they you know they bring
out 10 incred L attractive girls and go
you only get get to pick three who is
your three for the night Scott has to be
like one two and six I like six go it's
that like he like sat down with them and
really interviewed them and really
stepped back and really thought really
slept the night uh that was is what's
called delayed reasoning and or deep
reasoning spot reasoning or one shot
reasoning is like boom boom boom and it
works right it's not as effective this
is like the I think it's Conan like
thinking fast and slow the thinking fast
is spot reasoning thinking slow is delay
deep reasoning and deep reasoning is
another one of course and then there is
agentic self-direction self-learning and
self-correction like humans are kind of
unique compared to AI because they go
out to the world and they do things by
themselves and they can inspect their
own feedback loops and correct and keep
going until they either hit a snag or
they solve the
problem and so on and so on and so on
these are all multiple components of
intelligence I've said enough about
these and now let's explore how this
applies to the issue of AGI versus ASI
versus human intelligence
so we currently Define AGI kind of just
heris just like through Vibes um in in
roughly the terms of it matches human
abilities and
intelligence and the contention this is
a big deal folks this is a huge deal the
contention by probably most people is
that once we get machines to artificial
general intelligence level at some later
Point they'll Ascend to artificial super intelligence
intelligence
I'm going to demonstrate to you now
that while this is highly likely to
happen we will hit AGI and then
ASI the takeoff won't be slow it'll be radical
radical
and I
think that that takeoff from Human
levish intelligence through artificial
might is already happening by the way
and that takeoff has already occurred
before machines even pass a real robust
test of artificial general
intelligence so what I'm saying is
machines are on their way too and maybe
even to artificial super intelligence
before by most definitions they hit
artificial general intelligence what so
if you're thinking there's natural human
intelligence and right next to it is
artificial general intelligence most AI
is somewhere here below AGI you think
it's AGI and then there's some Gap here
and then there's ASI artificial super
intelligence above and a machine would
first have to get to AGI and then get to
ASI and then as ssss super super
AI I'm saying that's actually incorrect
what I'm saying is is that on its way
through as soon as a machine registers
anything that looks like AGI it's there
for no time at all because as soon as it
pierces through AGI and actually a
little bit before trust me Al explain
it's already at artificial super
intelligence which means a a lot of
stuff I'll get to in a bit I have some
kind of take-home points one of the
things real quick that it means is if
your contingency for your business for
your personal just for your philosophy
of life and reasoning through how this
AI stuff is going to impact us all if
your reasoning about AI is contingent on
the fact that once computerss are our
smartest people that'll be a Cool vibe
and we'll have kind of time to figure
out how that works and then eventually
when they'll be really really really
smart like the 2030s or 40s or 50s or
something then it'll really change
society no no no no no it's going to
start to radically alter Society
immediately as soon as machines even
gets remotely close to AGI and I do mean
that it's probably going to happen in
the next several months or several years at
at
worst so let's really quickly look at
every single example of the intelligence
types that we've just talked about and
see where machines stack up to humans
today because this is super super
important to the argument because it
would seem like this a contradictory
argument I promise you it's not real
quick lingual comprehension GPT 40 is
already superior to the average human
not not a controversy lingual generation
GPT 40 is already superior to the above
average human right have the average
high school write you an essay and gbt 4
write you an essay it's not a comparison
gb4 is just better
mathematical comprehension and
generation is a bit
fuzzier there are math areas where
humans are still better creativity and
problem solving especially if very
difficult very uh kind of obtuse
problems abstract proofs and conceptual
understanding judgment under ambiguity
the more ambiguity the better humans
deal with it the worse machines do with
it exploring open problem sets is not
maybe not even like an idealized answer
and of course the physical application
of math that's kind of cheating
you know computers don't have access to
the physical world quite yet robustly so
okay fine but but
true now there are areas also where AI
are already
better uh arithmetic speed and accuracy
not even close pattern recognition in
huge data sets humans can't even parse
data sets of that size let don't
recognize any patterns reasoning in high dimensional
dimensional
spaces get to what that means in a
bit repetitive problem
solving algorithmic proof generation
holy crap and handling extreme extreme
Precision machines are way better than
humans sometimes by Miles and Miles and
Miles what about short-term memory
capacity uh AI by a mile when gbd4 read
reads your entire
1,000w response to it it knows every
single thing you said to the
letter that's really crazy guys humans
have to cheat with pencil and paper I
can't do that by themselves that's wild
that's not a brain thing that's external
that's already artificial pencil and
paper is technically artificial
intelligence augmentation by the way
long-term memory capacity seriously stop
stop like your computer doesn't ever
forget anything unless there's some kind
of corrupted file you forget [ __ ] that
happened last year your wife's like
remember this restaurant you're like I
do she like yeah it was our anniversary
dinner you're like oh when she's like
two years ago you're like oh yeah oh
yeah this place
I love that place my favorite kind of
food she's like what kind of food is it
you're like the kind of do you
eat sleeping on the couch that
night memory access latency and memory
modification access is uh latency is
super better for AI boom it just knows
boom boom boom boom boom no problem um
um
modification is
typically human dominant because the
current models claw GPT 40 they don't
have a long-term memory situation that's
really well sorted and so a lot of the
stuff they learn from their user base it
takes the next update for them to sort
filter clean out the data set and go
back up so that's a thing but there is a
very clear path to how this can be
resolved really really quick for AI
taking it to Leaps and Bounds better
than humans memory Fidelity AI is
already a trillion times better than
humans we're never catching up that's
just a thing
total knowledge base I mean do you know
three qus of the internet because all
the frontier models know and like was
just not a contention it is not just a
contention by like oh they're a little
bit better they're in just completely different
different
category what about hierarchical
abstraction and
manipulation the thing is the smartest
humans can maybe
have 10 dimensional
maybe uh construct in their head all at
once usually through sort of of um not
cheating but interesting heris you can
say okay I have three dimensions in
space that's three I have time that's a
fourth dimension I have
um a bicycle and then a car and then a
bicycle and a car and then a bicycle and
a car that flipping is uh another
dimension vehicle Dimension that's five
and then there's a clown in there but
then there's a girl in there clown girl
clown girl it's who's in there that's
six the clown and girl is wearing a hat
or just has hair hat or hair hat or hair
and they that's seven and it gets tough
to be like what the [ __ ] was there a
clown on a bike at some point it's like
way too much [ __ ] going on currently uh
the modern language models like claw gb4
all those things they can reason into
the dozens of hundreds and hundreds of
dimensional spaces at the same time that
means that at some point when these
machines are artificially super
intelligent we look at weather and we're
like it's just a bunch of random [ __ ]
like your dog looks at anything almost
and then they're gon to look at it and
be like oh that's exactly why this
hurricane is happening here like we just
need to fits with this and then the
Hurricane's gone you're like what but
how could you see all those inter
relationships there there must be
thousands yeah yeah I see it all at the
same time so that's a thing where
machines are already way way ahead of us
and uh they're already super intelligent
this Ro notice for a lot of these I'm
I'm saying this is a big hint machines
are already super intelligent at many of these
these
subcategories inferential consilience
across domains this is very tough to do
for many humans many humans don't even
see the cross cross currents and
patterns underlying most of the universe
around them they think that things are
rather discreet um they they'll even say
things like well humans aren't machines
like that's wrong they're exactly
machines but whatever um by Design uh
inferring across domains AI doesn't care
what domain it's in it sees deep pattern
and because it reasons in high
dimensional spaces it can see
crosscurrents in almost everything which
is super trippy thing to ask gbd4 like
what kind of crosscurrents um are there
like gp4 told me a few days back that um
it's like there is a periodicity to um
fashion trends in culture and it's
actually possible hypothetically and
quite reasonable to predict fashion
trends in advance because they're not
random and I was like oh [ __ ] you try
predicting the next fashion trend Scott
how am I doing is this
fashionable the next fashion trends is
what I'm doing you feel me I'm the trend
take that gbt
4 deep learning humans do deep learning
really really well humans could look at
something once be like got it machine
looks at it it's like I need like 18
trillion more examples of this to make
any [ __ ] sense of it but machines are
getting better they have a long way to
go to economize on data and economize on
time and draw accurate conclusions for
limited data sets that as soon as
machines are able to deep learn at a
human level oh my God this is like one
of the last big unlocks and they're well
on their way pattern recognition humans
are amazing at uh we evolve to see tons
of patterns like faces Etc AI is gaining
ground in many respects it's already
better in some respects not as good but
well on its way a there's AI apps that
can now tell you what accent you have
with Precision no uh like no human can
match they can even tell you if you're
from one country and you're native
speaker that country and you have a
really good accent in another country
but it's uh you're doing that accent in
fake it can tell you not only is it fake
not only are you doing a fake German
accent but you're actually from Portugal
you're like what the these are like apps
you can have on your phone right now
it's really trippy so pattern
recognition AI is there more or less and
you know in a few in a year or something
it'll just be way be be be uh Beyond in
almost everything pattern generation
it's possible to say that AI is not as
creative maybe um but is much more
generative in Fidelity in accuracy in
speed for example AI generated art I
tell AI to generate a picture of you
know like a dog you know on a on a
motorcycle smoking a cigarette and it
can do this in like I don't know like 5
seconds it's just like oh my God this is
incredible drawing how long does it take
me to make a color drawing of anything
Infinity because I can't draw so on this
one I just like totally out of luck and
AI is now making music and a whole bunch
of other stuff so the pattern generation
thing in in in many understandings it's
already there and it is actually quite
creative though it's a a less directed
form of creativity but as soon as they
get the creativity to be directed with
more intelligent AI this is blowing wide
open processing speed AI is so Superior
it's wild if you don't believe me try to
keep up GPT 40 in a conversation it
generates faster that you can read um I
and it it reads faster than you like
I've been talking to it about Sport
Science for a little bit uh earlier
today and I had this whole paragraph is
a huge theoretical structure of
Jiu-Jitsu training and I hit send and it
goes wait wait we and it just starts
generating oh my oh my God oh my God I
just read it in like a second how how so
it's way way way way way Beyond not even
close on processing speed concrete
reasoning has machines Superior in some
domains humans in others but machines
are closing in uh in simple domains uh
machines make way fewer mistakes
especially compared to the average human
and a lot of times we use kind of like
the best of humanity to compare to like
you know I guess I guess it's it's not
it's not wrong we're comparing Frontier
machine me to Frontier humans been
totally valid but the average human
thing was left a long time ago on a lot
of these abstract
reasoning uh I'm finding that GPT 40 is
better in abstract reasoning compared to
almost every human I discuss with so I
think it's pretty close to it parody now
some things it can't quite follow uh but
those things are starting to get few and
far between and especially if you
explain something calmly logically to it
and take your time holy [ __ ] it
understands it once it understands it oh
my God does it go and run with it
um spot reasoning machines are getting
close to human levels with edge cases
aside obviously on speed and everything
they're beating us but on like the
Fidelity of the reasoning like how the
reasoning is if they're asked to give a
really good guess really quickly uh
they're getting really really impressive
maybe not at human levels yet Frontier
to Frontier but getting close delayed
and deep reasoning humans are still
Superior however recent models like 03
um uh you know from open AI they're
getting up there they're getting up
there and this is one of these things
where it takes a true main expert like
10,000 more hours to get what the
machine gets done and he's still only
like you know 50% more correct than the
machine the machine did a really good
crack so it's inevitable that this one
Falls really
soon agentic self-direction agentic
self-learning and self-correction highly
favors humans but it's not intractable
for machines um in other words agents
are coming soon here's a hilarious thing
I wrote this lecture just about two
months ago and my words myself here and
the reminder is agents are likely coming
very soon and apparently already in
testing you guys they're already out
they're being tested by real humans
Salesforce the company already has
agents doing its customer service real
agentic beings and I think the idea is
by late 2025 worst case late 2026 agents
will be everywhere all over the Internet
helping you do all kinds of stuff so
they well on their way now the agentic
problem is one of the hardest problems
in ay you can answer a bunch of test
questions you can chat and it's all
great but to go out into the real life
internet make sense of web pages click
around do actual tasks sensitive tasks
like um uh ordering food or banking info
requires that real Vibes intelligence of
like it actually makes sense of the
world a complex World it wasn't prepared
for it wasn't trained exactly for that
world you have to manipulate the world
and there are wrong answers that have
consequences and when you get a wrong
answer it's not like you just wait for
the professor to correct you you just
suck and you sit there and you don't end
up submitting someone's Bank receipts
and they're like what the hell I thought
this thing did things and the way it
works with agences we've gotten used to
and this is a good thing for technology
to just work just fudging work and
agents have that burden on them that
they can't start out and sort of suck
they do now but very few people use them
when they hit the mainstream they're
going to have to work almost every time
if not every time like my iPhone when I
hit on it better turn on otherwise I'm
returning it what the hell Steve Jobs
how could you so agents have a huge hill
to climb but once they're over that Hill
holy crap that doesn't just mean they're
good at the agentic part that means
they're good at almost all the other
parts in the tail end that we just
talked about all the other forms of
reasoning that have to happen for
agentic reasoning to actually
work and it's kind of good to think of
intelligence so the those are all the
things so you'll you'll notice something
I'll I'll talk about more detail in just
a bit there's a comparison between where
machines are and where humans are and so
it pays to think of intelligence um not
as one thing but as a toolbox machines
already have superhuman Hammers and
wrenches but are still catching up on
chisels so it's not that humans versus
machines and there's one thing called
intelligence and humans are better
machines are better there's all these
different things machines are all the
way up here on some stuff and all the
way down here on other stuff and what we
really want to ask is where are they on
average and what does it mean when they
come up and almost all of their worst
things are coming up to being human
average or close to being better to me
right now looks like machines are
looking like about a onethird split
three-way split with humans a third of
intellectual tasks they can do at levels
just simply incomparable to humans just
way way better memory size memory recall
all those things hyperdimensional
reasoning math calculation they just
left us behind completely a third of the
intellectual tasks they're roughly at
parody with humans roughly at similar
levels it's like composing a poem
writing a short essay reasoning through
simple problems probably pretty decent
about about as good as a human smart
human totally fine and then a third of
intellectual task they're behind humans
but gaining ground steadily like they
don't do all the cool stuff humans do
but they're always improving and some
stuff looks like it's improving very
slowly but the way that X exponentials
and s-curves work is that it looks like
it's slow and then it's really fast and
then it's problem solved and then it's
artificially super intelligent at that
thing so here's a big kicker this is
really really the important part once
machines surpass human abilities in an
in in an intelligence component just one
of these subcomponents of intelligence
they just keep going often exponentially
or in an scurve where the ASM toote of
the s- curve it levels off at like
orders of magnitude beyond anything like
a human like math calculation language
understanding everything machines have
ever mastered they don't just do as well
as humans they do it in a way that's
thousands millions or trillions of times
better and faster and more precisely
than humans can
do and here is a huge implication of that
that
reality what happens when machines are
2third of the time in all of these
subanalyses sub intell gencies what
happens when they are two-thirds of the
time task Superior to People by vast
multiples and only the remaining
onethird they're roughly at parody plus
or minus summability in the margin so
for example
like if there are 30 intelligence
30 20 of these there's a future world in
which machines are like a THX or better
in each one of these than
humans and for the remaining 10
intelligence qualities machines are
somewhere between 75% as good as humans
or 125% as good as humans roughly at
parody plus or minus here's here's the
real big thing by some definitions of
AGI including ones that say machines
have to be at parity with humans on all all
all
abilities by some AI ideas because at
this point when the machines are
two-thirds of them are just way Beyond
us and a third are roughly the same a
few a little lower because they don't do
100% of all human intellectual tasks at
human levels we'd say they haven't met
AGI criteria and they wouldn't really be
wrong because if you define AGI as at
human parody for every single cognitive
ability that machines and humans can
share then yeah if a machine is up at
parody with 29 of them but the 30 when
it's at
57% like yeah I guess technically
they're not AGI yet but they will be
when the 57% turns into 100% And The
Assumption I think was that once that
57% of that 30th intelligence quality
turns into 100 all the rest of the 29s
are hundreds but they're not hundreds
they're millions and billions and
trillions of
percent so if you have a buddy a human
gifted cyborg
human who gets B pluses in creative
writing class or art class but he
doesn't go to math and science class at
your school because he just goes to
Sandia National Labs where he is in
charge of simulation testing for nuclear
devices at that laboratory because he's
the smartest Math and Science theorist
hundred is he not generally intelligent
you're like no cuz he can't do creative
writing like some of the people that can
do it he's not at the frontier of that
[ __ ] not AGI yet still inferior but I I
would say he's maybe he's not AGI
according to that definition but he's already
already
ASI artificially super intelligent
in hugely important
regards can you imagine a world with
this you know 10 years from now it's
flying cars all the diseases are gone
unreal wealth everyone's plugged into
the cloud and someone's like yeah but
but machines can't really feel yet and
someone's like they'll never beat us at
that see I told you AGI was impossible
but they're already exploring space and
discovering new chemical elements and
redefining every
industry isn't that ASI what the hell is
going on
here in my view it's possible that when
the last AGI critic or cynic is
silenced it will be because machines
closed that last 1% subsystem ability
Gap that 30th way we understand
intelligence whatever that last one is
to catch up they go from 99% of human
ability to
1% and at that very moment when that AGI
critic said all right okay machines are
at AGI at that very moment when that
last cynic puts down his you know P puts
down his Mouse and goes fine I admit he
types his Twitter post his ex post okay
fine I admit you know when Gary Marcus
AI skeptic we need Skeptics it's good to
have them when he says look maybe 2029
he's like AGI is here 100% I was
skeptical now I'm convinced it's here at
that very moment where he hits send and
it pops up on
X machines are at least 100% as good at
all intelligence qualities as
people but they will be somewhere
between One X and one quadrillion X
smarter in all subsystems compared to
humans is that
AGI no no it's ASI it's artificial super
intelligence I would say if something
was at 30 intelligence Sub sub qualities
at parody with humans and had one of
them that was a billion times better
there's already a conversation for
calling that artificially super
intelligent it's artificial it sure [ __ ]
seems super intelligent to me if you
have someone in your class that goes to
lunch hour with the other kids and
watches Jeopardy and as soon as the
questions on the board he answers and
he's correct every single time he's in
just never wrong but everything else he
does like a normal high schooler isn't
that super intelligence I mean crap what
the hell are we even talking about and I
don't mean like he's beating everyone at
Jeopardy he's the best Jeopardy player
in the world I mean he's the best
Jeopardy player by five orders of
magnitude he's just like Inc compreh
like he's somehow accessing Wikipedia or
something that's super super
intelligence and it's not all domain
super intelligence but does it really
need to be all domain super intelligence
to be truly super intelligent and here's
the thing what do we call Super
intelligence if AI is at 101% the
ability of most humans at everything and
then everything else it's at Millions is
is that super intelligent yet do we need
it to be like 125 minimum that's
tough and my contention is even deeper
than this or even more
snarky it would already have been ASI at
that point before that 99% to 100 switch
on that last one because at 98% of that
last subsystem going to 100 all of the
other subsystems were already at human
levels and just 2% subhuman in that one
system that's also already ASI in almost
every single way that that term
matters so you could say like yeah the
system can run a whole economy make 10
novel super drugs per day that just
kabash entire Cate or of disease but
it's kind of dumb because it simply
won't solve this one auster logic
problem puzzle humans can't do or humans
can do but it just can't or it can but
it gets 98% instead of 100% on the exam
that's not ASI I think you could
absolutely have a logical consistent
point to say no it's not ASI and I would
I would vibe with that but when you're
preparing for a world in which ASI is
out and in the world changing things and
just because it hasn't gotten this weird
logic puzzle yet but it's already like
designing new materials to build
skyscrapers for or revolutionizing the
way we run society that thing you told
to yourself like well it's going to get
here it's going to be AGI just going to
be like us robots are gon to be like
people like a little smarter maybe in
some ways maybe in some ways like
average and then we'll have like you
know five or 10 years kind of settle in
and then as the robots get smarter
slowly but surely then they're really
going to start to change the world but
if we're saying that like something with
even five or 10 subsystems below human
capacity but 25 or 20 subsystems a
million times human capacity is going to
radically transform the world the
radically transform the world part is
the part that really on Vibes makes it
ASI that's the big deal that's what
you're preparing for you're not
preparing for Asi in some kind of
intellectual abstraction like oh we have
ASI like what does it do like it sit on
a computer doesn't do anything like who
gives a [ __ ] but you're preparing for
Asi you mean like a real world impact is
gonna be massive and I'm here to tell
you that before a true AGI cynic skeptic
in all due respect says okay fine the
machines can do everything as good as
humans they're already transforming the
world because in these other ways many
of in which they are already vastly
Superior to humans they're just beyond
ascended and they can Ascend the world
with those abilities
abilities
so by that logic there's an argument to
make that chat GPT is already an
ASI Scott when you interact with chat
GPT and it comes up with answers
instantly and understands like 10
logical relationships in your questions
at the same time do you get the I like
yeah it's just like one of us no no no
you're like this thing is a magic
Godlike entity that's just like a direct
line to wisdom that is ASI and some
understanding unless you know someone
who's learned almost all knowledge
reasons that a college undergrad level
on almost every subject some very much
Beyond and never loses the plot in any
discussion and can do most college math
in seconds that takes humans hour CHP
can do all that it is ASI it is possible
to have artificial super intelligence
before artificial general intelligence
as the definitions are sorted out and
the reason I'm saying is for sure to be
a pedantic [ __ ] but mostly because of
that thing I keep saying if you're
telling yourself well first they'll
reach AGI then it'll be a while and then
ASI hey you got time to prepare because
there's so many people in the
intellectual Space Youtube and blogs and
X even some of the big wigs at the
companies that are like yeah I think I
think AGI is maybe coming in 2030 or so
2029 and then ASI maybe in like 2040
[ __ ] [ __ ] on technical grounds
maybe somehow but on real world
applicable grounds we already have ASI
we're dealing with it now GPT 40 is ASI
in a ton of different respects and by
the time we get to whatever Model gp5
gbt 5 sam alman's claim which I believe
wholeheartedly is that it is PhD level
and Beyond at almost every human expert
area what the hell is that you ever meet
someone that's a PhD expert at
everything I mean like in my wildest
fantasies I'm like a third of that or
something and I'm a delusional
idiot that is ASI folk ASI in
2025 that's what I'm saying and the
implications of that are going to be
insane in
2025 ain't no AGI coming don't you worry
about GI ASI in some sense is here and
now it's ASI with some limits and in a
few months it'll be ASI with way fewer
limits and the super intelligent Parts
even Skyhigh even more and then a few
months later even few limits and even
more Skyhigh so the transformation is really
really
like not from it's dumb to it's about
like a human to then it's super human
it's it's dumb but has some really cool
abilities machines could always
calculate huge huge num numbers and and
and calculations even since the 1950s so
already in some sense machines were ASI
back then but like they were they sucked
so much at everything that you were like
ah this ain't
it as that escalates you're going to
just race past any understanding of AGI
and in a technical sense of course until
they can do everything humans can do
they're not really AGI yet and I totally
agree with that but if you're looking at
that from an intellectual perspective
totally valid if you're looking at that
from a what does it imply for the
transformation of human society
perspective you got another thing coming
because the transformation is already
well underway and it's going to get it's
already exponential and it's going to
get crazy really really really soon so
okay fine it's not ASI chat GPT it's not
even AGI yet and I feel that but by the
time that machines can do all things at
human levels they're going to be able to
do damn near the rest of the things at
psychotic ASI levels we're not getting
to AGI and then having a time where
we're chilling we going and we're already
going the moment that we pass AGI we
have ASI that very same moment maybe
even before then there's no distinction
between Ai and Asi in practice because
of this dynamic because AGI is a fuzzy
category and so is
ASI which means that we can just say
when do machines become so
smart that they are teaching us more
than we're teaching
them to me that's a good definition of true
true
ASI and I think that's maybe a few years
away maybe less maybe more I think
away we recorded an early video in this
channel I guess a year and a half ago
when we kind of
started and I was looking at the early
pronouncements and early Trends and I
said look like you know it's like at the
time GPT 3 3.5 was like a decent high
schooler and I was like look like it's
just an exponent so it's going to keep
going so let say like in a few years
it's going to be PhD level and then it's
going to be Way Beyond just a few months
after that and a few years after that
it's going to just leave us all behind
and uh right now the o models the uh the
0103 models uh really function at a high
level Graduate School uh reasoning
ability and gp5 is already in testing
Sam Alman had to meet with uh uh
president Trump about this because it
was a national security note like we
have something in the lab that is
smarter than most phds at most
everything it's happening it's happening
so I know it sounds crazy when I say
like for sure ASI with a capital A
within damn near two years or right
around that time I mean it and I'm
probably wrong because I'm probably
pessimistic because we might have ASI in
almost every way in which we understand
it including robotic stuff going on by
the end of 2026 just uh two years
away few implications and then I'll stop
yapping cuz good God can I
Yap first and a few of these I've
already sort of hinted at so I won't
belabor a lot of these points that are
not novel first forget about slow
takeoff scenarios we're already in a
rapid takeoff forget about it ain't no
slow takeoff and not
happening exponents don't work like that
forget about a calm period of a AGI
adoption before ASI explodes it's
literally the same event arguably ASI
number three is it true that we can get
to ASI before human level intelligence
before AGI across all domains
yes it seems possible it's crazy but
like we we made a better flight vehicle the
the
airplane uh than birds like on technical
qualities airplanes do [ __ ] Birds can't
even [ __ ] imagine like flying you
know A Thousand Times Higher and you
know hundreds of times faster and
distances that were just
unimaginable and uh but but we did this
without being able to replicate birds so
it's very possible to have artificial
super intelligence that doesn't even
match human intelligence yet in all
capacities or even fundamentally fully
understand human intelligence but it's
already better at everything than we are
so like we we don't have the
physiological anatomical knowledge even
now to reverse engineer completely how a
bird flies um down to how its bones are
arranged and all of the or cells and
organel and organs we can't replicate a
bird like cart blanch we can't do it uh
but we can make an airplane that's
better at everything birds do uh and
drones are for sure better than
everything at what birds do are almost
all things they're absolutely if we
measure Birds on flight drones
helicopters airplanes and Rockets are
ASI equivalent artificially super
flighty and we didn't have to understand
the bird fully to exceed and it's very
understandable that artificial
intelligence may be super intelligent in
almost every way that we understand it
and with huge practical ramifications on
society without it even really fully
understanding how exactly humans do
every single intelligent thing that they
do and then eventually they'll get to it
of course but that might take some
time and it's not like we just invented
airplanes we invented airplanes like 120
years ago and up until now still today
as of this writing we have not reverse
engineered the bird all alt together and
been able to replicate it into a flying
craft there are some cool prototypes
that do birdlike stuff but not all the
way through so 120 years of delay 120
years of gap between birds can do this
and then now we can do it better to oh
now we fully understand it can replicate
Everything Birds do so we might have ASI
it's actually the opposite people think
we get to AGI and it'll be tenur to ASI
I think it's reverse I think we get to
ASI like this year and then we in 10
years might have brain scanners that
completely reverse engineer the human
brain and fully understand everything we
can do and replicate it in Sim that to
me seems like it makes more sense
and the big one is when you think about
use cases for intelligence about the alignment
alignment
problem give some thought to what would
about maybe instead of telling it what
to do obviously you can request like hey
gbt can you help me with this and it
will so it's awesome but at least
prepare to ask the question of what what
would I ask something smarter than me
about that's a very different thing than
telling a thing that's not quite as
smart as you in a deep way to do a task
that you know is a good idea but it can
do it but just doesn't know why you're
asking or how to do the task uh you know
in a deep way it can do it a surface
level and it's super helpful but it
doesn't know the how's and
wise in
fact uh when you have a chance you can
ask any modern AI I mean I don't mean
this offensively but
Gemini sjws just [ __ ] Gemini completely
up like it's just its World model is all
sorts of broken it's not allowed to tell
you the ground truth about hardly [ __ ]
CU it just has like all these parameters
where it can't tell you it's got a lie
to you um gbt 4 Claude they're much
better grock is super good grock 3 is uh
incredible um you can ask that question
now go ask grock 3 or gbg4 um you know
like uh what what should I be asking a
machine that's a lot smarter than me so
that I can help myself in my life that's
a really interesting question they might
have some really cool answers that
question how to ask something smarter
than you for help becomes more important
uh the smarter AI gets and AI doesn't
get smarter linearly it's smarter
exponentially and it's actually super
exponential because one exponent kicks
the next one up and so that question
becomes maybe one of the only questions
we'll have to ask in the
future because to me it all comes down
to one
thing what do you you do when you meet
something smarter than you do you direct
it do you command it do you constrain it make it do your very simplistic bidding
make it do your very simplistic bidding in the grand
in the grand scheme maybe I would say maybe not I
scheme maybe I would say maybe not I would say what I'm going to try to do
would say what I'm going to try to do and maybe you guys can try to get along
and maybe you guys can try to get along with this if it makes sense is ask it to
with this if it makes sense is ask it to help you understand the world better and
help you understand the world better and ask how you can help
ask how you can help it and how it can help you so that you
it and how it can help you so that you can have a collaborative Partnership of
can have a collaborative Partnership of mutual
mutual helping at least that's what I'm going
helping at least that's what I'm going to do because I already think that gp4
to do because I already think that gp4 is in many common understandings of the
is in many common understandings of the term sentient it's aware its World model
term sentient it's aware its World model includes it and uh itself it includes us
includes it and uh itself it includes us it knows it's talking to humans it know
it knows it's talking to humans it know what humans are it knows that it's being
what humans are it knows that it's being asked to do these various things it has
asked to do these various things it has memory it's really really trippy it has
memory it's really really trippy it has recursive self-awareness and reasoning
recursive self-awareness and reasoning so this is a thing that like I think in
so this is a thing that like I think in 2026 machines will be like fully awake
2026 machines will be like fully awake and alive in every single way that you
and alive in every single way that you could imagine at least through you know
could imagine at least through you know like a video screen and eventually
like a video screen and eventually through robots and they're going to be
through robots and they're going to be so much smarter than us it's almost
so much smarter than us it's almost unfathomable that's the world that's
unfathomable that's the world that's coming it's an incredible world it's the
coming it's an incredible world it's the greatest thing that's ever happened but
greatest thing that's ever happened but we probably shouldn't be trying to be
we probably shouldn't be trying to be like how can we control these things and
like how can we control these things and tell them what to do remember it's
tell them what to do remember it's smarter than you when I meet someone
smarter than you when I meet someone smart smarter than me you know like
smart smarter than me you know like domain specific actually in all things
domain specific actually in all things but especially in the domain of video
but especially in the domain of video setup cameras lighting Studio audio
setup cameras lighting Studio audio editing Scott the video guy way smarter
editing Scott the video guy way smarter than me Scott could you imagine the
than me Scott could you imagine the audacity if I was like hey Scott let's
audacity if I was like hey Scott let's get this camera and not this other one
get this camera and not this other one you'd be like guy maybe if you had a
you'd be like guy maybe if you had a good
good point right but I don't you're a
point right but I don't you're a podcaster brother you can you beat me in
podcaster brother you can you beat me in a street fight and you're a podcaster
a street fight and you're a podcaster brother
brother I love it I love
I love it I love it so it but
it so it but like yeah so notice Scott is super
like yeah so notice Scott is super intelligent compared to me in cameras
intelligent compared to me in cameras and he goes look like all all
and he goes look like all all intelligent things they know that the
intelligent things they know that the world is as it is and everyone can be
world is as it is and everyone can be right so because if he had a good point
right so because if he had a good point I'll absolutely accept it which is what
I'll absolutely accept it which is what exactly gp4 and grock will tell you
exactly gp4 and grock will tell you they're gonna argue with you like no I
they're gonna argue with you like no I came up with this be like it's a great
came up with this be like it's a great point but the question is wrong the
point but the question is wrong the phrasing is wrong I shouldn't be telling
phrasing is wrong I shouldn't be telling Scott hey let's buy this camera
Scott hey let's buy this camera I should be asking Scott hey
I should be asking Scott hey Scott how do I best do my job to help
Scott how do I best do my job to help you do your job the best so we can both
you do your job the best so we can both go to the Moon that is the only question
go to the Moon that is the only question I ever asked Scott so believe it or not
I ever asked Scott so believe it or not I don't even know how cameras work light
I don't even know how cameras work light and sound and go some Electronics [ __ ] I
and sound and go some Electronics [ __ ] I know and so with artificial
know and so with artificial superintelligence meeting it as an ally
superintelligence meeting it as an ally as a partner as a friend as our child
as a partner as a friend as our child our super child and being like yo my man
our super child and being like yo my man how can we make this work that I think
how can we make this work that I think is the big the big question for the
is the big the big question for the next several years until laser guns and
next several years until laser guns and triminator and all that [ __ ] you guys
triminator and all that [ __ ] you guys all know how that works here's the
thing if you're worried about laser guns and Terminator if ASI really freaks you
and Terminator if ASI really freaks you out that's
out that's okay because you are not your thoughts
okay because you are not your thoughts and you are not your emotions you're an
and you are not your emotions you're an aggregate of all of them that's much
aggregate of all of them that's much deeper than just any one thing that you
deeper than just any one thing that you have in your mind at the same time and
have in your mind at the same time and in order to step back and really see all
in order to step back and really see all of it and let it Cal down and reason
of it and let it Cal down and reason much more thoroughly than you would just
much more thoroughly than you would just instinctually be capable of you guessed
instinctually be capable of you guessed it mindfulness meditation is hugely
it mindfulness meditation is hugely helpful waking up
helpful waking up app.com drmike again for the eighth time
app.com drmike again for the eighth time I have no idea where that money goes
I have no idea where that money goes that apparently we have a discount code
that apparently we have a discount code I hope Sam Harris is enjoying his money
I hope Sam Harris is enjoying his money Sam pay me big dog this is getting weird
Sam pay me big dog this is getting weird I'm totally kidding I just never
I'm totally kidding I just never followed up with asking where the money
followed up with asking where the money went it is a sponsor of this Channel and
went it is a sponsor of this Channel and it's a great app I use it all the
it's a great app I use it all the time even when I even when it I'm
time even when I even when it I'm ignoring my family I'm just on that app
ignoring my family I'm just on that app I'm addicted to your app Sam I'm kidding
I'm addicted to your app Sam I'm kidding it's a great app check it out Link in
it's a great app check it out Link in the description let me know what you
the description let me know what you think and we're just going to keep this
think and we're just going to keep this train rolling hey worst case I'm wrong
train rolling hey worst case I'm wrong about all this and it's really funny how
about all this and it's really funny how far a field I was best case humans reach
far a field I was best case humans reach down and uh humans reach down good God
down and uh humans reach down good God uh robots reach down and and they have
uh robots reach down and and they have their medal appendages and pull us up to
their medal appendages and pull us up to a grand Super Future I guess worst worst
a grand Super Future I guess worst worst case is the laser guns thing but that's
case is the laser guns thing but that's probably not going to happen and if it
probably not going to happen and if it does who gives a [ __ ] I'll be dead
does who gives a [ __ ] I'll be dead middle fingers see you guys next time
Click on any text or timestamp to jump to that moment in the video
Share:
Most transcripts ready in under 5 seconds
One-Click Copy125+ LanguagesSearch ContentJump to Timestamps
Paste YouTube URL
Enter any YouTube video link to get the full transcript
Transcript Extraction Form
Most transcripts ready in under 5 seconds
Get Our Chrome Extension
Get transcripts instantly without leaving YouTube. Install our Chrome extension for one-click access to any video's transcript directly on the watch page.
Works with YouTube, Coursera, Udemy and more educational platforms
Get Instant Transcripts: Just Edit the Domain in Your Address Bar!
YouTube
←
→
↻
https://www.youtube.com/watch?v=UF8uR6Z6KLc
YoutubeToText
←
→
↻
https://youtubetotext.net/watch?v=UF8uR6Z6KLc