episode
stringlengths 45
100
| text
stringlengths 1
528
| timestamp_link
stringlengths 56
56
|
|---|---|---|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
my hunch is that we're going to understand
|
https://karpathy.ai/lexicap/0001-large.html#01:00:54.720
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
how to build AGI before we fully understand
|
https://karpathy.ai/lexicap/0001-large.html#01:00:57.040
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
how our brains work, just like we understood
|
https://karpathy.ai/lexicap/0001-large.html#01:00:59.520
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
how to build flying machines long before
|
https://karpathy.ai/lexicap/0001-large.html#01:01:02.600
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
we were able to build a mechanical bird.
|
https://karpathy.ai/lexicap/0001-large.html#01:01:05.560
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
Yeah, that's right.
|
https://karpathy.ai/lexicap/0001-large.html#01:01:07.800
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
You've given the example exactly of mechanical birds
|
https://karpathy.ai/lexicap/0001-large.html#01:01:08.640
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
and airplanes and airplanes do a pretty good job
|
https://karpathy.ai/lexicap/0001-large.html#01:01:13.280
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
of flying without really mimicking bird flight.
|
https://karpathy.ai/lexicap/0001-large.html#01:01:15.680
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
And even now after 100 years later,
|
https://karpathy.ai/lexicap/0001-large.html#01:01:18.560
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
did you see the Ted talk with this German mechanical bird?
|
https://karpathy.ai/lexicap/0001-large.html#01:01:20.920
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
I heard you mention it.
|
https://karpathy.ai/lexicap/0001-large.html#01:01:23.880
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
Check it out, it's amazing.
|
https://karpathy.ai/lexicap/0001-large.html#01:01:25.040
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
But even after that, right,
|
https://karpathy.ai/lexicap/0001-large.html#01:01:26.520
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
we still don't fly in mechanical birds
|
https://karpathy.ai/lexicap/0001-large.html#01:01:27.760
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
because it turned out the way we came up with was simpler
|
https://karpathy.ai/lexicap/0001-large.html#01:01:29.360
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
and it's better for our purposes.
|
https://karpathy.ai/lexicap/0001-large.html#01:01:32.720
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
And I think it might be the same there.
|
https://karpathy.ai/lexicap/0001-large.html#01:01:33.840
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
That's one lesson.
|
https://karpathy.ai/lexicap/0001-large.html#01:01:35.280
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
And another lesson, it's more what our paper was about.
|
https://karpathy.ai/lexicap/0001-large.html#01:01:37.520
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
First, as a physicist thought it was fascinating
|
https://karpathy.ai/lexicap/0001-large.html#01:01:42.640
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
how there's a very close mathematical relationship
|
https://karpathy.ai/lexicap/0001-large.html#01:01:45.800
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
actually between our artificial neural networks
|
https://karpathy.ai/lexicap/0001-large.html#01:01:48.240
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
and a lot of things that we've studied for in physics
|
https://karpathy.ai/lexicap/0001-large.html#01:01:50.800
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
go by nerdy names like the renormalization group equation
|
https://karpathy.ai/lexicap/0001-large.html#01:01:54.560
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
and Hamiltonians and yada, yada, yada.
|
https://karpathy.ai/lexicap/0001-large.html#01:01:57.520
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
And when you look a little more closely at this,
|
https://karpathy.ai/lexicap/0001-large.html#01:01:59.800
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
you have,
|
https://karpathy.ai/lexicap/0001-large.html#01:02:05.720
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
at first I was like, well, there's something crazy here
|
https://karpathy.ai/lexicap/0001-large.html#01:02:10.320
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
that doesn't make sense.
|
https://karpathy.ai/lexicap/0001-large.html#01:02:12.360
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
Because we know that if you even want to build
|
https://karpathy.ai/lexicap/0001-large.html#01:02:13.520
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
a super simple neural network to tell apart cat pictures
|
https://karpathy.ai/lexicap/0001-large.html#01:02:19.200
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
and dog pictures, right,
|
https://karpathy.ai/lexicap/0001-large.html#01:02:22.560
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
that you can do that very, very well now.
|
https://karpathy.ai/lexicap/0001-large.html#01:02:23.400
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
But if you think about it a little bit,
|
https://karpathy.ai/lexicap/0001-large.html#01:02:25.400
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
you convince yourself it must be impossible
|
https://karpathy.ai/lexicap/0001-large.html#01:02:27.520
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
because if I have one megapixel,
|
https://karpathy.ai/lexicap/0001-large.html#01:02:29.080
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
even if each pixel is just black or white,
|
https://karpathy.ai/lexicap/0001-large.html#01:02:31.920
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
there's two to the power of 1 million possible images,
|
https://karpathy.ai/lexicap/0001-large.html#01:02:34.160
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
which is way more than there are atoms in our universe,
|
https://karpathy.ai/lexicap/0001-large.html#01:02:36.960
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
right, so in order to,
|
https://karpathy.ai/lexicap/0001-large.html#01:02:38.960
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
and then for each one of those,
|
https://karpathy.ai/lexicap/0001-large.html#01:02:42.040
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
I have to assign a number,
|
https://karpathy.ai/lexicap/0001-large.html#01:02:43.200
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
which is the probability that it's a dog.
|
https://karpathy.ai/lexicap/0001-large.html#01:02:44.640
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
So an arbitrary function of images
|
https://karpathy.ai/lexicap/0001-large.html#01:02:47.080
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
is a list of more numbers than there are atoms in our universe.
|
https://karpathy.ai/lexicap/0001-large.html#01:02:49.440
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
So clearly I can't store that under the hood of my GPU
|
https://karpathy.ai/lexicap/0001-large.html#01:02:54.440
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
or my computer, yet somehow it works.
|
https://karpathy.ai/lexicap/0001-large.html#01:02:57.360
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
So what does that mean?
|
https://karpathy.ai/lexicap/0001-large.html#01:03:00.640
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
Well, it means that out of all of the problems
|
https://karpathy.ai/lexicap/0001-large.html#01:03:01.480
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
that you could try to solve with a neural network,
|
https://karpathy.ai/lexicap/0001-large.html#01:03:04.960
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
almost all of them are impossible to solve
|
https://karpathy.ai/lexicap/0001-large.html#01:03:10.120
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
with a reasonably sized one.
|
https://karpathy.ai/lexicap/0001-large.html#01:03:12.880
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
But then what we showed in our paper
|
https://karpathy.ai/lexicap/0001-large.html#01:03:15.480
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
was that the fraction, the kind of problems,
|
https://karpathy.ai/lexicap/0001-large.html#01:03:17.440
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
the fraction of all the problems
|
https://karpathy.ai/lexicap/0001-large.html#01:03:22.360
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
that you could possibly pose,
|
https://karpathy.ai/lexicap/0001-large.html#01:03:23.800
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
that we actually care about given the laws of physics
|
https://karpathy.ai/lexicap/0001-large.html#01:03:26.520
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
is also an infinite testimony, tiny little part.
|
https://karpathy.ai/lexicap/0001-large.html#01:03:29.480
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
And amazingly, they're basically the same part.
|
https://karpathy.ai/lexicap/0001-large.html#01:03:32.480
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
Yeah, it's almost like our world was created for,
|
https://karpathy.ai/lexicap/0001-large.html#01:03:35.440
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
I mean, they kind of come together.
|
https://karpathy.ai/lexicap/0001-large.html#01:03:37.560
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
Yeah, well, you could say maybe where the world was created
|
https://karpathy.ai/lexicap/0001-large.html#01:03:39.000
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
for us, but I have a more modest interpretation,
|
https://karpathy.ai/lexicap/0001-large.html#01:03:42.800
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
which is that the world was created for us,
|
https://karpathy.ai/lexicap/0001-large.html#01:03:44.960
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
but I have a more modest interpretation,
|
https://karpathy.ai/lexicap/0001-large.html#01:03:46.680
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
which is that instead evolution endowed us
|
https://karpathy.ai/lexicap/0001-large.html#01:03:48.040
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
with neural networks precisely for that reason.
|
https://karpathy.ai/lexicap/0001-large.html#01:03:50.360
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
Because this particular architecture,
|
https://karpathy.ai/lexicap/0001-large.html#01:03:53.120
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
as opposed to the one in your laptop,
|
https://karpathy.ai/lexicap/0001-large.html#01:03:54.640
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
is very, very well adapted to solving the kind of problems
|
https://karpathy.ai/lexicap/0001-large.html#01:03:56.040
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
that nature kept presenting our ancestors with.
|
https://karpathy.ai/lexicap/0001-large.html#01:04:02.480
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
So it makes sense that why do we have a brain
|
https://karpathy.ai/lexicap/0001-large.html#01:04:05.560
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
in the first place?
|
https://karpathy.ai/lexicap/0001-large.html#01:04:08.120
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
It's to be able to make predictions about the future
|
https://karpathy.ai/lexicap/0001-large.html#01:04:09.280
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
and so on.
|
https://karpathy.ai/lexicap/0001-large.html#01:04:11.880
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
So if we had a sucky system, which could never solve it,
|
https://karpathy.ai/lexicap/0001-large.html#01:04:12.880
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
we wouldn't have a world.
|
https://karpathy.ai/lexicap/0001-large.html#01:04:16.440
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
So this is, I think, a very beautiful fact.
|
https://karpathy.ai/lexicap/0001-large.html#01:04:18.280
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
Yeah.
|
https://karpathy.ai/lexicap/0001-large.html#01:04:23.680
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
We also realize that there's been earlier work
|
https://karpathy.ai/lexicap/0001-large.html#01:04:24.520
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
on why deeper networks are good,
|
https://karpathy.ai/lexicap/0001-large.html#01:04:29.000
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
but we were able to show an additional cool fact there,
|
https://karpathy.ai/lexicap/0001-large.html#01:04:32.040
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
which is that even incredibly simple problems,
|
https://karpathy.ai/lexicap/0001-large.html#01:04:34.680
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
like suppose I give you a thousand numbers
|
https://karpathy.ai/lexicap/0001-large.html#01:04:38.360
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
and ask you to multiply them together,
|
https://karpathy.ai/lexicap/0001-large.html#01:04:41.080
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
and you can write a few lines of code, boom, done, trivial.
|
https://karpathy.ai/lexicap/0001-large.html#01:04:42.720
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
If you just try to do that with a neural network
|
https://karpathy.ai/lexicap/0001-large.html#01:04:46.680
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
that has only one single hidden layer in it,
|
https://karpathy.ai/lexicap/0001-large.html#01:04:49.520
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
you can do it,
|
https://karpathy.ai/lexicap/0001-large.html#01:04:52.440
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
but you're going to need two to the power of a thousand
|
https://karpathy.ai/lexicap/0001-large.html#01:04:54.320
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
neurons to multiply a thousand numbers,
|
https://karpathy.ai/lexicap/0001-large.html#01:04:57.360
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
which is, again, more neurons than there are atoms
|
https://karpathy.ai/lexicap/0001-large.html#01:05:00.920
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
in our universe.
|
https://karpathy.ai/lexicap/0001-large.html#01:05:02.520
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
That's fascinating.
|
https://karpathy.ai/lexicap/0001-large.html#01:05:04.600
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
But if you allow yourself to make it a deep network
|
https://karpathy.ai/lexicap/0001-large.html#01:05:05.480
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
with many layers, you only need 4,000 neurons.
|
https://karpathy.ai/lexicap/0001-large.html#01:05:09.960
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
It's perfectly feasible.
|
https://karpathy.ai/lexicap/0001-large.html#01:05:13.240
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
That's really interesting.
|
https://karpathy.ai/lexicap/0001-large.html#01:05:16.400
|
Max Tegmark: Life 3.0 | Lex Fridman Podcast #1
|
Yeah.
|
https://karpathy.ai/lexicap/0001-large.html#01:05:17.960
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.