episode
stringlengths
45
100
text
stringlengths
1
528
timestamp_link
stringlengths
56
56
Guido van Rossum: Python | Lex Fridman Podcast #6
number, I, well, at my age I have to write it down, but I could imagine, I could remember
https://karpathy.ai/lexicap/0006-large.html#00:39:34.040
Guido van Rossum: Python | Lex Fridman Podcast #6
those seven numbers, or ten digits, and reproduce them in a while, if I sort of repeat them
https://karpathy.ai/lexicap/0006-large.html#00:39:39.560
Guido van Rossum: Python | Lex Fridman Podcast #6
to myself a few times, so that's a fairly conscious form of memorization.
https://karpathy.ai/lexicap/0006-large.html#00:39:46.240
Guido van Rossum: Python | Lex Fridman Podcast #6
On the other hand, how do I recognize your face, I have no idea.
https://karpathy.ai/lexicap/0006-large.html#00:39:53.320
Guido van Rossum: Python | Lex Fridman Podcast #6
My brain has a whole bunch of specialized hardware that knows how to recognize faces,
https://karpathy.ai/lexicap/0006-large.html#00:39:57.800
Guido van Rossum: Python | Lex Fridman Podcast #6
I don't know how much of that is sort of coded in our DNA, and how much of that is
https://karpathy.ai/lexicap/0006-large.html#00:40:04.080
Guido van Rossum: Python | Lex Fridman Podcast #6
trained over and over between the ages of zero and three, but somehow our brains know
https://karpathy.ai/lexicap/0006-large.html#00:40:10.200
Guido van Rossum: Python | Lex Fridman Podcast #6
how to do lots of things like that, that are useful in our interactions with other humans,
https://karpathy.ai/lexicap/0006-large.html#00:40:17.960
Guido van Rossum: Python | Lex Fridman Podcast #6
without really being conscious of how it's done anymore.
https://karpathy.ai/lexicap/0006-large.html#00:40:26.000
Guido van Rossum: Python | Lex Fridman Podcast #6
Right, so our actual day to day lives, we're operating at the very highest level of abstraction,
https://karpathy.ai/lexicap/0006-large.html#00:40:29.880
Guido van Rossum: Python | Lex Fridman Podcast #6
we're just not even conscious of all the little details underlying it.
https://karpathy.ai/lexicap/0006-large.html#00:40:36.200
Guido van Rossum: Python | Lex Fridman Podcast #6
There's compilers on top of, it's like turtles on top of turtles, or turtles all the way
https://karpathy.ai/lexicap/0006-large.html#00:40:39.760
Guido van Rossum: Python | Lex Fridman Podcast #6
down, there's compilers all the way down, but that's essentially, you say that there's
https://karpathy.ai/lexicap/0006-large.html#00:40:43.360
Guido van Rossum: Python | Lex Fridman Podcast #6
no magic, that's what I, what I was trying to get at, I think, is with Descartes started
https://karpathy.ai/lexicap/0006-large.html#00:40:48.200
Guido van Rossum: Python | Lex Fridman Podcast #6
this whole train of saying that there's no magic, I mean, there's all this beforehand.
https://karpathy.ai/lexicap/0006-large.html#00:40:54.920
Guido van Rossum: Python | Lex Fridman Podcast #6
Well didn't Descartes also have the notion though that the soul and the body were fundamentally
https://karpathy.ai/lexicap/0006-large.html#00:40:59.600
Guido van Rossum: Python | Lex Fridman Podcast #6
separate?
https://karpathy.ai/lexicap/0006-large.html#00:41:06.120
Guido van Rossum: Python | Lex Fridman Podcast #6
Separate, yeah, I think he had to write in God in there for political reasons, so I don't
https://karpathy.ai/lexicap/0006-large.html#00:41:07.120
Guido van Rossum: Python | Lex Fridman Podcast #6
know actually, I'm not a historian, but there's notions in there that all of reasoning, all
https://karpathy.ai/lexicap/0006-large.html#00:41:11.800
Guido van Rossum: Python | Lex Fridman Podcast #6
of human thought can be formalized.
https://karpathy.ai/lexicap/0006-large.html#00:41:17.880
Guido van Rossum: Python | Lex Fridman Podcast #6
I think that continued in the 20th century with Russell and with Gadot's incompleteness
https://karpathy.ai/lexicap/0006-large.html#00:41:20.120
Guido van Rossum: Python | Lex Fridman Podcast #6
theorem, this debate of what are the limits of the things that could be formalized, that's
https://karpathy.ai/lexicap/0006-large.html#00:41:28.480
Guido van Rossum: Python | Lex Fridman Podcast #6
where the Turing machine came along, and this exciting idea, I mean, underlying a lot of
https://karpathy.ai/lexicap/0006-large.html#00:41:33.120
Guido van Rossum: Python | Lex Fridman Podcast #6
computing that you can do quite a lot with a computer.
https://karpathy.ai/lexicap/0006-large.html#00:41:37.960
Guido van Rossum: Python | Lex Fridman Podcast #6
You can encode a lot of the stuff we're talking about in terms of recognizing faces and so
https://karpathy.ai/lexicap/0006-large.html#00:41:43.160
Guido van Rossum: Python | Lex Fridman Podcast #6
on, theoretically, in an algorithm that can then run on a computer.
https://karpathy.ai/lexicap/0006-large.html#00:41:47.640
Guido van Rossum: Python | Lex Fridman Podcast #6
And in that context, I'd like to ask programming in a philosophical way, what does it mean
https://karpathy.ai/lexicap/0006-large.html#00:41:53.960
Guido van Rossum: Python | Lex Fridman Podcast #6
to program a computer?
https://karpathy.ai/lexicap/0006-large.html#00:42:05.040
Guido van Rossum: Python | Lex Fridman Podcast #6
So you said you write a Python program or compiled a C++ program that compiles to some
https://karpathy.ai/lexicap/0006-large.html#00:42:06.480
Guido van Rossum: Python | Lex Fridman Podcast #6
byte code, it's forming layers, you're programming a layer of abstraction that's higher, how
https://karpathy.ai/lexicap/0006-large.html#00:42:13.360
Guido van Rossum: Python | Lex Fridman Podcast #6
do you see programming in that context?
https://karpathy.ai/lexicap/0006-large.html#00:42:21.200
Guido van Rossum: Python | Lex Fridman Podcast #6
Can it keep getting higher and higher levels of abstraction?
https://karpathy.ai/lexicap/0006-large.html#00:42:24.920
Guido van Rossum: Python | Lex Fridman Podcast #6
I think at some point the higher levels of abstraction will not be called programming
https://karpathy.ai/lexicap/0006-large.html#00:42:29.800
Guido van Rossum: Python | Lex Fridman Podcast #6
and they will not resemble what we call programming at the moment.
https://karpathy.ai/lexicap/0006-large.html#00:42:35.960
Guido van Rossum: Python | Lex Fridman Podcast #6
There will not be source code, I mean, there will still be source code sort of at a lower
https://karpathy.ai/lexicap/0006-large.html#00:42:44.720
Guido van Rossum: Python | Lex Fridman Podcast #6
level of the machine, just like there are still molecules and electrons and sort of
https://karpathy.ai/lexicap/0006-large.html#00:42:52.080
Guido van Rossum: Python | Lex Fridman Podcast #6
proteins in our brains, but, and so there's still programming and system administration
https://karpathy.ai/lexicap/0006-large.html#00:42:59.320
Guido van Rossum: Python | Lex Fridman Podcast #6
and who knows what, to keep the machine running, but what the machine does is a different level
https://karpathy.ai/lexicap/0006-large.html#00:43:09.120
Guido van Rossum: Python | Lex Fridman Podcast #6
of abstraction in a sense, and as far as I understand the way that for the last decade
https://karpathy.ai/lexicap/0006-large.html#00:43:15.960
Guido van Rossum: Python | Lex Fridman Podcast #6
or more people have made progress with things like facial recognition or the self driving
https://karpathy.ai/lexicap/0006-large.html#00:43:23.060
Guido van Rossum: Python | Lex Fridman Podcast #6
cars is all by endless, endless amounts of training data where at least as a lay person,
https://karpathy.ai/lexicap/0006-large.html#00:43:28.440
Guido van Rossum: Python | Lex Fridman Podcast #6
and I feel myself totally as a lay person in that field, it looks like the researchers
https://karpathy.ai/lexicap/0006-large.html#00:43:38.200
Guido van Rossum: Python | Lex Fridman Podcast #6
who publish the results don't necessarily know exactly how their algorithms work, and
https://karpathy.ai/lexicap/0006-large.html#00:43:47.420
Guido van Rossum: Python | Lex Fridman Podcast #6
I often get upset when I sort of read a sort of a fluff piece about Facebook in the newspaper
https://karpathy.ai/lexicap/0006-large.html#00:43:57.400
Guido van Rossum: Python | Lex Fridman Podcast #6
or social networks and they say, well, algorithms, and that's like a totally different interpretation
https://karpathy.ai/lexicap/0006-large.html#00:44:04.840
Guido van Rossum: Python | Lex Fridman Podcast #6
of the word algorithm, because for me, the way I was trained or what I learned when I
https://karpathy.ai/lexicap/0006-large.html#00:44:12.680
Guido van Rossum: Python | Lex Fridman Podcast #6
was eight or ten years old, an algorithm is a set of rules that you completely understand
https://karpathy.ai/lexicap/0006-large.html#00:44:19.240
Guido van Rossum: Python | Lex Fridman Podcast #6
that can be mathematically analyzed and you can prove things.
https://karpathy.ai/lexicap/0006-large.html#00:44:25.920
Guido van Rossum: Python | Lex Fridman Podcast #6
You can like prove that Aristotelian sieve produces all prime numbers and only prime
https://karpathy.ai/lexicap/0006-large.html#00:44:30.720
Guido van Rossum: Python | Lex Fridman Podcast #6
numbers.
https://karpathy.ai/lexicap/0006-large.html#00:44:37.840
Guido van Rossum: Python | Lex Fridman Podcast #6
Yeah.
https://karpathy.ai/lexicap/0006-large.html#00:44:38.840
Guido van Rossum: Python | Lex Fridman Podcast #6
So I don't know if you know who Andrej Karpathy is, I'm afraid not.
https://karpathy.ai/lexicap/0006-large.html#00:44:39.840
Guido van Rossum: Python | Lex Fridman Podcast #6
So he's a head of AI at Tesla now, but he was at Stanford before and he has this cheeky
https://karpathy.ai/lexicap/0006-large.html#00:44:44.360
Guido van Rossum: Python | Lex Fridman Podcast #6
way of calling this concept software 2.0.
https://karpathy.ai/lexicap/0006-large.html#00:44:51.980
Guido van Rossum: Python | Lex Fridman Podcast #6
So let me disentangle that for a second.
https://karpathy.ai/lexicap/0006-large.html#00:44:56.480
Guido van Rossum: Python | Lex Fridman Podcast #6
So kind of what you're referring to is the traditional, the algorithm, the concept of
https://karpathy.ai/lexicap/0006-large.html#00:45:00.120
Guido van Rossum: Python | Lex Fridman Podcast #6
an algorithm, something that's there, it's clear, you can read it, you understand it,
https://karpathy.ai/lexicap/0006-large.html#00:45:06.080
Guido van Rossum: Python | Lex Fridman Podcast #6
you can prove it's functioning as kind of software 1.0.
https://karpathy.ai/lexicap/0006-large.html#00:45:09.560
Guido van Rossum: Python | Lex Fridman Podcast #6
And what software 2.0 is, is exactly what you described, which is you have neural networks,
https://karpathy.ai/lexicap/0006-large.html#00:45:14.800
Guido van Rossum: Python | Lex Fridman Podcast #6
which is a type of machine learning that you feed a bunch of data and that neural network
https://karpathy.ai/lexicap/0006-large.html#00:45:21.920
Guido van Rossum: Python | Lex Fridman Podcast #6
learns to do a function.
https://karpathy.ai/lexicap/0006-large.html#00:45:26.600
Guido van Rossum: Python | Lex Fridman Podcast #6
All you specify is the inputs and the outputs you want and you can't look inside.
https://karpathy.ai/lexicap/0006-large.html#00:45:30.200
Guido van Rossum: Python | Lex Fridman Podcast #6
You can't analyze it.
https://karpathy.ai/lexicap/0006-large.html#00:45:35.220
Guido van Rossum: Python | Lex Fridman Podcast #6
All you can do is train this function to map the inputs to the outputs by giving a lot
https://karpathy.ai/lexicap/0006-large.html#00:45:37.040
Guido van Rossum: Python | Lex Fridman Podcast #6
of data.
https://karpathy.ai/lexicap/0006-large.html#00:45:41.920
Guido van Rossum: Python | Lex Fridman Podcast #6
And that's as programming becomes getting a lot of data.
https://karpathy.ai/lexicap/0006-large.html#00:45:42.920
Guido van Rossum: Python | Lex Fridman Podcast #6
That's what programming is.
https://karpathy.ai/lexicap/0006-large.html#00:45:47.040
Guido van Rossum: Python | Lex Fridman Podcast #6
Well, that would be programming 2.0.
https://karpathy.ai/lexicap/0006-large.html#00:45:48.920
Guido van Rossum: Python | Lex Fridman Podcast #6
To programming 2.0.
https://karpathy.ai/lexicap/0006-large.html#00:45:52.120
Guido van Rossum: Python | Lex Fridman Podcast #6
I wouldn't call that programming.
https://karpathy.ai/lexicap/0006-large.html#00:45:53.800
Guido van Rossum: Python | Lex Fridman Podcast #6
It's just a different activity.
https://karpathy.ai/lexicap/0006-large.html#00:45:55.600
Guido van Rossum: Python | Lex Fridman Podcast #6
Just like building organs out of cells is not called chemistry.
https://karpathy.ai/lexicap/0006-large.html#00:45:57.480
Guido van Rossum: Python | Lex Fridman Podcast #6
Well, so let's just step back and think sort of more generally, of course.
https://karpathy.ai/lexicap/0006-large.html#00:46:02.640
Guido van Rossum: Python | Lex Fridman Podcast #6
But you know, it's like as a parent teaching your kids, things can be called programming.
https://karpathy.ai/lexicap/0006-large.html#00:46:09.680
Guido van Rossum: Python | Lex Fridman Podcast #6
In that same sense, that's how programming is being used.
https://karpathy.ai/lexicap/0006-large.html#00:46:18.080
Guido van Rossum: Python | Lex Fridman Podcast #6
You're providing them data, examples, use cases.
https://karpathy.ai/lexicap/0006-large.html#00:46:22.720
Guido van Rossum: Python | Lex Fridman Podcast #6
So imagine writing a function not by, not with for loops and clearly readable text,
https://karpathy.ai/lexicap/0006-large.html#00:46:27.080
Guido van Rossum: Python | Lex Fridman Podcast #6
but more saying, well, here's a lot of examples of what this function should take.
https://karpathy.ai/lexicap/0006-large.html#00:46:36.680
Guido van Rossum: Python | Lex Fridman Podcast #6
And here's a lot of examples of when it takes those functions, it should do this.
https://karpathy.ai/lexicap/0006-large.html#00:46:42.760
Guido van Rossum: Python | Lex Fridman Podcast #6
And then figure out the rest.
https://karpathy.ai/lexicap/0006-large.html#00:46:47.860
Guido van Rossum: Python | Lex Fridman Podcast #6
So that's the 2.0 concept.
https://karpathy.ai/lexicap/0006-large.html#00:46:50.280
Guido van Rossum: Python | Lex Fridman Podcast #6
And so the question I have for you is like, it's a very fuzzy way.
https://karpathy.ai/lexicap/0006-large.html#00:46:52.640
Guido van Rossum: Python | Lex Fridman Podcast #6
This is the reality of a lot of these pattern recognition systems and so on.
https://karpathy.ai/lexicap/0006-large.html#00:46:58.560
Guido van Rossum: Python | Lex Fridman Podcast #6
It's a fuzzy way of quote unquote programming.
https://karpathy.ai/lexicap/0006-large.html#00:47:01.680
Guido van Rossum: Python | Lex Fridman Podcast #6
What do you think about this kind of world?
https://karpathy.ai/lexicap/0006-large.html#00:47:05.400
Guido van Rossum: Python | Lex Fridman Podcast #6
Should it be called something totally different than programming?
https://karpathy.ai/lexicap/0006-large.html#00:47:09.160
Guido van Rossum: Python | Lex Fridman Podcast #6
If you're a software engineer, does that mean you're designing systems that are very, can
https://karpathy.ai/lexicap/0006-large.html#00:47:13.640
Guido van Rossum: Python | Lex Fridman Podcast #6
be systematically tested, evaluated, they have a very specific specification and then this
https://karpathy.ai/lexicap/0006-large.html#00:47:21.000
Guido van Rossum: Python | Lex Fridman Podcast #6
other fuzzy software 2.0 world, machine learning world, that's something else totally?
https://karpathy.ai/lexicap/0006-large.html#00:47:28.140
Guido van Rossum: Python | Lex Fridman Podcast #6
Or is there some intermixing that's possible?
https://karpathy.ai/lexicap/0006-large.html#00:47:33.520
Guido van Rossum: Python | Lex Fridman Podcast #6
Well the question is probably only being asked because we don't quite know what that software
https://karpathy.ai/lexicap/0006-large.html#00:47:41.000
Guido van Rossum: Python | Lex Fridman Podcast #6
2.0 actually is.
https://karpathy.ai/lexicap/0006-large.html#00:47:48.600
Guido van Rossum: Python | Lex Fridman Podcast #6
And I think there is a truism that every task that AI has tackled in the past, at some point
https://karpathy.ai/lexicap/0006-large.html#00:47:51.400
Guido van Rossum: Python | Lex Fridman Podcast #6
we realized how it was done and then it was no longer considered part of artificial intelligence
https://karpathy.ai/lexicap/0006-large.html#00:48:02.960
Guido van Rossum: Python | Lex Fridman Podcast #6
because it was no longer necessary to use that term.
https://karpathy.ai/lexicap/0006-large.html#00:48:09.160
Guido van Rossum: Python | Lex Fridman Podcast #6
It was just, oh now we know how to do this.
https://karpathy.ai/lexicap/0006-large.html#00:48:15.200
Guido van Rossum: Python | Lex Fridman Podcast #6
And a new field of science or engineering has been developed and I don't know if sort
https://karpathy.ai/lexicap/0006-large.html#00:48:21.600
Guido van Rossum: Python | Lex Fridman Podcast #6
of every form of learning or sort of controlling computer systems should always be called programming.
https://karpathy.ai/lexicap/0006-large.html#00:48:30.320
Guido van Rossum: Python | Lex Fridman Podcast #6
So I don't know, maybe I'm focused too much on the terminology.
https://karpathy.ai/lexicap/0006-large.html#00:48:39.000
Guido van Rossum: Python | Lex Fridman Podcast #6
But I expect that there just will be different concepts where people with sort of different
https://karpathy.ai/lexicap/0006-large.html#00:48:43.720