episode
stringlengths
45
100
text
stringlengths
1
528
timestamp_link
stringlengths
56
56
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
is going to figure out the pseudo random generator
https://karpathy.ai/lexicap/0011-large.html#00:19:02.440
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
which is computing whether the measurement
https://karpathy.ai/lexicap/0011-large.html#00:19:06.240
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
of that spin up or down thing here
https://karpathy.ai/lexicap/0011-large.html#00:19:11.800
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
is going to be positive or negative.
https://karpathy.ai/lexicap/0011-large.html#00:19:15.320
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Underlying quantum mechanics.
https://karpathy.ai/lexicap/0011-large.html#00:19:17.840
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Yes.
https://karpathy.ai/lexicap/0011-large.html#00:19:19.280
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Do you ultimately think quantum mechanics
https://karpathy.ai/lexicap/0011-large.html#00:19:20.120
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
is a pseudo random number generator?
https://karpathy.ai/lexicap/0011-large.html#00:19:22.600
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So it's all deterministic.
https://karpathy.ai/lexicap/0011-large.html#00:19:24.640
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
There's no randomness in our universe.
https://karpathy.ai/lexicap/0011-large.html#00:19:26.320
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Does God play dice?
https://karpathy.ai/lexicap/0011-large.html#00:19:28.200
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So a couple of years ago, a famous physicist,
https://karpathy.ai/lexicap/0011-large.html#00:19:31.120
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
quantum physicist, Anton Zeilinger,
https://karpathy.ai/lexicap/0011-large.html#00:19:34.680
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
he wrote an essay in nature
https://karpathy.ai/lexicap/0011-large.html#00:19:37.680
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
and it started more or less like that.
https://karpathy.ai/lexicap/0011-large.html#00:19:40.080
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
One of the fundamental insights of the 20th century
https://karpathy.ai/lexicap/0011-large.html#00:19:45.360
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
was that the universe is fundamentally random
https://karpathy.ai/lexicap/0011-large.html#00:19:50.360
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
on the quantum level.
https://karpathy.ai/lexicap/0011-large.html#00:19:57.360
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
And that whenever you measure spin up or down
https://karpathy.ai/lexicap/0011-large.html#00:20:00.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
or something like that,
https://karpathy.ai/lexicap/0011-large.html#00:20:04.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
a new bit of information enters the history of the universe.
https://karpathy.ai/lexicap/0011-large.html#00:20:05.440
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
And while I was reading that,
https://karpathy.ai/lexicap/0011-large.html#00:20:12.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
I was already typing the response
https://karpathy.ai/lexicap/0011-large.html#00:20:13.200
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
and they had to publish it.
https://karpathy.ai/lexicap/0011-large.html#00:20:16.560
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Because I was right, that there is no evidence,
https://karpathy.ai/lexicap/0011-large.html#00:20:17.880
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
no physical evidence for that.
https://karpathy.ai/lexicap/0011-large.html#00:20:21.640
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So there's an alternative explanation
https://karpathy.ai/lexicap/0011-large.html#00:20:25.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
where everything that we consider random
https://karpathy.ai/lexicap/0011-large.html#00:20:27.720
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
is actually pseudo random,
https://karpathy.ai/lexicap/0011-large.html#00:20:30.680
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
such as the decimal expansion of pi,
https://karpathy.ai/lexicap/0011-large.html#00:20:33.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
3.141 and so on, which looks random, but isn't.
https://karpathy.ai/lexicap/0011-large.html#00:20:35.960
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So pi is interesting because every three digits
https://karpathy.ai/lexicap/0011-large.html#00:20:41.680
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
sequence, every sequence of three digits
https://karpathy.ai/lexicap/0011-large.html#00:20:45.400
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
appears roughly one in a thousand times.
https://karpathy.ai/lexicap/0011-large.html#00:20:50.400
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
And every five digit sequence
https://karpathy.ai/lexicap/0011-large.html#00:20:53.400
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
appears roughly one in 10,000 times,
https://karpathy.ai/lexicap/0011-large.html#00:20:57.400
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
what you would expect if it was random.
https://karpathy.ai/lexicap/0011-large.html#00:21:01.080
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
But there's a very short algorithm,
https://karpathy.ai/lexicap/0011-large.html#00:21:04.400
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
a short program that computes all of that.
https://karpathy.ai/lexicap/0011-large.html#00:21:06.680
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So it's extremely compressible.
https://karpathy.ai/lexicap/0011-large.html#00:21:09.120
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
And who knows, maybe tomorrow,
https://karpathy.ai/lexicap/0011-large.html#00:21:11.320
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
somebody, some grad student at CERN goes back
https://karpathy.ai/lexicap/0011-large.html#00:21:12.640
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
over all these data points, better decay and whatever,
https://karpathy.ai/lexicap/0011-large.html#00:21:15.640
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
and figures out, oh, it's the second billion digits of pi
https://karpathy.ai/lexicap/0011-large.html#00:21:20.120
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
or something like that.
https://karpathy.ai/lexicap/0011-large.html#00:21:24.760
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
We don't have any fundamental reason at the moment
https://karpathy.ai/lexicap/0011-large.html#00:21:26.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
to believe that this is truly random
https://karpathy.ai/lexicap/0011-large.html#00:21:29.080
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
and not just a deterministic video game.
https://karpathy.ai/lexicap/0011-large.html#00:21:33.600
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
If it was a deterministic video game,
https://karpathy.ai/lexicap/0011-large.html#00:21:36.680
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
it would be much more beautiful.
https://karpathy.ai/lexicap/0011-large.html#00:21:38.680
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Because beauty is simplicity.
https://karpathy.ai/lexicap/0011-large.html#00:21:40.360
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
And many of the basic laws of the universe,
https://karpathy.ai/lexicap/0011-large.html#00:21:43.840
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
like gravity and the other basic forces are very simple.
https://karpathy.ai/lexicap/0011-large.html#00:21:47.000
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So very short programs can explain what these are doing.
https://karpathy.ai/lexicap/0011-large.html#00:21:51.360
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
And it would be awful and ugly.
https://karpathy.ai/lexicap/0011-large.html#00:21:56.760
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
The universe would be ugly.
https://karpathy.ai/lexicap/0011-large.html#00:22:00.560
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
The history of the universe would be ugly
https://karpathy.ai/lexicap/0011-large.html#00:22:01.720
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
if for the extra things, the random,
https://karpathy.ai/lexicap/0011-large.html#00:22:04.000
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
the seemingly random data points that we get all the time,
https://karpathy.ai/lexicap/0011-large.html#00:22:06.240
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
that we really need a huge number of extra bits
https://karpathy.ai/lexicap/0011-large.html#00:22:11.080
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
to describe all these extra bits of information.
https://karpathy.ai/lexicap/0011-large.html#00:22:15.160
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So as long as we don't have evidence
https://karpathy.ai/lexicap/0011-large.html#00:22:22.160
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
that there is no short program
https://karpathy.ai/lexicap/0011-large.html#00:22:24.800
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
that computes the entire history of the entire universe,
https://karpathy.ai/lexicap/0011-large.html#00:22:26.920
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
we are, as scientists, compelled to look further
https://karpathy.ai/lexicap/0011-large.html#00:22:31.240
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
for that shortest program.
https://karpathy.ai/lexicap/0011-large.html#00:22:36.600
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Your intuition says there exists a program
https://karpathy.ai/lexicap/0011-large.html#00:22:39.760
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
that can backtrack to the creation of the universe.
https://karpathy.ai/lexicap/0011-large.html#00:22:43.760
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Yeah.
https://karpathy.ai/lexicap/0011-large.html#00:22:47.760
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So it can give the shortest path
https://karpathy.ai/lexicap/0011-large.html#00:22:48.600
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
to the creation of the universe.
https://karpathy.ai/lexicap/0011-large.html#00:22:49.440
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Yes.
https://karpathy.ai/lexicap/0011-large.html#00:22:50.480
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Including all the entanglement things
https://karpathy.ai/lexicap/0011-large.html#00:22:51.320
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
and all the spin up and down measures
https://karpathy.ai/lexicap/0011-large.html#00:22:54.480
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
that have been taken place since 13.8 billion years ago.
https://karpathy.ai/lexicap/0011-large.html#00:22:57.800
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So we don't have a proof that it is random.
https://karpathy.ai/lexicap/0011-large.html#00:23:06.840
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
We don't have a proof that it is compressible
https://karpathy.ai/lexicap/0011-large.html#00:23:11.840
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
to a short program.
https://karpathy.ai/lexicap/0011-large.html#00:23:15.600
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
But as long as we don't have that proof,
https://karpathy.ai/lexicap/0011-large.html#00:23:16.760
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
we are obliged as scientists to keep looking
https://karpathy.ai/lexicap/0011-large.html#00:23:18.240
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
for that simple explanation.
https://karpathy.ai/lexicap/0011-large.html#00:23:21.680
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Absolutely.
https://karpathy.ai/lexicap/0011-large.html#00:23:23.600
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So you said the simplicity is beautiful or beauty is simple.
https://karpathy.ai/lexicap/0011-large.html#00:23:24.440
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Either one works.
https://karpathy.ai/lexicap/0011-large.html#00:23:27.880
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
But you also work on curiosity, discovery,
https://karpathy.ai/lexicap/0011-large.html#00:23:29.440
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
the romantic notion of randomness, of serendipity,
https://karpathy.ai/lexicap/0011-large.html#00:23:34.560
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
of being surprised by things that are about you.
https://karpathy.ai/lexicap/0011-large.html#00:23:39.000
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
In our poetic notion of reality,
https://karpathy.ai/lexicap/0011-large.html#00:23:45.920
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
we think it's kind of like,
https://karpathy.ai/lexicap/0011-large.html#00:23:49.600
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
poetic notion of reality, we think as humans
https://karpathy.ai/lexicap/0011-large.html#00:23:51.640
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
require randomness.
https://karpathy.ai/lexicap/0011-large.html#00:23:54.920
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So you don't find randomness beautiful.
https://karpathy.ai/lexicap/0011-large.html#00:23:56.400
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
You find simple determinism beautiful.
https://karpathy.ai/lexicap/0011-large.html#00:23:59.000
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Yeah.
https://karpathy.ai/lexicap/0011-large.html#00:24:04.880
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Okay.
https://karpathy.ai/lexicap/0011-large.html#00:24:05.720
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So why?
https://karpathy.ai/lexicap/0011-large.html#00:24:07.520
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Why?
https://karpathy.ai/lexicap/0011-large.html#00:24:08.560
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Because the explanation becomes shorter.
https://karpathy.ai/lexicap/0011-large.html#00:24:09.400
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
A universe that is compressible to a short program
https://karpathy.ai/lexicap/0011-large.html#00:24:13.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
is much more elegant and much more beautiful
https://karpathy.ai/lexicap/0011-large.html#00:24:18.040