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
or can we just be super impressed
https://karpathy.ai/lexicap/0011-large.html#00:13:29.200
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
by the thing working extremely well and that's sufficient?
https://karpathy.ai/lexicap/0011-large.html#00:13:31.160
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
There's an old saying,
https://karpathy.ai/lexicap/0011-large.html#00:13:35.120
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
and I don't know who brought it up first,
https://karpathy.ai/lexicap/0011-large.html#00:13:36.760
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
which says, there's nothing more practical
https://karpathy.ai/lexicap/0011-large.html#00:13:39.360
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
than a good theory.
https://karpathy.ai/lexicap/0011-large.html#00:13:42.480
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
And a good theory of problem solving
https://karpathy.ai/lexicap/0011-large.html#00:13:43.720
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
under limited resources,
https://karpathy.ai/lexicap/0011-large.html#00:13:52.800
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
like here in this universe or on this little planet,
https://karpathy.ai/lexicap/0011-large.html#00:13:54.360
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
has to take into account these limited resources.
https://karpathy.ai/lexicap/0011-large.html#00:13:58.480
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
And so probably there is locking a theory,
https://karpathy.ai/lexicap/0011-large.html#00:14:01.800
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
which is related to what we already have,
https://karpathy.ai/lexicap/0011-large.html#00:14:08.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
these asymptotically optimal problem solvers,
https://karpathy.ai/lexicap/0011-large.html#00:14:10.760
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
which tells us what we need in addition to that
https://karpathy.ai/lexicap/0011-large.html#00:14:14.400
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
to come up with a practically optimal problem solver.
https://karpathy.ai/lexicap/0011-large.html#00:14:18.520
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So I believe we will have something like that.
https://karpathy.ai/lexicap/0011-large.html#00:14:22.640
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
And maybe just a few little tiny twists are necessary
https://karpathy.ai/lexicap/0011-large.html#00:14:27.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
to change what we already have,
https://karpathy.ai/lexicap/0011-large.html#00:14:30.520
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
to come up with that as well.
https://karpathy.ai/lexicap/0011-large.html#00:14:34.280
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
As long as we don't have that,
https://karpathy.ai/lexicap/0011-large.html#00:14:36.320
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
we admit that we are taking suboptimal ways
https://karpathy.ai/lexicap/0011-large.html#00:14:37.720
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
and recurrent neural networks and long short term memory
https://karpathy.ai/lexicap/0011-large.html#00:14:42.560
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
for equipped with local search techniques.
https://karpathy.ai/lexicap/0011-large.html#00:14:45.960
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
And we are happy that it works better
https://karpathy.ai/lexicap/0011-large.html#00:14:50.400
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
than any competing methods,
https://karpathy.ai/lexicap/0011-large.html#00:14:53.520
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
but that doesn't mean that we think we are done.
https://karpathy.ai/lexicap/0011-large.html#00:14:55.480
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
You've said that an AGI system
https://karpathy.ai/lexicap/0011-large.html#00:15:00.800
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
will ultimately be a simple one.
https://karpathy.ai/lexicap/0011-large.html#00:15:02.720
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
A general intelligence system
https://karpathy.ai/lexicap/0011-large.html#00:15:05.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
will ultimately be a simple one.
https://karpathy.ai/lexicap/0011-large.html#00:15:06.200
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Maybe a pseudocode of a few lines
https://karpathy.ai/lexicap/0011-large.html#00:15:08.000
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
will be able to describe it.
https://karpathy.ai/lexicap/0011-large.html#00:15:10.240
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Can you talk through your intuition behind this idea,
https://karpathy.ai/lexicap/0011-large.html#00:15:11.840
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
why you feel that at its core,
https://karpathy.ai/lexicap/0011-large.html#00:15:16.760
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
intelligence is a simple algorithm?
https://karpathy.ai/lexicap/0011-large.html#00:15:21.480
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Experience tells us that the stuff that works best
https://karpathy.ai/lexicap/0011-large.html#00:15:26.920
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
is really simple.
https://karpathy.ai/lexicap/0011-large.html#00:15:31.680
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So the asymptotically optimal ways of solving problems,
https://karpathy.ai/lexicap/0011-large.html#00:15:33.120
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
if you look at them,
https://karpathy.ai/lexicap/0011-large.html#00:15:37.680
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
they're just a few lines of code, it's really true.
https://karpathy.ai/lexicap/0011-large.html#00:15:38.800
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Although they are these amazing properties,
https://karpathy.ai/lexicap/0011-large.html#00:15:41.840
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
just a few lines of code.
https://karpathy.ai/lexicap/0011-large.html#00:15:44.000
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Then the most promising and most useful practical things,
https://karpathy.ai/lexicap/0011-large.html#00:15:45.800
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
maybe don't have this proof of optimality
https://karpathy.ai/lexicap/0011-large.html#00:15:53.760
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
associated with them.
https://karpathy.ai/lexicap/0011-large.html#00:15:56.360
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
However, they are also just a few lines of code.
https://karpathy.ai/lexicap/0011-large.html#00:15:57.800
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
The most successful recurrent neural networks,
https://karpathy.ai/lexicap/0011-large.html#00:16:00.880
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
you can write them down in five lines of pseudocode.
https://karpathy.ai/lexicap/0011-large.html#00:16:05.080
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
That's a beautiful, almost poetic idea,
https://karpathy.ai/lexicap/0011-large.html#00:16:08.320
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
but what you're describing there
https://karpathy.ai/lexicap/0011-large.html#00:16:10.920
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
is the lines of pseudocode are sitting on top
https://karpathy.ai/lexicap/0011-large.html#00:16:15.640
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
of layers and layers of abstractions in a sense.
https://karpathy.ai/lexicap/0011-large.html#00:16:18.200
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So you're saying at the very top,
https://karpathy.ai/lexicap/0011-large.html#00:16:22.280
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
it'll be a beautifully written sort of algorithm.
https://karpathy.ai/lexicap/0011-large.html#00:16:25.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
But do you think that there's many layers of abstractions
https://karpathy.ai/lexicap/0011-large.html#00:16:31.120
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
we have to first learn to construct?
https://karpathy.ai/lexicap/0011-large.html#00:16:33.960
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Yeah, of course, we are building on all these
https://karpathy.ai/lexicap/0011-large.html#00:16:36.880
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
great abstractions that people have invented over the millennia,
https://karpathy.ai/lexicap/0011-large.html#00:16:40.400
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
such as matrix multiplications and real numbers
https://karpathy.ai/lexicap/0011-large.html#00:16:45.080
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
and basic arithmetics and calculus
https://karpathy.ai/lexicap/0011-large.html#00:16:50.520
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
and derivations of error functions
https://karpathy.ai/lexicap/0011-large.html#00:16:56.440
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
and derivatives of error functions and stuff like that.
https://karpathy.ai/lexicap/0011-large.html#00:17:00.240
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So without that language that greatly simplifies
https://karpathy.ai/lexicap/0011-large.html#00:17:04.400
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
our way of thinking about these problems,
https://karpathy.ai/lexicap/0011-large.html#00:17:09.400
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
we couldn't do anything.
https://karpathy.ai/lexicap/0011-large.html#00:17:13.840
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So in that sense, as always,
https://karpathy.ai/lexicap/0011-large.html#00:17:14.760
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
we are standing on the shoulders of the giants
https://karpathy.ai/lexicap/0011-large.html#00:17:16.520
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
who in the past simplified the problem
https://karpathy.ai/lexicap/0011-large.html#00:17:19.520
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
of problem solving so much
https://karpathy.ai/lexicap/0011-large.html#00:17:24.200
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
that now we have a chance to do the final step.
https://karpathy.ai/lexicap/0011-large.html#00:17:26.320
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So the final step will be a simple one.
https://karpathy.ai/lexicap/0011-large.html#00:17:29.960
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
If we take a step back through all of human civilization
https://karpathy.ai/lexicap/0011-large.html#00:17:33.960
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
and just the universe in general,
https://karpathy.ai/lexicap/0011-large.html#00:17:36.680
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
how do you think about evolution
https://karpathy.ai/lexicap/0011-large.html#00:17:40.000
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
and what if creating a universe
https://karpathy.ai/lexicap/0011-large.html#00:17:41.400
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
is required to achieve this final step?
https://karpathy.ai/lexicap/0011-large.html#00:17:44.480
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
What if going through the very painful
https://karpathy.ai/lexicap/0011-large.html#00:17:47.240
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
and inefficient process of evolution is needed
https://karpathy.ai/lexicap/0011-large.html#00:17:50.880
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
to come up with this set of abstractions
https://karpathy.ai/lexicap/0011-large.html#00:17:53.800
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
that ultimately lead to intelligence?
https://karpathy.ai/lexicap/0011-large.html#00:17:55.800
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Do you think there's a shortcut
https://karpathy.ai/lexicap/0011-large.html#00:17:57.720
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
or do you think we have to create something like our universe
https://karpathy.ai/lexicap/0011-large.html#00:18:00.720
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
in order to create something like human level intelligence?
https://karpathy.ai/lexicap/0011-large.html#00:18:04.560
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So far, the only example we have is this one,
https://karpathy.ai/lexicap/0011-large.html#00:18:09.400
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
this universe in which we are living.
https://karpathy.ai/lexicap/0011-large.html#00:18:13.080
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Do you think we can do better?
https://karpathy.ai/lexicap/0011-large.html#00:18:14.880
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Maybe not, but we are part of this whole process.
https://karpathy.ai/lexicap/0011-large.html#00:18:20.800
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So apparently, so it might be the case
https://karpathy.ai/lexicap/0011-large.html#00:18:24.960
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
that the code that runs the universe
https://karpathy.ai/lexicap/0011-large.html#00:18:29.920
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
is really, really simple.
https://karpathy.ai/lexicap/0011-large.html#00:18:32.080
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Everything points to that possibility
https://karpathy.ai/lexicap/0011-large.html#00:18:33.600
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
because gravity and other basic forces
https://karpathy.ai/lexicap/0011-large.html#00:18:35.760
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
are really simple laws that can be easily described
https://karpathy.ai/lexicap/0011-large.html#00:18:39.080
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
also in just a few lines of code basically.
https://karpathy.ai/lexicap/0011-large.html#00:18:43.280
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
And then there are these other events
https://karpathy.ai/lexicap/0011-large.html#00:18:46.240
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
that the apparently random events
https://karpathy.ai/lexicap/0011-large.html#00:18:51.360
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
in the history of the universe,
https://karpathy.ai/lexicap/0011-large.html#00:18:54.280
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
which as far as we know at the moment
https://karpathy.ai/lexicap/0011-large.html#00:18:55.760
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
don't have a compact code, but who knows?
https://karpathy.ai/lexicap/0011-large.html#00:18:58.000
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Maybe somebody in the near future
https://karpathy.ai/lexicap/0011-large.html#00:19:00.600