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
No, because that's the nature of power play that
https://karpathy.ai/lexicap/0011-large.html#00:34:47.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
it's always trying to break its current
https://karpathy.ai/lexicap/0011-large.html#00:34:50.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
generalization abilities by coming up with a new
https://karpathy.ai/lexicap/0011-large.html#00:34:53.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
problem which is beyond the current horizon.
https://karpathy.ai/lexicap/0011-large.html#00:34:57.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Just shifting the horizon of knowledge a little
https://karpathy.ai/lexicap/0011-large.html#00:35:00.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
bit out there, breaking the existing rules such
https://karpathy.ai/lexicap/0011-large.html#00:35:04.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
that the new thing becomes solvable but wasn't
https://karpathy.ai/lexicap/0011-large.html#00:35:08.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
solvable by the old thing.
https://karpathy.ai/lexicap/0011-large.html#00:35:11.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So like adding a new axiom like what Gödel did
https://karpathy.ai/lexicap/0011-large.html#00:35:13.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
when he came up with these new sentences, new
https://karpathy.ai/lexicap/0011-large.html#00:35:17.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
theorems that didn't have a proof in the formal
https://karpathy.ai/lexicap/0011-large.html#00:35:21.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
system which means you can add them to the
https://karpathy.ai/lexicap/0011-large.html#00:35:23.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
repertoire hoping that they are not going to
https://karpathy.ai/lexicap/0011-large.html#00:35:25.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
damage the consistency of the whole thing.
https://karpathy.ai/lexicap/0011-large.html#00:35:31.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So in the paper with the amazing title,
https://karpathy.ai/lexicap/0011-large.html#00:35:35.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Formal Theory of Creativity, Fun and Intrinsic
https://karpathy.ai/lexicap/0011-large.html#00:35:39.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Motivation, you talk about discovery as intrinsic
https://karpathy.ai/lexicap/0011-large.html#00:35:43.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
reward, so if you view humans as intelligent
https://karpathy.ai/lexicap/0011-large.html#00:35:46.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
agents, what do you think is the purpose and
https://karpathy.ai/lexicap/0011-large.html#00:35:50.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
meaning of life for us humans?
https://karpathy.ai/lexicap/0011-large.html#00:35:53.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
You've talked about this discovery, do you see
https://karpathy.ai/lexicap/0011-large.html#00:35:56.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
humans as an instance of power play, agents?
https://karpathy.ai/lexicap/0011-large.html#00:35:59.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Humans are curious and that means they behave
https://karpathy.ai/lexicap/0011-large.html#00:36:04.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
like scientists, not only the official scientists
https://karpathy.ai/lexicap/0011-large.html#00:36:10.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
but even the babies behave like scientists and
https://karpathy.ai/lexicap/0011-large.html#00:36:13.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
they play around with their toys to figure out
https://karpathy.ai/lexicap/0011-large.html#00:36:16.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
how the world works and how it is responding to
https://karpathy.ai/lexicap/0011-large.html#00:36:19.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
their actions and that's how they learn about
https://karpathy.ai/lexicap/0011-large.html#00:36:22.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
gravity and everything.
https://karpathy.ai/lexicap/0011-large.html#00:36:25.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
In 1990 we had the first systems like that which
https://karpathy.ai/lexicap/0011-large.html#00:36:27.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
would just try to play around with the environment
https://karpathy.ai/lexicap/0011-large.html#00:36:31.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
and come up with situations that go beyond what
https://karpathy.ai/lexicap/0011-large.html#00:36:34.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
they knew at that time and then get a reward for
https://karpathy.ai/lexicap/0011-large.html#00:36:38.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
creating these situations and then becoming more
https://karpathy.ai/lexicap/0011-large.html#00:36:41.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
general problem solvers and being able to understand
https://karpathy.ai/lexicap/0011-large.html#00:36:44.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
more of the world.
https://karpathy.ai/lexicap/0011-large.html#00:36:47.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
I think in principle that curiosity strategy or
https://karpathy.ai/lexicap/0011-large.html#00:36:49.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
more sophisticated versions of what I just
https://karpathy.ai/lexicap/0011-large.html#00:37:01.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
described, they are what we have built in as well
https://karpathy.ai/lexicap/0011-large.html#00:37:03.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
because evolution discovered that's a good way of
https://karpathy.ai/lexicap/0011-large.html#00:37:07.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
exploring the unknown world and a guy who explores
https://karpathy.ai/lexicap/0011-large.html#00:37:10.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
the unknown world has a higher chance of solving
https://karpathy.ai/lexicap/0011-large.html#00:37:13.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
the mystery that he needs to survive in this world.
https://karpathy.ai/lexicap/0011-large.html#00:37:17.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
On the other hand, those guys who were too curious
https://karpathy.ai/lexicap/0011-large.html#00:37:20.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
they were weeded out as well so you have to find
https://karpathy.ai/lexicap/0011-large.html#00:37:24.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
this trade off.
https://karpathy.ai/lexicap/0011-large.html#00:37:27.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Evolution found a certain trade off.
https://karpathy.ai/lexicap/0011-large.html#00:37:28.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Apparently in our society there is a certain
https://karpathy.ai/lexicap/0011-large.html#00:37:30.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
percentage of extremely explorative guys and it
https://karpathy.ai/lexicap/0011-large.html#00:37:33.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
doesn't matter if they die because many of the
https://karpathy.ai/lexicap/0011-large.html#00:37:37.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
others are more conservative.
https://karpathy.ai/lexicap/0011-large.html#00:37:40.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
It would be surprising to me if that principle of
https://karpathy.ai/lexicap/0011-large.html#00:37:45.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
artificial curiosity wouldn't be present in almost
https://karpathy.ai/lexicap/0011-large.html#00:37:51.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
exactly the same form here.
https://karpathy.ai/lexicap/0011-large.html#00:37:56.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
In our brains.
https://karpathy.ai/lexicap/0011-large.html#00:37:58.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
You are a bit of a musician and an artist.
https://karpathy.ai/lexicap/0011-large.html#00:38:00.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Continuing on this topic of creativity, what do you
https://karpathy.ai/lexicap/0011-large.html#00:38:03.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
think is the role of creativity and intelligence?
https://karpathy.ai/lexicap/0011-large.html#00:38:08.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So you've kind of implied that it's essential for
https://karpathy.ai/lexicap/0011-large.html#00:38:10.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
intelligence if you think of intelligence as a
https://karpathy.ai/lexicap/0011-large.html#00:38:15.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
problem solving system, as ability to solve problems.
https://karpathy.ai/lexicap/0011-large.html#00:38:18.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
But do you think it's essential, this idea of
https://karpathy.ai/lexicap/0011-large.html#00:38:23.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
creativity?
https://karpathy.ai/lexicap/0011-large.html#00:38:26.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
We never have a program, a sub program that is
https://karpathy.ai/lexicap/0011-large.html#00:38:27.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
called creativity or something.
https://karpathy.ai/lexicap/0011-large.html#00:38:32.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
It's just a side effect of what our problem solvers
https://karpathy.ai/lexicap/0011-large.html#00:38:34.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
do. They are searching a space of problems, a space
https://karpathy.ai/lexicap/0011-large.html#00:38:37.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
of candidates, of solution candidates until they
https://karpathy.ai/lexicap/0011-large.html#00:38:41.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
hopefully find a solution to a given problem.
https://karpathy.ai/lexicap/0011-large.html#00:38:45.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
But then there are these two types of creativity
https://karpathy.ai/lexicap/0011-large.html#00:38:48.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
and both of them are now present in our machines.
https://karpathy.ai/lexicap/0011-large.html#00:38:50.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
The first one has been around for a long time,
https://karpathy.ai/lexicap/0011-large.html#00:38:54.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
which is human gives problem to machine, machine
https://karpathy.ai/lexicap/0011-large.html#00:38:56.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
tries to find a solution to that.
https://karpathy.ai/lexicap/0011-large.html#00:39:00.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
And this has been happening for many decades and
https://karpathy.ai/lexicap/0011-large.html#00:39:03.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
for many decades machines have found creative
https://karpathy.ai/lexicap/0011-large.html#00:39:06.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
solutions to interesting problems where humans were
https://karpathy.ai/lexicap/0011-large.html#00:39:09.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
not aware of these particularly creative solutions
https://karpathy.ai/lexicap/0011-large.html#00:39:12.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
but then appreciated that the machine found that.
https://karpathy.ai/lexicap/0011-large.html#00:39:17.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
The second is the pure creativity.
https://karpathy.ai/lexicap/0011-large.html#00:39:20.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
That I would call, what I just mentioned, I would
https://karpathy.ai/lexicap/0011-large.html#00:39:23.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
call the applied creativity, like applied art where
https://karpathy.ai/lexicap/0011-large.html#00:39:26.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
somebody tells you now make a nice picture of this
https://karpathy.ai/lexicap/0011-large.html#00:39:30.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Pope and you will get money for that.
https://karpathy.ai/lexicap/0011-large.html#00:39:34.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So here is the artist and he makes a convincing
https://karpathy.ai/lexicap/0011-large.html#00:39:37.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
picture of the Pope and the Pope likes it and gives
https://karpathy.ai/lexicap/0011-large.html#00:39:40.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
him the money.
https://karpathy.ai/lexicap/0011-large.html#00:39:43.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
And then there is the pure creativity which is
https://karpathy.ai/lexicap/0011-large.html#00:39:46.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
more like the power play and the artificial
https://karpathy.ai/lexicap/0011-large.html#00:39:49.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
curiosity thing where you have the freedom to
https://karpathy.ai/lexicap/0011-large.html#00:39:51.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
select your own problem.
https://karpathy.ai/lexicap/0011-large.html#00:39:54.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
Like a scientist who defines his own question
https://karpathy.ai/lexicap/0011-large.html#00:39:57.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
to study and so that is the pure creativity if you
https://karpathy.ai/lexicap/0011-large.html#00:40:02.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
will as opposed to the applied creativity which
https://karpathy.ai/lexicap/0011-large.html#00:40:06.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
serves another.
https://karpathy.ai/lexicap/0011-large.html#00:40:11.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
And in that distinction there is almost echoes of
https://karpathy.ai/lexicap/0011-large.html#00:40:14.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
narrow AI versus general AI.
https://karpathy.ai/lexicap/0011-large.html#00:40:16.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
So this kind of constrained painting of a Pope
https://karpathy.ai/lexicap/0011-large.html#00:40:19.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
seems like the approaches of what people are
https://karpathy.ai/lexicap/0011-large.html#00:40:22.040
Juergen Schmidhuber: Godel Machines, Meta-Learning, and LSTMs | Lex Fridman Podcast #11
calling narrow AI and pure creativity seems to be,
https://karpathy.ai/lexicap/0011-large.html#00:40:28.040