episode
stringlengths
45
100
text
stringlengths
1
528
timestamp_link
stringlengths
56
56
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
for game solving and applying them into the real world.
https://karpathy.ai/lexicap/0012-large.html#01:01:26.360
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
I'm still very interested in market design as well.
https://karpathy.ai/lexicap/0012-large.html#01:01:30.440
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
And we're doing that in the optimized markets,
https://karpathy.ai/lexicap/0012-large.html#01:01:33.160
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
but I'm most interested if number one right now
https://karpathy.ai/lexicap/0012-large.html#01:01:35.400
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
is strategic machine strategy robot,
https://karpathy.ai/lexicap/0012-large.html#01:01:37.560
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
getting that technology out there
https://karpathy.ai/lexicap/0012-large.html#01:01:40.000
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
and seeing as you were in the trenches doing applications,
https://karpathy.ai/lexicap/0012-large.html#01:01:41.440
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
what needs to be actually filled,
https://karpathy.ai/lexicap/0012-large.html#01:01:45.560
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
what technology gaps still need to be filled.
https://karpathy.ai/lexicap/0012-large.html#01:01:47.120
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So it's so hard to just put your feet on the table
https://karpathy.ai/lexicap/0012-large.html#01:01:49.800
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
and imagine what needs to be done.
https://karpathy.ai/lexicap/0012-large.html#01:01:52.040
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
But when you're actually doing real applications,
https://karpathy.ai/lexicap/0012-large.html#01:01:53.800
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
the applications tell you what needs to be done.
https://karpathy.ai/lexicap/0012-large.html#01:01:56.280
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
And I really enjoy that interaction.
https://karpathy.ai/lexicap/0012-large.html#01:01:59.120
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Is it a challenging process to apply
https://karpathy.ai/lexicap/0012-large.html#01:02:00.840
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
some of the state of the art techniques you're working on
https://karpathy.ai/lexicap/0012-large.html#01:02:04.480
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
and having the various players in industry
https://karpathy.ai/lexicap/0012-large.html#01:02:07.760
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
or the military or people who could really benefit from it
https://karpathy.ai/lexicap/0012-large.html#01:02:14.080
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
actually use it?
https://karpathy.ai/lexicap/0012-large.html#01:02:17.720
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
What's that process like of,
https://karpathy.ai/lexicap/0012-large.html#01:02:19.040
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
autonomous vehicles work with automotive companies
https://karpathy.ai/lexicap/0012-large.html#01:02:21.400
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
and they're in many ways are a little bit old fashioned.
https://karpathy.ai/lexicap/0012-large.html#01:02:23.680
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
It's difficult.
https://karpathy.ai/lexicap/0012-large.html#01:02:28.200
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
They really want to use this technology.
https://karpathy.ai/lexicap/0012-large.html#01:02:29.240
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
There's clearly will have a significant benefit,
https://karpathy.ai/lexicap/0012-large.html#01:02:31.840
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
but the systems aren't quite in place
https://karpathy.ai/lexicap/0012-large.html#01:02:34.640
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
to easily have them integrated in terms of data,
https://karpathy.ai/lexicap/0012-large.html#01:02:37.480
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
in terms of compute, in terms of all these kinds of things.
https://karpathy.ai/lexicap/0012-large.html#01:02:41.080
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So is that one of the bigger challenges that you're facing
https://karpathy.ai/lexicap/0012-large.html#01:02:43.760
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
and how do you tackle that challenge?
https://karpathy.ai/lexicap/0012-large.html#01:02:48.680
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Yeah, I think that's always a challenge.
https://karpathy.ai/lexicap/0012-large.html#01:02:50.000
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
That's kind of slowness and inertia really
https://karpathy.ai/lexicap/0012-large.html#01:02:52.360
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
of let's do things the way we've always done it.
https://karpathy.ai/lexicap/0012-large.html#01:02:55.560
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
You just have to find the internal champions
https://karpathy.ai/lexicap/0012-large.html#01:02:57.920
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
at the customer who understand that,
https://karpathy.ai/lexicap/0012-large.html#01:03:00.120
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
hey, things can't be the same way in the future.
https://karpathy.ai/lexicap/0012-large.html#01:03:02.120
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Otherwise bad things are going to happen.
https://karpathy.ai/lexicap/0012-large.html#01:03:04.680
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
And it's in autonomous vehicles.
https://karpathy.ai/lexicap/0012-large.html#01:03:06.960
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
It's actually very interesting
https://karpathy.ai/lexicap/0012-large.html#01:03:08.600
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
that the car makers are doing that
https://karpathy.ai/lexicap/0012-large.html#01:03:09.680
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
and they're very traditional,
https://karpathy.ai/lexicap/0012-large.html#01:03:11.120
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
but at the same time you have tech companies
https://karpathy.ai/lexicap/0012-large.html#01:03:12.440
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
who have nothing to do with cars or transportation
https://karpathy.ai/lexicap/0012-large.html#01:03:14.360
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
like Google and Baidu really pushing on autonomous cars.
https://karpathy.ai/lexicap/0012-large.html#01:03:17.120
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
I find that fascinating.
https://karpathy.ai/lexicap/0012-large.html#01:03:21.880
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Clearly you're super excited
https://karpathy.ai/lexicap/0012-large.html#01:03:23.240
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
about actually these ideas having an impact in the world.
https://karpathy.ai/lexicap/0012-large.html#01:03:25.160
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
In terms of the technology, in terms of ideas and research,
https://karpathy.ai/lexicap/0012-large.html#01:03:29.320
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
are there directions that you're also excited about?
https://karpathy.ai/lexicap/0012-large.html#01:03:32.680
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Whether that's on some of the approaches you talked about
https://karpathy.ai/lexicap/0012-large.html#01:03:36.600
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
for the imperfect information games,
https://karpathy.ai/lexicap/0012-large.html#01:03:40.840
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
whether it's applying deep learning
https://karpathy.ai/lexicap/0012-large.html#01:03:42.760
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
to some of these problems,
https://karpathy.ai/lexicap/0012-large.html#01:03:44.000
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
is there something that you're excited
https://karpathy.ai/lexicap/0012-large.html#01:03:45.120
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
in the research side of things?
https://karpathy.ai/lexicap/0012-large.html#01:03:46.520
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Yeah, yeah, lots of different things
https://karpathy.ai/lexicap/0012-large.html#01:03:48.840
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
in the game solving.
https://karpathy.ai/lexicap/0012-large.html#01:03:51.120
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So solving even bigger games,
https://karpathy.ai/lexicap/0012-large.html#01:03:53.240
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
games where you have more hidden action
https://karpathy.ai/lexicap/0012-large.html#01:03:56.400
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
of the player actions as well.
https://karpathy.ai/lexicap/0012-large.html#01:03:59.760
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Poker is a game where really the chance actions are hidden
https://karpathy.ai/lexicap/0012-large.html#01:04:02.040
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
or some of them are hidden,
https://karpathy.ai/lexicap/0012-large.html#01:04:05.880
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
but the player actions are public.
https://karpathy.ai/lexicap/0012-large.html#01:04:07.080
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Multiplayer games of various sorts,
https://karpathy.ai/lexicap/0012-large.html#01:04:11.440
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
collusion, opponent exploitation,
https://karpathy.ai/lexicap/0012-large.html#01:04:14.000
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
all and even longer games.
https://karpathy.ai/lexicap/0012-large.html#01:04:18.080
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So games that basically go forever,
https://karpathy.ai/lexicap/0012-large.html#01:04:21.280
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
but they're not repeated.
https://karpathy.ai/lexicap/0012-large.html#01:04:23.160
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So see extensive fun games that go forever.
https://karpathy.ai/lexicap/0012-large.html#01:04:24.680
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
What would that even look like?
https://karpathy.ai/lexicap/0012-large.html#01:04:27.880
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
How do you represent that?
https://karpathy.ai/lexicap/0012-large.html#01:04:30.080
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
How do you solve that?
https://karpathy.ai/lexicap/0012-large.html#01:04:31.040
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
What's an example of a game like that?
https://karpathy.ai/lexicap/0012-large.html#01:04:32.040
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Or is this some of the stochastic games
https://karpathy.ai/lexicap/0012-large.html#01:04:33.440
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
that you mentioned?
https://karpathy.ai/lexicap/0012-large.html#01:04:35.600
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Let's say business strategy.
https://karpathy.ai/lexicap/0012-large.html#01:04:36.440
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So it's not just modeling like a particular interaction,
https://karpathy.ai/lexicap/0012-large.html#01:04:37.320
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
but thinking about the business from here to eternity.
https://karpathy.ai/lexicap/0012-large.html#01:04:40.840
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Or let's say military strategy.
https://karpathy.ai/lexicap/0012-large.html#01:04:44.440
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So it's not like war is gonna go away.
https://karpathy.ai/lexicap/0012-large.html#01:04:49.040
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
How do you think about military strategy
https://karpathy.ai/lexicap/0012-large.html#01:04:51.000
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
that's gonna go forever?
https://karpathy.ai/lexicap/0012-large.html#01:04:54.280
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
How do you even model that?
https://karpathy.ai/lexicap/0012-large.html#01:04:56.680
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
How do you know whether a move was good
https://karpathy.ai/lexicap/0012-large.html#01:04:58.080
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
that somebody made and so on?
https://karpathy.ai/lexicap/0012-large.html#01:05:01.000
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So that's kind of one direction.
https://karpathy.ai/lexicap/0012-large.html#01:05:05.200
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
I'm also very interested in learning
https://karpathy.ai/lexicap/0012-large.html#01:05:06.960
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
much more scalable techniques for integer programming.
https://karpathy.ai/lexicap/0012-large.html#01:05:09.800
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So we had an ICML paper this summer on that.
https://karpathy.ai/lexicap/0012-large.html#01:05:13.440
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
The first automated algorithm configuration paper
https://karpathy.ai/lexicap/0012-large.html#01:05:16.560
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
that has theoretical generalization guarantees.
https://karpathy.ai/lexicap/0012-large.html#01:05:20.280
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So if I see this many training examples
https://karpathy.ai/lexicap/0012-large.html#01:05:23.560
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
and I told my algorithm in this way,
https://karpathy.ai/lexicap/0012-large.html#01:05:26.200
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
it's going to have good performance
https://karpathy.ai/lexicap/0012-large.html#01:05:28.560
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
on the real distribution, which I've not seen.
https://karpathy.ai/lexicap/0012-large.html#01:05:30.560
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So, which is kind of interesting
https://karpathy.ai/lexicap/0012-large.html#01:05:33.200
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
that algorithm configuration has been going on now
https://karpathy.ai/lexicap/0012-large.html#01:05:34.840
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
for at least 17 years seriously.
https://karpathy.ai/lexicap/0012-large.html#01:05:37.680
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
And there has not been any generalization theory before.
https://karpathy.ai/lexicap/0012-large.html#01:05:41.200
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Well, this is really exciting
https://karpathy.ai/lexicap/0012-large.html#01:05:45.960