episode
stringlengths
45
100
text
stringlengths
1
528
timestamp_link
stringlengths
56
56
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
but then they cease to look good in the large.
https://karpathy.ai/lexicap/0012-large.html#00:39:05.200
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
And we've also seen that there are a lot of techniques
https://karpathy.ai/lexicap/0012-large.html#00:39:07.120
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
that look superior in theory.
https://karpathy.ai/lexicap/0012-large.html#00:39:10.160
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
And I really mean in terms of convergence rates,
https://karpathy.ai/lexicap/0012-large.html#00:39:13.280
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
like first order methods, better convergence rates,
https://karpathy.ai/lexicap/0012-large.html#00:39:16.200
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
like the CFR based algorithms,
https://karpathy.ai/lexicap/0012-large.html#00:39:18.440
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
yet the CFR based algorithms are the fastest in practice.
https://karpathy.ai/lexicap/0012-large.html#00:39:20.880
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So it really tells me that you have to test this in reality.
https://karpathy.ai/lexicap/0012-large.html#00:39:24.880
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
The theory isn't tight enough, if you will,
https://karpathy.ai/lexicap/0012-large.html#00:39:28.240
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
to tell you which algorithms are better than the others.
https://karpathy.ai/lexicap/0012-large.html#00:39:30.880
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
And you have to look at these things in the large,
https://karpathy.ai/lexicap/0012-large.html#00:39:34.360
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
because any sort of projections you do from the small
https://karpathy.ai/lexicap/0012-large.html#00:39:38.600
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
can at least in this domain be very misleading.
https://karpathy.ai/lexicap/0012-large.html#00:39:41.480
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So that's kind of from a kind of a science
https://karpathy.ai/lexicap/0012-large.html#00:39:43.800
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
and engineering perspective, from a personal perspective,
https://karpathy.ai/lexicap/0012-large.html#00:39:46.240
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
it's been just a wild experience
https://karpathy.ai/lexicap/0012-large.html#00:39:49.120
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
in that with the first poker competition,
https://karpathy.ai/lexicap/0012-large.html#00:39:51.280
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
the first brains versus AI,
https://karpathy.ai/lexicap/0012-large.html#00:39:54.160
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
man machine poker competition that we organized.
https://karpathy.ai/lexicap/0012-large.html#00:39:57.200
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
There had been, by the way, for other poker games,
https://karpathy.ai/lexicap/0012-large.html#00:39:59.840
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
there had been previous competitions,
https://karpathy.ai/lexicap/0012-large.html#00:40:01.760
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
but this was for Heads Up No Limit, this was the first.
https://karpathy.ai/lexicap/0012-large.html#00:40:03.240
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
And I probably became the most hated person
https://karpathy.ai/lexicap/0012-large.html#00:40:06.360
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
in the world of poker.
https://karpathy.ai/lexicap/0012-large.html#00:40:09.560
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
And I didn't mean to, I just saw.
https://karpathy.ai/lexicap/0012-large.html#00:40:10.880
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Why is that?
https://karpathy.ai/lexicap/0012-large.html#00:40:12.880
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
For cracking the game, for something.
https://karpathy.ai/lexicap/0012-large.html#00:40:13.720
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Yeah, a lot of people felt that it was a real threat
https://karpathy.ai/lexicap/0012-large.html#00:40:15.840
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
to the whole game, the whole existence of the game.
https://karpathy.ai/lexicap/0012-large.html#00:40:20.000
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
If AI becomes better than humans,
https://karpathy.ai/lexicap/0012-large.html#00:40:22.760
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
people would be scared to play poker
https://karpathy.ai/lexicap/0012-large.html#00:40:26.080
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
because there are these superhuman AIs running around
https://karpathy.ai/lexicap/0012-large.html#00:40:28.520
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
taking their money and all of that.
https://karpathy.ai/lexicap/0012-large.html#00:40:30.680
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So I just, it's just really aggressive.
https://karpathy.ai/lexicap/0012-large.html#00:40:32.760
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
The comments were super aggressive.
https://karpathy.ai/lexicap/0012-large.html#00:40:36.200
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
I got everything just short of death threats.
https://karpathy.ai/lexicap/0012-large.html#00:40:37.880
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Do you think the same was true for chess?
https://karpathy.ai/lexicap/0012-large.html#00:40:40.920
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Because right now they just completed
https://karpathy.ai/lexicap/0012-large.html#00:40:44.000
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
the world championships in chess,
https://karpathy.ai/lexicap/0012-large.html#00:40:45.760
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
and humans just started ignoring the fact
https://karpathy.ai/lexicap/0012-large.html#00:40:47.720
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
that there's AI systems now that outperform humans
https://karpathy.ai/lexicap/0012-large.html#00:40:49.560
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
and they still enjoy the game, it's still a beautiful game.
https://karpathy.ai/lexicap/0012-large.html#00:40:52.920
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
That's what I think.
https://karpathy.ai/lexicap/0012-large.html#00:40:55.520
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
And I think the same thing happens in poker.
https://karpathy.ai/lexicap/0012-large.html#00:40:56.360
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
And so I didn't think of myself
https://karpathy.ai/lexicap/0012-large.html#00:40:58.800
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
as somebody who was gonna kill the game,
https://karpathy.ai/lexicap/0012-large.html#00:41:01.040
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
and I don't think I did.
https://karpathy.ai/lexicap/0012-large.html#00:41:02.360
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
I've really learned to love this game.
https://karpathy.ai/lexicap/0012-large.html#00:41:03.800
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
I wasn't a poker player before,
https://karpathy.ai/lexicap/0012-large.html#00:41:05.600
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
but learned so many nuances about it from these AIs,
https://karpathy.ai/lexicap/0012-large.html#00:41:06.960
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
and they've really changed how the game is played,
https://karpathy.ai/lexicap/0012-large.html#00:41:10.520
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
by the way.
https://karpathy.ai/lexicap/0012-large.html#00:41:12.480
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So they have these very Martian ways of playing poker,
https://karpathy.ai/lexicap/0012-large.html#00:41:13.320
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
and the top humans are now incorporating
https://karpathy.ai/lexicap/0012-large.html#00:41:16.240
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
those types of strategies into their own play.
https://karpathy.ai/lexicap/0012-large.html#00:41:18.400
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So if anything, to me, our work has made poker
https://karpathy.ai/lexicap/0012-large.html#00:41:21.400
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
a richer, more interesting game for humans to play,
https://karpathy.ai/lexicap/0012-large.html#00:41:26.560
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
not something that is gonna steer humans
https://karpathy.ai/lexicap/0012-large.html#00:41:29.800
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
away from it entirely.
https://karpathy.ai/lexicap/0012-large.html#00:41:32.160
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Just a quick comment on something you said,
https://karpathy.ai/lexicap/0012-large.html#00:41:34.200
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
which is, if I may say so,
https://karpathy.ai/lexicap/0012-large.html#00:41:35.960
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
in academia is a little bit rare sometimes.
https://karpathy.ai/lexicap/0012-large.html#00:41:39.400
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
It's pretty brave to put your ideas to the test
https://karpathy.ai/lexicap/0012-large.html#00:41:42.400
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
in the way you described,
https://karpathy.ai/lexicap/0012-large.html#00:41:45.520
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
saying that sometimes good ideas don't work
https://karpathy.ai/lexicap/0012-large.html#00:41:47.200
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
when you actually try to apply them at scale.
https://karpathy.ai/lexicap/0012-large.html#00:41:49.360
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So where does that come from?
https://karpathy.ai/lexicap/0012-large.html#00:41:52.760
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
I mean, if you could do advice for people,
https://karpathy.ai/lexicap/0012-large.html#00:41:54.200
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
what drives you in that sense?
https://karpathy.ai/lexicap/0012-large.html#00:41:58.880
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Were you always this way?
https://karpathy.ai/lexicap/0012-large.html#00:42:00.760
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
I mean, it takes a brave person.
https://karpathy.ai/lexicap/0012-large.html#00:42:02.360
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
I guess is what I'm saying, to test their ideas
https://karpathy.ai/lexicap/0012-large.html#00:42:04.080
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
and to see if this thing actually works
https://karpathy.ai/lexicap/0012-large.html#00:42:06.760
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
against human top human players and so on.
https://karpathy.ai/lexicap/0012-large.html#00:42:08.640
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Yeah, I don't know about brave,
https://karpathy.ai/lexicap/0012-large.html#00:42:11.680
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
but it takes a lot of work.
https://karpathy.ai/lexicap/0012-large.html#00:42:12.960
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
It takes a lot of work and a lot of time
https://karpathy.ai/lexicap/0012-large.html#00:42:15.000
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
to organize, to make something big
https://karpathy.ai/lexicap/0012-large.html#00:42:18.400
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
and to organize an event and stuff like that.
https://karpathy.ai/lexicap/0012-large.html#00:42:20.360
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
And what drives you in that effort?
https://karpathy.ai/lexicap/0012-large.html#00:42:22.920
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Because you could still, I would argue,
https://karpathy.ai/lexicap/0012-large.html#00:42:24.760
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
get a best paper award at NIPS as you did in 17
https://karpathy.ai/lexicap/0012-large.html#00:42:26.880
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
without doing this.
https://karpathy.ai/lexicap/0012-large.html#00:42:30.280
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
That's right, yes.
https://karpathy.ai/lexicap/0012-large.html#00:42:31.440
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
And so in general, I believe it's very important
https://karpathy.ai/lexicap/0012-large.html#00:42:32.960
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
to do things in the real world and at scale.
https://karpathy.ai/lexicap/0012-large.html#00:42:37.640
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
And that's really where the pudding, if you will,
https://karpathy.ai/lexicap/0012-large.html#00:42:41.480
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
proof is in the pudding, that's where it is.
https://karpathy.ai/lexicap/0012-large.html#00:42:46.160
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
In this particular case,
https://karpathy.ai/lexicap/0012-large.html#00:42:48.400
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
it was kind of a competition between different groups
https://karpathy.ai/lexicap/0012-large.html#00:42:50.080
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
and for many years as to who can be the first one
https://karpathy.ai/lexicap/0012-large.html#00:42:55.160
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
to beat the top humans at Heads Up No Limit, Texas Holdem.
https://karpathy.ai/lexicap/0012-large.html#00:42:59.080
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So it became kind of like a competition who can get there.
https://karpathy.ai/lexicap/0012-large.html#00:43:02.040
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Yeah, so a little friendly competition
https://karpathy.ai/lexicap/0012-large.html#00:43:09.560
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
could do wonders for progress.
https://karpathy.ai/lexicap/0012-large.html#00:43:11.800
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Yes, absolutely.
https://karpathy.ai/lexicap/0012-large.html#00:43:14.040
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So the topic of mechanism design,
https://karpathy.ai/lexicap/0012-large.html#00:43:16.400
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
which is really interesting, also kind of new to me,
https://karpathy.ai/lexicap/0012-large.html#00:43:19.040
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
except as an observer of, I don't know, politics and any,
https://karpathy.ai/lexicap/0012-large.html#00:43:22.280
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
I'm an observer of mechanisms,
https://karpathy.ai/lexicap/0012-large.html#00:43:25.680