episode
stringlengths
45
100
text
stringlengths
1
528
timestamp_link
stringlengths
56
56
Daniel Negreanu: Poker | Lex Fridman Podcast #324
Where artificial intelligence created solvers
https://karpathy.ai/lexicap/0324-large.html#00:59:04.580
Daniel Negreanu: Poker | Lex Fridman Podcast #324
that will specifically say, okay, this is the optimal play.
https://karpathy.ai/lexicap/0324-large.html#00:59:07.500
Daniel Negreanu: Poker | Lex Fridman Podcast #324
This is the game theory optimal play.
https://karpathy.ai/lexicap/0324-large.html#00:59:11.020
Daniel Negreanu: Poker | Lex Fridman Podcast #324
So now it introduced poker to a whole new group
https://karpathy.ai/lexicap/0324-large.html#00:59:12.780
Daniel Negreanu: Poker | Lex Fridman Podcast #324
of like personality types.
https://karpathy.ai/lexicap/0324-large.html#00:59:16.180
Daniel Negreanu: Poker | Lex Fridman Podcast #324
In my day, it was people that were dregs of society
https://karpathy.ai/lexicap/0324-large.html#00:59:18.500
Daniel Negreanu: Poker | Lex Fridman Podcast #324
that didn't fit in, not college goers with a degree.
https://karpathy.ai/lexicap/0324-large.html#00:59:21.440
Daniel Negreanu: Poker | Lex Fridman Podcast #324
These are people who were street hustlers playing pool.
https://karpathy.ai/lexicap/0324-large.html#00:59:24.220
Daniel Negreanu: Poker | Lex Fridman Podcast #324
They found poker and they had these unique lives, right?
https://karpathy.ai/lexicap/0324-large.html#00:59:26.540
Daniel Negreanu: Poker | Lex Fridman Podcast #324
But now because poker can be studied,
https://karpathy.ai/lexicap/0324-large.html#00:59:29.640
Daniel Negreanu: Poker | Lex Fridman Podcast #324
much like you study university or college,
https://karpathy.ai/lexicap/0324-large.html#00:59:32.740
Daniel Negreanu: Poker | Lex Fridman Podcast #324
you had, for example, the German contingent
https://karpathy.ai/lexicap/0324-large.html#00:59:35.300
Daniel Negreanu: Poker | Lex Fridman Podcast #324
who was literally analyzing data
https://karpathy.ai/lexicap/0324-large.html#00:59:37.440
Daniel Negreanu: Poker | Lex Fridman Podcast #324
and coming up with strategies based on this.
https://karpathy.ai/lexicap/0324-large.html#00:59:39.300
Daniel Negreanu: Poker | Lex Fridman Podcast #324
And it's like, what?
https://karpathy.ai/lexicap/0324-large.html#00:59:41.300
Daniel Negreanu: Poker | Lex Fridman Podcast #324
And the old guy, got to play by feel or whatever.
https://karpathy.ai/lexicap/0324-large.html#00:59:42.620
Daniel Negreanu: Poker | Lex Fridman Podcast #324
And they're like, they're learning.
https://karpathy.ai/lexicap/0324-large.html#00:59:45.780
Daniel Negreanu: Poker | Lex Fridman Podcast #324
So I guess the way that you describe it is like
https://karpathy.ai/lexicap/0324-large.html#00:59:47.340
Daniel Negreanu: Poker | Lex Fridman Podcast #324
in the old days, it required skill and talent,
https://karpathy.ai/lexicap/0324-large.html#00:59:49.460
Daniel Negreanu: Poker | Lex Fridman Podcast #324
a card sense, right?
https://karpathy.ai/lexicap/0324-large.html#00:59:52.140
Daniel Negreanu: Poker | Lex Fridman Podcast #324
That was the only way to become good.
https://karpathy.ai/lexicap/0324-large.html#00:59:53.860
Daniel Negreanu: Poker | Lex Fridman Podcast #324
And today that's not the case.
https://karpathy.ai/lexicap/0324-large.html#00:59:55.100
Daniel Negreanu: Poker | Lex Fridman Podcast #324
Good study habits, a good work ethic in that regard
https://karpathy.ai/lexicap/0324-large.html#00:59:56.960
Daniel Negreanu: Poker | Lex Fridman Podcast #324
can make you like a really good player.
https://karpathy.ai/lexicap/0324-large.html#00:59:59.540
Daniel Negreanu: Poker | Lex Fridman Podcast #324
Even if you aren't all that talented or gifted,
https://karpathy.ai/lexicap/0324-large.html#01:00:01.220
Daniel Negreanu: Poker | Lex Fridman Podcast #324
having a good work ethic is a talent, right?
https://karpathy.ai/lexicap/0324-large.html#01:00:05.020
Daniel Negreanu: Poker | Lex Fridman Podcast #324
Not necessarily card sense, but if you're able to put
https://karpathy.ai/lexicap/0324-large.html#01:00:07.460
Daniel Negreanu: Poker | Lex Fridman Podcast #324
in the work and study from these solvers,
https://karpathy.ai/lexicap/0324-large.html#01:00:09.860
Daniel Negreanu: Poker | Lex Fridman Podcast #324
you essentially have the perfect study tool now
https://karpathy.ai/lexicap/0324-large.html#01:00:12.180
Daniel Negreanu: Poker | Lex Fridman Podcast #324
that we didn't have in my day.
https://karpathy.ai/lexicap/0324-large.html#01:00:15.420
Daniel Negreanu: Poker | Lex Fridman Podcast #324
So what do the solvers give you?
https://karpathy.ai/lexicap/0324-large.html#01:00:17.160
Daniel Negreanu: Poker | Lex Fridman Podcast #324
Do you start to memorize the optimal play
https://karpathy.ai/lexicap/0324-large.html#01:00:19.900
Daniel Negreanu: Poker | Lex Fridman Podcast #324
for every single hand?
https://karpathy.ai/lexicap/0324-large.html#01:00:22.120
Daniel Negreanu: Poker | Lex Fridman Podcast #324
You try your best.
https://karpathy.ai/lexicap/0324-large.html#01:00:24.120
Daniel Negreanu: Poker | Lex Fridman Podcast #324
So again, solvers are imperfect as well,
https://karpathy.ai/lexicap/0324-large.html#01:00:25.300
Daniel Negreanu: Poker | Lex Fridman Podcast #324
in terms of the way the humans utilize them, right?
https://karpathy.ai/lexicap/0324-large.html#01:00:28.720
Daniel Negreanu: Poker | Lex Fridman Podcast #324
Because you can give solvers a certain number of inputs
https://karpathy.ai/lexicap/0324-large.html#01:00:32.700
Daniel Negreanu: Poker | Lex Fridman Podcast #324
in terms of what you want it to solve,
https://karpathy.ai/lexicap/0324-large.html#01:00:36.860
Daniel Negreanu: Poker | Lex Fridman Podcast #324
but a solver can think on many, many levels.
https://karpathy.ai/lexicap/0324-large.html#01:00:38.500
Daniel Negreanu: Poker | Lex Fridman Podcast #324
So for example, the way that a typical player
https://karpathy.ai/lexicap/0324-large.html#01:00:40.900
Daniel Negreanu: Poker | Lex Fridman Podcast #324
would do a solve is to say, okay,
https://karpathy.ai/lexicap/0324-large.html#01:00:43.240
Daniel Negreanu: Poker | Lex Fridman Podcast #324
what does a solver think is the best play here?
https://karpathy.ai/lexicap/0324-large.html#01:00:45.580
Daniel Negreanu: Poker | Lex Fridman Podcast #324
Bet one third pot, bet two thirds pot,
https://karpathy.ai/lexicap/0324-large.html#01:00:47.860
Daniel Negreanu: Poker | Lex Fridman Podcast #324
or bet one and a half times pot, okay?
https://karpathy.ai/lexicap/0324-large.html#01:00:50.700
Daniel Negreanu: Poker | Lex Fridman Podcast #324
You give it three parameters, it comes out with an output
https://karpathy.ai/lexicap/0324-large.html#01:00:52.580
Daniel Negreanu: Poker | Lex Fridman Podcast #324
and it tells you what you should do
https://karpathy.ai/lexicap/0324-large.html#01:00:55.820
Daniel Negreanu: Poker | Lex Fridman Podcast #324
with all the different hands you have.
https://karpathy.ai/lexicap/0324-large.html#01:00:57.420
Daniel Negreanu: Poker | Lex Fridman Podcast #324
However, that's a simplified version
https://karpathy.ai/lexicap/0324-large.html#01:00:59.320
Daniel Negreanu: Poker | Lex Fridman Podcast #324
of what a solver would really do,
https://karpathy.ai/lexicap/0324-large.html#01:01:01.820
Daniel Negreanu: Poker | Lex Fridman Podcast #324
because a solver might decide
https://karpathy.ai/lexicap/0324-large.html#01:01:02.940
Daniel Negreanu: Poker | Lex Fridman Podcast #324
that seven times the pot is best, 10% of the pot.
https://karpathy.ai/lexicap/0324-large.html#01:01:04.580
Daniel Negreanu: Poker | Lex Fridman Podcast #324
But when you're putting in a solve,
https://karpathy.ai/lexicap/0324-large.html#01:01:08.100
Daniel Negreanu: Poker | Lex Fridman Podcast #324
you can only put in specific parameters.
https://karpathy.ai/lexicap/0324-large.html#01:01:09.860
Daniel Negreanu: Poker | Lex Fridman Podcast #324
So that's why, frankly, that's typically the number,
https://karpathy.ai/lexicap/0324-large.html#01:01:12.540
Daniel Negreanu: Poker | Lex Fridman Podcast #324
one third, two third, and one and a half times pot
https://karpathy.ai/lexicap/0324-large.html#01:01:14.900
Daniel Negreanu: Poker | Lex Fridman Podcast #324
is what people often do.
https://karpathy.ai/lexicap/0324-large.html#01:01:17.140
Daniel Negreanu: Poker | Lex Fridman Podcast #324
So they sort of have a vague idea of what a solver wants.
https://karpathy.ai/lexicap/0324-large.html#01:01:18.500
Daniel Negreanu: Poker | Lex Fridman Podcast #324
But again, imperfect in terms of the implementation of it.
https://karpathy.ai/lexicap/0324-large.html#01:01:21.860
Daniel Negreanu: Poker | Lex Fridman Podcast #324
And memorizing all the variables,
https://karpathy.ai/lexicap/0324-large.html#01:01:26.660
Daniel Negreanu: Poker | Lex Fridman Podcast #324
like that King Jack offsuit with the King of Diamonds
https://karpathy.ai/lexicap/0324-large.html#01:01:29.060
Daniel Negreanu: Poker | Lex Fridman Podcast #324
is 13%, no human brain can do that.
https://karpathy.ai/lexicap/0324-large.html#01:01:31.780
Daniel Negreanu: Poker | Lex Fridman Podcast #324
So what you do is you bucket it.
https://karpathy.ai/lexicap/0324-large.html#01:01:35.800
Daniel Negreanu: Poker | Lex Fridman Podcast #324
Like you bucket it into say, instead of 10,000 variables,
https://karpathy.ai/lexicap/0324-large.html#01:01:37.500
Daniel Negreanu: Poker | Lex Fridman Podcast #324
you have 10 buckets and you say, okay,
https://karpathy.ai/lexicap/0324-large.html#01:01:41.000
Daniel Negreanu: Poker | Lex Fridman Podcast #324
with these hands, we do roughly this,
https://karpathy.ai/lexicap/0324-large.html#01:01:42.580
Daniel Negreanu: Poker | Lex Fridman Podcast #324
and we do roughly this.
https://karpathy.ai/lexicap/0324-large.html#01:01:44.100
Daniel Negreanu: Poker | Lex Fridman Podcast #324
And you try your best to stay within those lines.
https://karpathy.ai/lexicap/0324-large.html#01:01:45.140
Daniel Negreanu: Poker | Lex Fridman Podcast #324
But again, what I love about live poker partly
https://karpathy.ai/lexicap/0324-large.html#01:01:48.260
Daniel Negreanu: Poker | Lex Fridman Podcast #324
is that nobody will ever be able to master game theory,
https://karpathy.ai/lexicap/0324-large.html#01:01:50.460
Daniel Negreanu: Poker | Lex Fridman Podcast #324
and mimic a solver.
https://karpathy.ai/lexicap/0324-large.html#01:01:54.560
Daniel Negreanu: Poker | Lex Fridman Podcast #324
But you also have to incorporate your position,
https://karpathy.ai/lexicap/0324-large.html#01:01:57.800
Daniel Negreanu: Poker | Lex Fridman Podcast #324
where you are, and obviously what cards you have,
https://karpathy.ai/lexicap/0324-large.html#01:02:03.080
Daniel Negreanu: Poker | Lex Fridman Podcast #324
but also the size of your stack,
https://karpathy.ai/lexicap/0324-large.html#01:02:06.460
Daniel Negreanu: Poker | Lex Fridman Podcast #324
how much money you have,
https://karpathy.ai/lexicap/0324-large.html#01:02:08.920
Daniel Negreanu: Poker | Lex Fridman Podcast #324
and also whether you have the ability
https://karpathy.ai/lexicap/0324-large.html#01:02:09.920
Daniel Negreanu: Poker | Lex Fridman Podcast #324
or desire to buy in, all those kinds of things.
https://karpathy.ai/lexicap/0324-large.html#01:02:12.440
Daniel Negreanu: Poker | Lex Fridman Podcast #324
So you have to calculate all of that, right?
https://karpathy.ai/lexicap/0324-large.html#01:02:15.460
Daniel Negreanu: Poker | Lex Fridman Podcast #324
So the solver will do that, right?
https://karpathy.ai/lexicap/0324-large.html#01:02:17.600
Daniel Negreanu: Poker | Lex Fridman Podcast #324
And essentially, you don't input your hand.
https://karpathy.ai/lexicap/0324-large.html#01:02:20.440
Daniel Negreanu: Poker | Lex Fridman Podcast #324
It tells you, you'll look at the grade,
https://karpathy.ai/lexicap/0324-large.html#01:02:23.120
Daniel Negreanu: Poker | Lex Fridman Podcast #324
and be like, all right, this is my hand,
https://karpathy.ai/lexicap/0324-large.html#01:02:25.600
Daniel Negreanu: Poker | Lex Fridman Podcast #324
and it tells you what it is.
https://karpathy.ai/lexicap/0324-large.html#01:02:27.320
Daniel Negreanu: Poker | Lex Fridman Podcast #324
But it tells you what you would do with any hand, right?
https://karpathy.ai/lexicap/0324-large.html#01:02:28.160
Daniel Negreanu: Poker | Lex Fridman Podcast #324
It gives you the full output.
https://karpathy.ai/lexicap/0324-large.html#01:02:30.960
Daniel Negreanu: Poker | Lex Fridman Podcast #324
And that actually gives you a better idea,
https://karpathy.ai/lexicap/0324-large.html#01:02:32.280
Daniel Negreanu: Poker | Lex Fridman Podcast #324
because you're ultimately, like you said,
https://karpathy.ai/lexicap/0324-large.html#01:02:34.040
Daniel Negreanu: Poker | Lex Fridman Podcast #324
playing a range of hands, not a hand.
https://karpathy.ai/lexicap/0324-large.html#01:02:36.800
Daniel Negreanu: Poker | Lex Fridman Podcast #324
And the solvers do things that are really interesting.
https://karpathy.ai/lexicap/0324-large.html#01:02:39.400
Daniel Negreanu: Poker | Lex Fridman Podcast #324
You've seen AlphaGo, I would imagine.
https://karpathy.ai/lexicap/0324-large.html#01:02:41.240
Daniel Negreanu: Poker | Lex Fridman Podcast #324
Brilliant film, right, I thought.
https://karpathy.ai/lexicap/0324-large.html#01:02:43.040
Daniel Negreanu: Poker | Lex Fridman Podcast #324
And I thought what was interesting is there was,
https://karpathy.ai/lexicap/0324-large.html#01:02:44.680
Daniel Negreanu: Poker | Lex Fridman Podcast #324
you know, accepted theory from all the top Go players,
https://karpathy.ai/lexicap/0324-large.html#01:02:46.920
Daniel Negreanu: Poker | Lex Fridman Podcast #324
this is what you do.
https://karpathy.ai/lexicap/0324-large.html#01:02:50.280
Daniel Negreanu: Poker | Lex Fridman Podcast #324
But the AI was doing things way different,
https://karpathy.ai/lexicap/0324-large.html#01:02:51.160
Daniel Negreanu: Poker | Lex Fridman Podcast #324
and they're like, this has to be wrong,
https://karpathy.ai/lexicap/0324-large.html#01:02:54.360
Daniel Negreanu: Poker | Lex Fridman Podcast #324
but really it wasn't.
https://karpathy.ai/lexicap/0324-large.html#01:02:55.720
Daniel Negreanu: Poker | Lex Fridman Podcast #324
So for example, a solver may say this, right?
https://karpathy.ai/lexicap/0324-large.html#01:02:56.880
Daniel Negreanu: Poker | Lex Fridman Podcast #324
Let's say you bet on the end, and you bet a lot,
https://karpathy.ai/lexicap/0324-large.html#01:03:00.760
Daniel Negreanu: Poker | Lex Fridman Podcast #324
and a solver may say, you should fold here
https://karpathy.ai/lexicap/0324-large.html#01:03:04.360
Daniel Negreanu: Poker | Lex Fridman Podcast #324
with a pair of kings and a queen kicker,
https://karpathy.ai/lexicap/0324-large.html#01:03:07.460