Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss
What is the highest human win rate against top bots?
20 views
Skip to first unread message
mu...@compuplus.net
May 17, 2012, 4:38:42 AM5/17/12
to
Without being specific, I myself claimed over 50% and later backtracked trying to get bets based on bot calculated ratings. Others reported here 38%, 40%, etc...
Is there any records of the highest win rates by humans against the top 2-3 bots, in a "humanly long" series of long matches (i.e. 50-100 matches of 15-25 points)...?
And also, whether the highest human win rates against bots were achieved by trying to play like the bots, as advocated by world-class gamblers, by mediocre math phd's, by book peddlers, private tutors, etc. or by other strategies...?
MK
Is there any records of the highest win rates by humans against the top 2-3 bots, in a "humanly long" series of long matches (i.e. 50-100 matches of 15-25 points)...?
And also, whether the highest human win rates against bots were achieved by trying to play like the bots, as advocated by world-class gamblers, by mediocre math phd's, by book peddlers, private tutors, etc. or by other strategies...?
MK
Tim Chow
May 17, 2012, 8:32:40 PM5/17/12
to
On May 17, 4:38 am, mu...@compuplus.net wrote:
> And also, whether the highest human win rates against bots were achieved by trying to play like the bots
I don't have any hard data. However, from what I've heard from
> And also, whether the highest human win rates against bots were achieved by trying to play like the bots
others, the best way to achieve high win rates against the bots is as
follows.
1. Play money games if possible. If not, play long matches, as long
as possible, so that you can jack up the cube. Avoid short matches.
2. Don't play like the bot. Get into a massive backgame that the bot
doesn't understand, and jack up the cube.
The bot will outplay you in "normal" games, but the hope is that you
can make the games that you understand better count for a lot, by
using the cube appropriately.
For money games, this strategy should let you win consistently in the
long run. For long matches, I haven't heard anyone make quantitative
claims about what is possible.
---
Tim Chow
mu...@compuplus.net
May 18, 2012, 7:31:49 AM5/18/12
to
On Thursday, May 17, 2012 6:32:40 PM UTC-6, Tim Chow wrote:
> On May 17, 4:38 am, mu...@compuplus.net wrote:
>> And also, whether the highest human win rates against bots were achieved by trying to play like the bots
> On May 17, 4:38 am, mu...@compuplus.net wrote:
>> And also, whether the highest human win rates against bots were achieved by trying to play like the bots
> 1. Play money games if possible. If not, play long matches, as long
> as possible, so that you can jack up the cube. Avoid short matches.
These two propositions seem to be contradictory...
> as possible, so that you can jack up the cube. Avoid short matches.
How is a money game similar to a long match?
Isn't a money game essentially a 1-point cubeful game?
In terms of a "match", it's the shortest possible match. How would a human benefit from playing the shortest possible match and long matches at the same time...?
I'm sure there must be a phd level mathematical explanation for this...??
> 2. Don't play like the bot. Get into a massive backgame that the bot
> doesn't understand, and jack up the cube.
As far as "neural nets" and how the bots play by looking ahead so many plies, what makes "backgames" so different that even n-ply look ahead can't help the bots win...?
Again, I'm sure there must be a phd level mathematical explanation for this...??
> The bot will outplay you in "normal" games, but the hope is that you
> can make the games that you understand better count for a lot, by
> using the cube appropriately.
Isn't the "roll-out" the ultimate answer to play any position in any kind of game "appropriately"...?
Isn't the "cubeful roll-out" the ultimate answer to how to make make the cube decisions at any position in any kind of game "appropriately"...??
MK
Tim Chow
May 18, 2012, 8:02:47 PM5/18/12
to
In a money game, there's no limit to how high the cube can go.
If you get the bot into a position that it doesn't understand, and
it's losing but thinks it's winning, then you can double and it will
beaver and send the cube back to you next turn. Then you can repeat
the process. Over a few rolls you can get the cube up to 64 or even
higher. Then you win a ton of points in just one game, which will
create a huge lead and the bot will have trouble catching up.
In a short match, say to 5 points, you can't get the cube higher than
8. In a match to 33 points, you can get the cube up to 64.
> Isn't the "roll-out" the ultimate answer to play any position
> in any kind of game "appropriately"...?
>
> Isn't the "cubeful roll-out" the ultimate answer to how to make make the cube
> decisions at any position in any kind of game "appropriately"...??
Not if the bot doesn't understand most of the positions that arise.
---
Tim Chow
If you get the bot into a position that it doesn't understand, and
it's losing but thinks it's winning, then you can double and it will
beaver and send the cube back to you next turn. Then you can repeat
the process. Over a few rolls you can get the cube up to 64 or even
higher. Then you win a ton of points in just one game, which will
create a huge lead and the bot will have trouble catching up.
In a short match, say to 5 points, you can't get the cube higher than
8. In a match to 33 points, you can get the cube up to 64.
> Isn't the "roll-out" the ultimate answer to play any position
> in any kind of game "appropriately"...?
>
> Isn't the "cubeful roll-out" the ultimate answer to how to make make the cube
> decisions at any position in any kind of game "appropriately"...??
---
Tim Chow
mu...@compuplus.net
May 19, 2012, 5:02:02 AM5/19/12
to
On Friday, May 18, 2012 6:02:47 PM UTC-6, Tim Chow wrote:
> On May 18, 7:31 am, mu...@compuplus.net wrote:
>> How is a money game similar to a long match?
> In a money game, there's no limit to how high the cube can go.
Then, they are not similar but to the contrary they are dissimilar.
> On May 18, 7:31 am, mu...@compuplus.net wrote:
>> How is a money game similar to a long match?
> In a money game, there's no limit to how high the cube can go.
> If you get the bot into a position that it doesn't understand, and
> it's losing but thinks it's winning, then you can double and it will
> beaver and send the cube back to you next turn. Then you can repeat
> the process. Over a few rolls you can get the cube up to 64 or even
> higher. Then you win a ton of points in just one game, which will
> create a huge lead and the bot will have trouble catching up.
Aren't we in the for the long haul and won't the luck even out eventually?
> In a short match, say to 5 points, you can't get the cube higher than
> 8. In a match to 33 points, you can get the cube up to 64.
>> Isn't the "roll-out" the ultimate answer to play any position
>> in any kind of game "appropriately"...?
>> Isn't the "cubeful roll-out" the ultimate answer to how to make make the cube
>> decisions at any position in any kind of game "appropriately"...??
> Not if the bot doesn't understand most of the positions that arise.
Don't worry, I am not taking you seriously. I am just exposing to any readers what kind of a scum you are.
You are the kind of people that I call "teflon coated assholes". Nothing sticks to you... :) Not even your own bullshit.
You and your ilks are amusing me... Thanks ;)
MK
Tim Chow
May 19, 2012, 11:38:01 AM5/19/12
to
On May 19, 5:02 am, mu...@compuplus.net wrote:
> On Friday, May 18, 2012 6:02:47 PM UTC-6, Tim Chow wrote:
> > On May 18, 7:31 am, mu...@compuplus.net wrote:
> >> How is a money game similar to a long match?
> > In a money game, there's no limit to how high the cube can go.
>
> Then, they are not similar but to the contrary they are dissimilar.
They're not exactly the same, of course, but they're similar in that
> On Friday, May 18, 2012 6:02:47 PM UTC-6, Tim Chow wrote:
> > On May 18, 7:31 am, mu...@compuplus.net wrote:
> >> How is a money game similar to a long match?
> > In a money game, there's no limit to how high the cube can go.
>
> Then, they are not similar but to the contrary they are dissimilar.
in both cases, the strategy of jacking up the cube very high is
available. That strategy is not available in a short match.
> Why will the bot have trouble catching up?
>
> Aren't we in the for the long haul and won't the luck even out eventually?
where you have a skill advantage, you win a ton of points. In the
games where the bot has a skill advantage, you give up only a few
points. Let's say that the games in which you have a skill advantage
are only 1/10 as likely as the games in which the bot has a skill
advantage, but you win 64 points on average when you have a skill
advantage and the bot wins only 2 points on average when it has a
skill advantage. In the long run, you'll come out ahead.
> > In a short match, say to 5 points, you can't get the cube higher than
> > 8. In a match to 33 points, you can get the cube up to 64.
>
> Once again now you are contradicting your previous statement.
illogical, but here I can't even do that. What are you trying to say?
> How can the bot not understand any position at all? After a given large number of trials,
> the bot should know what move results in the most wins, no?
> Don't worry, I am not taking you seriously. I am just exposing to any readers
> what kind of a scum you are.
>
> You are the kind of people that I call "teflon coated assholes".
> Nothing sticks to you... :) Not even your own bullshit.
>
> You and your ilks are amusing me... Thanks ;)
---
Tim Chow
0 new messages