On February 1, 2022 at 10:18:58 PM UTC-7, Tim Chow wrote:
Below is a position that just came up today. I "automatically"
played 20/14 and even Stick might call this a whopper....
XGID=-b----E-D--BbB---cbb-Ab-bA:0:0:1:65:3:0:0:7:10
1. XG Roller++ Bar/20 21/15 eq:-0.616
2. XG Roller++ Bar/14 eq:-0.716 (-0.100)
The third play is this:
3. XG Roller++ Bar/20 8/2 eq:-1.000 (-0.377)
I decided to dwell on this a little more because for someone
who tries to discredit the bots at every opportunity, this one
is too big of a fish to just catch/release and well worth frying.
What makes this such a big catch is that the bar/20 8/2 (third
option above and "my"correct play) is not even considered by
the bot. Showing that bots can be so terribly wrong may shake
the faith of even the most devout among the bot worshipping
gambling flock.
First, Chow's position is very different than BlueDice's original
position: XGID=-b---AD-C-A-dD----bcA-b-bA:0:0:1:64:0:0:0:0:10
which had already two other blots. In this one, there aren't other
blots but there are two towers on X's 8 and 6 points instead.
Playing the 6 from the 20 or 21 point is a "false similarity" and
it shows how stupid one can be to automatically follow stupid
rules without considering the rest of the board carefully first.
Chow's having already posted the rollout allows me to readily
state that the bot's rating bar/20 8/2 so low is a sure indication
that it will immediately double if X plays that.
And from there, what should jump at someone's face is that the
cube will almost never stay at 4, since O will necessarily double
again to 8. Thus, this is a perfect opportunity to jack up the cube,
especially if X considers himself the underdog against the bot.
Surely O can roll some numbers that will decimate X but with the
cube at 2, the resulting scores will be 2/7 vs 3/7 or 4/7 vs 3/7 and
extremely rarely 6/7 vs 3/7, which means that afterwards X will still
have an almost equal chance of winning the match on the average.
O can roll quite awkward numbers also and even in an in-between
case of hitting X on the 2 point accompanied by awkward numbers,
X's chances of winning could improve considerably. Some may even
wish that they could switch to Nackgammon by taking a piece each
from X's 8 and 6 points, to anchor them on O's 23 point, which would
also improve X's chances of winning, although by less.
Let's also consider that with the cube at 2, the bot still gives X about
35% chance of winning to make the score 0/7 vs 5/7 and 8% chance
of winning a gammon and the match with it. Since the bot's numbers
are grossly wrong, X will actually fare much better.
Now, if X gets to redouble to 4, O will be forced to redouble to 8 also
and it will become a DMP game. Assuming that X doubles correctly,
(per the bot's inaccurate equity calculations), X will win more since
O will be forced to double even if incorrectly, (again all per the bot's inaccurate equity calculations); and also assuming that both X and O
play checker moves like the bot would play.
Needless to say, I suggest that X played by "Murat mutant" would do
much better than the bot's estimates and I will leave it to you all to
convince yourselves of that, perhaps with a little help from Axel. ;)
Of couse, O will also win some matches with the cube at 8 in DMP
games. What we need to do is weigh the sides' chances of winning
at once with cube at 8 vs their chances of winning in playing on from
all other different scores scores reached after the current game.
Last night I played about 60 games from Chow's position against XG
Roller++. X won more of the DMP's. O won more of the games with
the cube at 2, which resulted in all combinations of N/away-N/away
scores. I din't want to waste time playing those matches to the end
since we can extrapolate the expected results, (even if by the bot's
inaccurate calculations, which would be good enough for me also, as
they would apply to both sides from that point on).
I may be surprising you all, as I'm surprising myself, by bothering to
write such stuff but don't get worried about my writing a book yet. :)
MK
--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)