BBO Discussion Forums: Maybe it was always broken ? - BBO Discussion Forums

Otherwise what was going on here ?

After trick 7, if GiB keeps even basic track of play it can count 6 top tricks: 2 spades, 1 heart, 1 diamond and 2 clubs.
Why should it take any different line than dropping these, even if "certain" due to autistic assumptions from the auction or early play?

I apologize in advance to smerriman who has probably explained this to me before, but GiB logic is not human logic and not winning logic here either (21% MP).

Well played by GIB. A lesser robot might have gone down.

pescetom, on 2024-February-10, 15:51, said:

pescetom, on 2024-February-10, 15:51, said:

The point exactly. Humans see 6 top tricks, while GIB doesn't have the remotest concept of what a top trick is. All it knows is that every time it tried dealing the remaining cards to the opposition, playing x, y, or z will result in it taking all of the remaining tricks, so it chooses one of x, y, or z.

Sure, you could program GIB to check if there are multiple ties for first, and then run a completely separate set of simulations independent from the bidding to determine whether any of those break the ties.. but running simulations is costly, and it simply doesn't.

My best guess is the GIB program is faulty in general. For instance, I played multiple hands where a bot defender was holding on to the suit nobody else held while its pitches in another suit would squeeze the other bot. For the hand 'at hand', If like someone above stated, bot simulates at every trick, the simulations are poorly programmed since they don't lead to taking all the tricks.

fuzzyquack, on 2024-February-10, 19:59, said:

West is 'guaranteed' to have more points for the 1♠ overcall than the human actually did so the simulations can't distinguish lines which are 100% to lines which are 100% based on its assumptions. That's what pescetom was referring to. It doesn't say, oh, I have all the top tricks via this other line even if the human bid differently, because it has no concept of what tricks are without dealing hands.

I don't understand what you mean.
All I see is that it leaves the CA stranded, and that's bad regardless off assumptions or simulations.
What else are you referring to? The risk of taking the heart finesse?

lorserker, on 2024-February-11, 00:59, said:

If you look at GIB's simulations at later tricks, you will discover they all assume West has both the ♥K and the ♦Q, due to the points it requires for an overcall. GIB therefore sees no risk in either the heart finesse (it believes it is "guaranteed" to work at trick 7) nor the ♣J at trick 9 (it believes the ♥K now is "guaranteed" to be singleton to make space for the known ♦Q alongside the known 4 remaining spades, so the club ace isn't stranded at all).

Obviously both of those are awful, but it's not bad "regardless of assumptions or simulations", it's specifically because of the simulations based on the assumptions, and that's just how GIB works. Unless you are planning to change the fundamentals of how GIB works.. but if you were planning on fixing things, there are far greater play issues that should be fixed first..

I understand, thanks.

The heart finesse is risky, but I think that it's ok to take because it can gain a trick, and it's positive EV even in a team game.
I have also seen GIB frequently take finesses when there can be no gain because there is a 100% line available already - this I plan to fix soon.

It looks like the later play in the posted hand (being sure that it will drop the HK) has the same cause as taking unnecessary finesses, so I want to fix it too.

About urgent card play issues which need to be fixed. If you can share your top 5, I am happy to listen.

#1 would definitely be fixing the issue where it regularly calculates double dummy scores incorrectly. This results in it making 0% plays - i.e. playing cards that are guaranteed to be strictly worse than any other possible card on literally every single possible distribution of the remaining cards, due to it giving wrong results for simple 3 or 4 card endings.

Full analysis here; this still pops up frequently in hands posted on the forum; in fact there was another occurrence reported on Facebook just a couple of days ago. Even if basic double dummy errors are a result of insufficient calculation time, there's no way this can't be fixed on small end positions at the very minimum (but double dummy solvers these days should be fast enough to work on all hands regardless..)

This doesn't just result in 0% plays, but likely impacts the calculations for many non-obvious decisions too.

The unrelated GIBSON 0% play bug is also well worth investigating, since that feels like GIB is caching incorrect data, which could have wide-ranging effects on its algorithm.

thanks. i am aware of the 0% plays and i think i have fixed it. it will come out in the next version.

i will look at the second example too.

maybe it was always broken is a massive understatement. Now with the new changes it is beyond trying to even simulate in the tiniest of ways of playing the game bridge.
Bridge is a great game and these robots are screwing up a whole generation of bridge players with the crap there programed to pull off. Robot BBO tournaments now are the biggest
waste of money you can throw down drain. There is no communication between the robot partner and the player at all, if any, it is an illusion to lead the human player away
from the joy and logic of a real bridge partner that is actually on your side. WHEN THEY ADVERTISE play against 3 robots they are not lying. You as the human in this sham rip off
event you've paid money for are the 'mark'. BBOs mark for more money. SAVE your time and money play with real people, at least then you'll learn and grow in real bridge experience.

lorserker, on 2024-February-11, 04:09, said:

Is this still in the pipeline, or have you discovered the mysterious and secretive roadblock that prevents all progress?

yes, it is. and i think that it will come out soon.