[pescetom]

This optimistic slam is down on a lead of hearts or even trumps, given the poor communication between hands. But here after a spades lead even GiB struggles to go down, succeeding only with a masterful play at trick 7.

[smerriman]

Really odd. 100% replicable with the old version of GIB, but it's also fully aware its plan will go down if South has 5 hearts.. yet continues to pick that over any alternative no matter how many times I try (even undo / redo which often fixes some other bugs). Will have to dig into this one some more..

[smerriman]

OK, figured it out. North led the ♠6 and later continued with the ♠2; GIB assumes opponents play the same lead conventions as itself (4th best here, low from xxx); the only valid holding is 62 doubleton. It deals a couple of mismatching hands, but almost all treat that as fact.

Under that assumption, its play is the best line, picking up a 4-0 diamond split where possible.

Relying on that assumption against unknown opponents is of course ridiculous, though I do have some sympathy for GIB here, since lead inferences are an important part of the game (and this isn't a case where it had a 100% alternative; it has to judge how likely the breaks are) - the programmers simply didn't tell it that others play differently to itself.

[johnu]

smerriman, on 2023-April-27, 17:32, said:

GIB assumes opponents play the same lead conventions as itself (4th best here, low from xxx); the only valid holding is 62 doubleton.

Apparently GIB isn't self aware enough to recognize that its leads are almost totally random a large percentage of the time.

[smerriman]

johnu, on 2023-April-27, 20:55, said:

Apparently GIB isn't self aware enough to recognize that its leads are almost totally random a large percentage of the time.

If by 'large percentage' you mean 3.9%.. the times it breaks the rules for no reason are super frustrating, but it still followed them on 11694/12164 hands in that study.. and half of those exceptions were underleading an ace vs a suit contract.

These exceptions I found back then don't seem to exist in the old version of GIB though; it gives the rule-based lead a 0.3 IMP bonus so when all else are equal, it never picks the wrong one.

[pescetom]

smerriman, on 2023-April-27, 17:32, said:

OK, figured it out. North led the ♠6 and later continued with the ♠2; GIB assumes opponents play the same lead conventions as itself (4th best here, low from xxx); the only valid holding is 62 doubleton. It deals a couple of mismatching hands, but almost all treat that as fact.

Under that assumption, its play is the best line, picking up a 4-0 diamond split where possible.

Relying on that assumption against unknown opponents is of course ridiculous, though I do have some sympathy for GIB here, since lead inferences are an important part of the game (and this isn't a case where it had a 100% alternative; it has to judge how likely the breaks are) - the programmers simply didn't tell it that others play differently to itself.

Thanks, I figured that out myself before waking up this morning The lead puzzled me at the time, but I didn't think it would be blindly trusted, especially as from a doubleton it would make little sense.
It's still madness by GIB of course, if only because it was IMPs and so no pressing need for the overtrick.

[smerriman]

pescetom, on 2023-April-28, 02:29, said:

It's still madness by GIB of course, if only because it was IMPs and so no pressing need for the overtrick.

It's not trying for the overtrick; if North has the promised doubleton of spades, the heart play is required to make the contract in case North holds all the trumps.