[allfail]

The bidding went like this:
Me GiB
2C 2D
3C 3D*
5C ?

3D by GiB is double negative and my 5C I guess should be to play. I have:
AKQ Qx AK AKQxxx.

GiB bid 5D having this:
Jxxx Txx Txxx xx


This must be a bug. I can't find any reasonable way to explain.

[TylerE]

I bid 3N over 3♦ and hope that they don't have 5 heart tricks. (The double neg denies A or K so I know we never have slam.)

[allfail]

I thought hoping for 2 Cs and a nice split would be an easier shot.

[helene_t]

Agree with Allfail, at least assuming we are not playing matchpoints. 5♣ just requires clubs to run, while 3NT requires that p has help in hearts or that hearts break 4-4. If clubs don't run we may not make 3NT anyway, say if p has xxx or xxxx.

5♦ may not be a real bug but rather an unfortunate effect of the algorithm GIB uses to make decisions. It simulates a bunch of deals consistent with the auction and then picks the contract that on average gives the best DD result. Maybe by chance a couple of 4♦6♣ hands came up in which 5♦ makes but 5♣ doesn't.

[TylerE]

The thing is that 3N will make on MANY layouts where 5♣ doesn't...any non ♥ lead and 3N is virtually cold, and if they allow you to slip a trick with Q♥ (very possible single dummy, esp against GIB), again, you're cold, regardless of the club situation.

[allfail]

I understand that GiB uses Monte Carlo, but in the light of this bid this process is evidently flawed (or bugged).
We should ask the question why human players would never make this bid. The answer is plainly simple: if partner is interested in playing in diamonds, he(she) would have bid it.
In Monte Carlo language, the bid 2C-3C-5C should deny 4 diamonds with suitable honors, or the player should bid 4D instead of 5C. The constraint of the sampling must have some problems for GiB to find 5D more playable.

As a side note, does anyone know how GiB samples the hand? If it just deal randomly and examine if the hands follows the constraint then it is very inefficient in these cases of a strong/distributional hand to find enough sample and this can also make its bid very unstable. On the other hand, I don't know a way to generate the distributions correctly with the constraint putting by hand.

[helene_t]

My guess would be that sampling isn't much of a problem. Even if GIB uses a suboptimal sampling algorithm (disclaimer: I don't know how it samples), most of the processor time will go to the DD analysis, not the sampling.

[hotShot]

My impression was that GIB is taking every jump as strong and forcing.

[helene_t]

Ah so 5♦ was meant as a help-suit GST?

[junyi_zhu]

allfail, on Jan 18 2010, 01:43 AM, said:

The bidding went like this:
Me GiB
2C 2D
3C 3D*
5C ?

3D by GiB is double negative and my 5C I guess should be to play. I have:
AKQ Qx AK AKQxxx.

GiB bid 5D having this:
Jxxx Txx Txxx xx


This must be a bug. I can't find any reasonable way to explain.

I think once gib has limited his hand, all jumps-to-game bids by partner should be a stop bid, and the algorithm shouldn't allow gib to bid his limited hand again. This kind of mistakes actually cost a lot of my money. You often observe gib bid something like:
1S 2H 2S 4H
p p 4S,
which is just wrong in the basic of bridge.
Also, there are other mistakes in the program, for example, 1S p 2S p 4S x, usually, gib takes this double as take out, which is just impossible. A simple programming rule would solve this problem, if the pass is at a lower level and the second double is at 3 level or higher, the double should be penalty and gib should never pull the double.

[zen]

I agree with the concern about GIB's bidding in these situations. I have seen many times GIB bid this way, and have made sure that when I play in Robot races I do not jump in these situations.

If i jump to the correct spot (which this would be in a total point race) GIB corrects to crazy suits w/support or raises a level a large majority of the time. In not jumping I may not get to the right level, but still go plus.

[barmar]

I think some of these auctions are due to GIB re-evaluating in light of the opponents' bidding and its own shape. And I've found that in many cases, its judgement turns out to be right. It will sometimes make a single raise with 5-card support, but if the opponents go to game it will bid 4-over-4, and it will be a make or good sacrifice. Some of these are hands where humans might have jumped straight to 4 earlier in the auction, but GIB sometimes likes to bid slower.