Play of the Hand and Defense by GIB

[otangu]

Hi,

 

I’ve been meaning to post on this for a while and what happened today in deal 16 was last straw.

 

I understand that GIB’s strategy is IMP not Matchpoints but most of us hires robots for the Pairs Tournaments, right?

 

Kindly check the link for deal 16, https://webutil.bridgebase.com/v2/tview.php?t=8146-1592910000&u=otangu&v3b=web&v3v=5.6.3 for otangu

 

Deal 14: A heart was ruffed with the trump K and later GIB plays trump 10 and overtakes with the Queen. After ruffing the diamond lead in the first trick, most of us would lay down the Ace at least to see how the suit breaks.

 

Deal 16: Contract is 1NT and after an unsuccessful club finesse, Robot gains the lead and plays another club to Dummy and finesses again. Had it won with the Jack, the contract would have been made.

 

There are a lot of playing techniques explained in the “Play of the Hand” textbooks and Robot accommodates almost none of them.

 

During the Pandemic, playing with a robot in free tourneys must be a considerable source of income for BBO and they really should invest to improve not only Play of the Hand but defensive play too.

 

Awaiting feedbacks

[smerriman]

I understand that GIB’s strategy is IMP not Matchpoints but most of us hires robots for the Pairs Tournaments, right?

No, GIB plays both IMPs and MPs. The general logic is the same, but when it calculates the outcome for possible simulations, it affects how they are scored. This therefore changes what it plays based on the scoring.

 

Accordingly, I don't see anything wrong with deal 16; obviously if it were IMPs, taking the Ace would be correct, but at MPs the potential overtricks are important.

 

Had you rented an advanced bot or a basic bot? Deal 14 is silly, but makes sense if it were just a basic bot. At the point it leads from hand, double dummy tells it all trumps are equivalent (all making +1), so it doesn't distinguish them. Only Advanced GIB has a single-dummy version built in which is able to realise that it does matter.

[otangu]

Advanced Robot as I pay 0.25 USD each and every time. I don't share your view that GIB plays MP as well. If it did he would have taken marked finesses when it was obvious in many occasions

[otangu]

Hi,

 

Since I started this post I decided to share my experiences with bidding, play and defense while playing with Robot and hope that BBO Management will take notice.

 

Based on GIB system notes published by BBO, GIB plays Lebensohl after 1NT opening. Today, (check out deal 2, https://webutil.bridgebase.com/v2/tview.php?th=3019~ozgull&u=otangu&v3b=web&v3v=5.6.4) Robot opened 1NT and RHO interfered with 2H. I bid 3D forcing with 9HCP and 11 total points with Diamond length. Robot rebid 3S.

 

Vul vs Non-Vul GIB

West North East South

1NT 2H

3D Pass 3S Pass

4S

 

If I had 4 spades I’d have bid 3H (Stayman) instead of 3D and thought that Robot had 5 spades and raised it to the game. As you can see, Robot had double stoppers in hearts and should have bid 3NT which was what was played im majority of the tables

[barmar]

Advanced Robot as I pay 0.25 USD each and every time. I don't share your view that GIB plays MP as well. If it did he would have taken marked finesses when it was obvious in many occasions

He didn't say it plays MP as well as IMPs, he just said it takes the form of scoring into account. But its strategy isn't always perfect. It only simulates a few dozen hands hands (it can't do hundreds of simulations in the time it has to make a decision), so a single outlier will sometimes bias the results.

[otangu]

Hi,

 

In of hands I played with GIB today, the contract was 1S and GIB was on lead with Q, Q986, QJ103, Q954 and laid down Q of Spades.

 

In GIB system notes, it claims that “Read the book Winning Suit Contract Leads for insight on the way GIB leads against suits.” and I have a copy of the book and it never advocates a singleton trump lead whatsoever. As a result, we got the lowest score for the board.

[otangu]

Hi,

 

In of hands I played with GIB today, the contract was 1S and GIB was on lead with Q, Q986, QJ103, Q954 and laid down Q of Spades.

 

In GIB system notes, it claims that “Read the book Winning Suit Contract Leads for insight on the way GIB leads against suits.” and I have a copy of the book and it never advocates a singleton trump lead whatsoever. As a result, we got the lowest score for the board.

 

 

Another hand in today's tournament and GIB holds 85, J1065, A1095, Q109 and the auction went:

1S - 2H, 4H and in between, Robot and I passed. Robot was on lead and attacked with 5 of diamonds under the Ace. Consequently, we got another lowest score.

[smerriman]

Hi,

 

In of hands I played with GIB today, the contract was 1S and GIB was on lead with Q, Q986, QJ103, Q954 and laid down Q of Spades.

 

In GIB system notes, it claims that “Read the book Winning Suit Contract Leads for insight on the way GIB leads against suits.” and I have a copy of the book and it never advocates a singleton trump lead whatsoever. As a result, we got the lowest score for the board.

GIB isn't programmed to follow generic advice from that book. It simply uses the same logic as the book did - generate hands at random, and see which card gives the best double dummy result on average. However, it simulates far less hands than the book did, so there is always a good chance the card it plays is not the best in the long run.

[otangu]

GIB isn't programmed to follow generic advice from that book. It simply uses the same logic as the book did - generate hands at random, and see which card gives the best double dummy result on average. However, it simulates far less hands than the book did, so there is always a good chance the card it plays is not the best in the long run.

 

Been playing with GIB since March and I'd write a novel re its leads and its defense in general.

 

For instance, today, I was dealt with Q2, Q10, J64, KQJ10xx and led K of clubs for a NT contract. Universally known in bridge circus that this kind of lead would be made from KQJ at least. GIB had Ax in clubs and won the 2nd club. The contract was made and if it had won the first lead and returned the 2nd club the game would have been set by 2 tricks

[otangu]

Been playing with GIB since March and I'd write a novel re its leads and its defense in general.

 

For instance, today, I was dealt with Q2, Q10, J64, KQJ10xx and led K of clubs for a NT contract. Universally known in bridge circus that this kind of lead would be made from KQJ at least. GIB had Ax in clubs and won the 2nd club. The contract was made and if it had won the first lead and returned the 2nd club the game would have been set by 2 tricks

 

The reason I bring up such glitches is that BBO management would take notice and make some improvements such as the player of Bridge Baron or Jack 6 or am I day-dreaming?

[otangu]

The reason I bring up such glitches is that BBO management would take notice and make some improvements such as the player of Bridge Baron or Jack 6 or am I day-dreaming?

 

Just played another tourney and check out this deal,

http://www.bridgebase.com/tools/handviewer.html?bbo=y&myhand=M-190376834-1594989634

 

Out of twenty tables, one pair sacrificed in 5D doubled and lost -800 and the rest made 4H or 4H +1 except that GIB couldn't make it.

[barmar]

The reason I bring up such glitches is that BBO management would take notice and make some improvements such as the player of Bridge Baron or Jack 6 or am I day-dreaming?

If we were to replace GIB it would probably be with Argine, the robot used by FunBridge, since we're part of the same company.

 

But substituting a new robot is not a simple task.

[Vampyr]

If we were to replace GIB it would probably be with Argine, the robot used by FunBridge, since we're part of the same company.

 

 

Is Argine as good as Jack? I really think that BBO should invest in the best robot available.

[barmar]

Is Argine as good as Jack? I really think that BBO should invest in the best robot available.

You're assuming Jack is for sale. I don't know if it is or not.

 

Plus, if we use Argine, we already have programmers who know how to integrate it. If we license something from another company, the development costs will be higher, on top of having to pay for the software.

 

Our programmer resources are VERY limited.

[Vampyr]

You're assuming Jack is for sale. I don't know if it is or not.

 

Plus, if we use Argine, we already have programmers who know how to integrate it. If we license something from another company, the development costs will be higher, on top of having to pay for the software.

 

Our programmer resources are VERY limited.

 

One solution would be to replace the current programmers with Jack’s programmers. Or hire them in addition; surely BBO is making record amounts of money at present.

[mycroft]

Not BarMar, but this is the problem they got into with GIB. When Jack's programmers get hired, even if BBO can do it, when they find a better job that they want to do (or have a better job that they do full-time, with Jack being a nice sideline), they leave, and now BBO has yet another play engine they can't fix and will decay (or be surpassed) over time.

[otangu]

Hi,

 

In a tournament yesterday, GIB attacked from a doubleton clubs against a 3NT contract,

 

http://www.bridgebase.com/tools/handviewer.html?bbo=y&myhand=M-206547772-1595446382

 

No expert view or a computer simulation justifies such a lead and I strongly suggest that BBO management should consider to revisit GIB’s defense. As always, I ended up with the lowest score for this board.

 

Regards,

Ozcan

[otangu]

Another mishap in bidding by GIB, http://www.bridgebase.com/tools/handviewer.html?bbo=y&myhand=M-221784415-1595878319

 

My Double after second Diamond rebid (3D) was surely for Penalty. If'd doubled for Takeout I'd have done it against 2D.

[johnu]

Another mishap in bidding by GIB, http://www.bridgebase.com/tools/handviewer.html?bbo=y&myhand=M-221784415-1595878319

 

My Double after second Diamond rebid (3D) was surely for Penalty. If'd doubled for Takeout I'd have done it against 2D.

My Double after second Diamond rebid (3D) was surely for Penalty.

And you would be wrong that the double of 3D was for penalty. The description for double is "11- total points". This is GIB you are talking about so meanings of bids/doubles/passes may only have a passing resemblance to real life bridge. There is no bridge logic in GIB's bids, just lines of code and datasets.

[otangu]

Hi,

 

Check out this, http://www.bridgebase.com/tools/handviewer.html?bbo=y&myhand=M-229193350-1596096001

 

GIB plays Cappelletti and I duly bid 2H with a 5 Diamond on the side. GIB bid 2S and I passed because I thought there was no support for hearts and neither for a minor showing at least a 6 Spades. The result was, as always, the lowest score for the board

[oldoc]

Let's all be honest. A GIB defensive opening lead is 6 to 5 to lose a trick. This is the one area that for years has screamed for significant change (that never comes). After a routine garbage opening lead Gib's subsequent defensive play improves.

[Stephen Tu]

GIB's opening leads aren't really that bad IMO in general, they are a bit "weird" looking in that they don't follow human rules of thumb, relying on statistical simulation, but the methodology is generally sound (except maybe should bias more against leading stiff honors etc. that tend to remove declarer's guesses). It's not like humans don't throw tricks left & right on opening lead also as it is the hardest card to select, with the least information available.

 

The main problem with GIB's defense is the near total lack of signalling, and lack of ability to interpret signalling, ability to bias its hypothetical possibilities of the rest of the cards based on partner's carding, line of attack, and inferences from declarer's line of play. The original designer had ideas about how to do such things, but it was computationally infeasible given the constraints of PC processing power at the time.

 

Anyway, it's a giant waste of time to complain about GIB's play/defense as their programming personnel just don't have the expertise and/or man hours to make it better. The real brains behind the project abandoned it almost 20 years ago in favor of his "day job" and other AI interests, sold it off to BBO who can't do much more than tweak the bidding database to fix a few bugs and plug some holes.

 

You can post stuff like that ridiculous Cappelletti bug in the post previous (preferably separate post per bug as is this forum's suggested policy), as that stuff is more fixable. Whether they ever get around to it, given their huge backlog and super slow release schedule the past few years is another question. Maybe barmar can leak some info on what BBO is thinking about robot plans these days, if it's not against company policy / considered important secret. I can see their side, they don't have a lot of competition and people still play a ton of robo tournaments despite their current limitations. So what is their incentive to sink a lot of money into improving bot or buying/integrating a different one, if people seem mostly content with the current mediocrity? The bot is a better player than the average BBO user.

[kontoleon]

I try the bbo gib robot in dublucation

 

Sometimes if I press undo, on exacly the some possision play much deference card. I am sure if try to cound information from pre undo info but this is little wild.

[kontoleon]

I try the bbo gib robot in dublucation

 

Sometimes if I press undo, on exacly the some possision play much deference card. I am sure if try to cound information from pre undo info but this is little wild.