GIB play

[Hyperon]

I received a hand where GIB's declarer play does not exceed. Any ideas on why GIB fails to take his 13 top tricks?

 

Play

[1eyedjack]

Which version of GIB?

[Hyperon]

Which version of GIB?

 

New version

[lycier]

I received a hand where GIB's declarer play does not exceed. Any ideas on why GIB fails to take his 13 top tricks?

 

Play

 

You said you use new version of Gib. This is a very strange thing, so I have to replay your hand .

1- The hand you played :

[hv=http://www.bridgebase.com/tools/handviewer.html?lin=nt|blink128%20has%20sent%20you%20a%20hand.%20%20slechtste%20afspel%20door%20bot%20ooit|pn%7CRobot%2CRobot%2Cblink128%2CRobot%7Cst%7C%7Cmd%7C1S985HT976D9432C76%2CSA6HK4DQ865CAQ953%2CS74HJ532DKT7CJT42%2CSKQJT32HAQ8DAJCK8%7Csv%7Cb%7Crh%7C%7Cah%7CBoard%207%7Cpg%7C%7Cmb%7CP%7Cmb%7C1N%7Can%7Cnotrump%20opener.%20Could%20have%205M.%20--%202-5%20%21C%3B%202-5%20%21D%3B%202-5%20%21H%3B%202-5%20%21S%3B%2015-17%20HCP%3B%2018-%20total%20points%7Cmb%7CP%7Cmb%7C7N%7Can%7C21%2B%20HCP%7Cmb%7CP%7Cmb%7CP%7Cmb%7CP%7Cpg%7C%7Cpc%7CCT%7Cpc%7CC8%7Cpc%7CC6%7Cpc%7CCQ%7Cpg%7C%7Cpc%7CD5%7Cpc%7CD7%7Cpc%7CDA%7Cpc%7CD9%7Cpg%7C%7Cpc%7CS3%7Cpc%7CS8%7Cpc%7CSA%7Cpc%7CS4%7Cpg%7C%7Cpc%7CS6%7Cpc%7CS7%7Cpc%7CSK%7Cpc%7CS9%7Cpg%7C%7Cpc%7CCK%7Cpc%7CC7%7Cpc%7CC3%7Cpc%7CC2%7Cpg%7C%7Cpc%7CH8%7Cpc%7CH7%7Cpc%7CHK%7Cpc%7CH2%7Cpg%7C%7Cpc%7CCA%7Cpc%7CC4%7Cpc%7CS2%7Cpc%7CD3%7Cpg%7C%7Cpc%7CH4%7Cpc%7CH3%7Cpc%7CHA%7Cpc%7CH9%7Cpg%7C%7Cpc%7CSQ%7Cpc%7CS5%7Cpc%7CD8%7Cpc%7CH5%7Cpg%7C%7Cpc%7CSJ%7Cpc%7CHT%7Cpc%7CD6%7Cpc%7CDT%7Cpg%7C%7Cpc%7CST%7Cpc%7CD4%7Cpc%7CC9%7Cpc%7CHJ%7Cpg%7C%7Cpc%7CHQ%7Cpc%7CH6%7Cpc%7CC5%7Cpc%7CCJ%7Cpg%7C%7Cpc%7CDJ%7Cpc%7CD2%7Cpc%7CDQ%7Cpc%7CDK%7Cpg%7C%7C]400|300[/hv]

Result : 7NTW-1

 

2- The hand Gibs played :

 

[hv=http://www.bridgebase.com/tools/handviewer.html?sn=机器人&s=S985HT976D9432C76&wn=机器人&w=SA6HK4DQ865CAQ953&nn=机器人&n=S74HJ532DKT7CJT42&en=机器人&e=SKQJT32HAQ8DAJCK8&d=s&v=o&b=11&a=P1N(notrump%20opener.%20Could%20have%205M.%20--%202-5%20%21C%3B%202-5%20%21D%3B%202-5%20%21H%3B%202-5%20%21S%3B%2015-17%20HCP%3B%2018-%20total%20points)P7N(21+%20HCP)PPP&p=H2HQH6H4S3S8SAS7C3C2CKC6H8H7HKH3D8D7DAD4C8C7CAC4CQCTDJD2S6S4SKS9SQS5D6H5SJHTD5DTSTD9C9CJS2D3C5HJHAH9DQDK]400|300[/hv]

Result : 7NTW=

 

3- The hand I replayed :

 

[hv=http://www.bridgebase.com/tools/handviewer.html?sn=机器人&s=S985HT976D9432C76&wn=机器人&w=SA6HK4DQ865CAQ953&nn=lycier&n=S74HJ532DKT7CJT42&en=机器人&e=SKQJT32HAQ8DAJCK8&d=s&v=o&b=11&a=P1N(notrump%20opener.%20Could%20have%205M.%20--%202-5%20%21C%3B%202-5%20%21D%3B%202-5%20%21H%3B%202-5%20%21S%3B%2015-17%20HCP%3B%2018-%20total%20points)P7N(21+%20HCP)PPP&p=CTC8C6CQD8D7DAD9S3S9SAS4C5C2CKC7H8H7HKH2CAC4S2D2H4H3HAH9HQHTD6H5SKS5S6S7SQS8C9DTSJD4D5CJSTH6C3HJDJD3DQDK]400|300[/hv]

Result : 7NTW-1

Same lead, same result. I don't know the reason,just reporting.

I guess perhaps BBO Gib experts or Stephen Tu can answer this question.

[lycier]

Just now I replay it again, but I have a different lead for this time, surprisedly, Gib got 13 tricks !

 

[hv=http://www.bridgebase.com/tools/handviewer.html?sn=机器人&s=S985HT976D9432C76&wn=机器人&w=SA6HK4DQ865CAQ953&nn=lycier&n=S74HJ532DKT7CJT42&en=机器人&e=SKQJT32HAQ8DAJCK8&d=s&v=o&b=11&a=P1N(notrump%20opener.%20Could%20have%205M.%20--%202-5%20%21C%3B%202-5%20%21D%3B%202-5%20%21H%3B%202-5%20%21S%3B%2015-17%20HCP%3B%2018-%20total%20points)P7N(21+%20HCP)PPP&p=CJC8C6CQD5D7DAD4S3S8SAS7C9C2CKC7H8H7HKH3CAC4DJH9S6S4SKS9SQS5D8DTSJD3C5CTSTHTD6H2S2D9H4H5HAH6C3HJHQD2DQDK]400|300[/hv]

 

Result : 7NTW=

This time I lead ♣J instead of ♣10, it turned out that W Gib played so well. If lead ♣10, W Gib only got 12 tricks! very very strange.

[Stephen Tu]

Obviously GIB is playing under the idea that the lead of the T 100% denies the J, therefore clubs split 3-3, therefore it can throw away spade winners it doesn't need.

 

While this is reasonable to assume in most contracts, obviously it should be tweaked for slams. Goes along with the idea of defending assuming declarer doesn't always have their bid. It would take more resources, essentially you'd have to run simulations twice, once with bidding/carding constraints, once with looser constraints, and then find if there is a play that is clearly best no matter what. Not necessarily the easiest thing to implement?

[1eyedjack]

New version

There are always at least two versions in play at any time. Each new release gives rise to a "basic" and "advanced" version. I never really got to grips with where they differ and where they behave the same. Maybe it makes no difference on this hand?

[johnu]

New version

 

The Basic version is basically the Advanced version with a frontal lobotomy. Based on statements from BBO developers, the Basic version has much more limited resources in time (and memory?) and doesn't use some of the more advanced programming. The Basic version makes strange bids and plays that make no sense all the time and the BBO explanation is basically that you get what you pay for and don't expect to get expert robot play from the Basic versions.

 

If this was an advanced robot, then it is worth sending in a report but play bugs are apparently very hard to fix so even then, this may never be fixed.

[Bbradley62]

This hand was played in MBC on 3/16. Only the player (and probably BBO's accounting system) knows whether he was renting a basic or advanced GIB at the time.

[Hyperon]

This hand was played in MBC on 3/16. Only the player (and probably BBO's accounting system) knows whether he was renting a basic or advanced GIB at the time.

 

It was a free junior bot and I'm informed that they are basic.

[iandayre]

The one thing I don't understand about these types of situations, is that GIB ALWAYS seems to choose the assumption based on opponents card play or bidding, rather that the sure thing of cashing top tricks. It is as though GIB wants to prove its intellectual superiority. But unfortunately it can't learn from experience that it is not infallible.

[Stephen Tu]

It's tricky because most of the time we DO want GIB to use inferences from the carding. It is restricting its study sample space, so from its point of view, running clubs is just as "sure thing" as running spades, it doesn't see there is a difference. In order for it to understand there is a difference, we have to force it to analyze a larger unconstrained sample space. This takes time and resources, and perhaps it is tricky to program it when to take the extra time.

[cloa513]

It's tricky because most of the time we DO want GIB to use inferences from the carding. It is restricting its study sample space, so from its point of view, running clubs is just as "sure thing" as running spades, it doesn't see there is a difference. In order for it to understand there is a difference, we have to force it to analyze a larger unconstrained sample space. This takes time and resources, and perhaps it is tricky to program it when to take the extra time.

Its not tricky because that would no difference to the play except it does because the programming is broken. Again its a lie that GIB follows DD.

[1eyedjack]

Again its a lie that GIB follows DD.

It would be a lie if anyone (with authority) were to claim it. But they haven't. And we should be grateful that it does not. That would give GIB a huge unfair advantage.

[lycier]

I can quite believe Stephen Tu, never think he would lie .

[1eyedjack]

I can quite believe Stephen Tu, never think he would lie .

What has this to do with the price of fish?

[Bbradley62]

It's tricky because most of the time we DO want GIB to use inferences from the carding. It is restricting its study sample space, so from its point of view, running clubs is just as "sure thing" as running spades, it doesn't see there is a difference. In order for it to understand there is a difference, we have to force it to analyze a larger unconstrained sample space. This takes time and resources, and perhaps it is tricky to program it when to take the extra time.

Yeah, but sometimes it would be nice if GIB could count 6S+3H+1D+3C=13 tricks guaranteed against any lie of the cards before considering any other option and stop its thinking there.