A few silly problems in the gib bidding

[xxhong]

Can gib really play some reliable bridge as its description?

One hand:

(gib)

KQxxx

Kx

Qxx

Kxx

 

(mine)

Ax

AT9xxxx

KJ9x

-

 

The bidding:

1S(gib) p 2H(me) 3C(human opp)

3H(!) 4C 4N 5C

Doulbe(claiming playing DOPI)

Here, gib should never raise to 3H with 2 hearts a horrible hand. Then later, it lied again, with one KC, it should pass instead of doubling.

 

Example 2:

Gib

Qxxxxx

x

xxx

AQT

 

Mine

AKJx

QJx

Kxx

xxx

 

The bidding went:

1H(opp's gib) 1S(my gib) 1N(opp's psyche) Double(it shows 18-19? This is just ridiculous and opens up a lot of psyche possibilities without good reasons)

3D(opp's gib, it shows 12+ points, another nonsense) Double(Another lie, this double shows 2 diamonds exactly, but gib holds 3) 3H(human opp holds 4 hearts and 2 jacks) Pass(which is forcing, but the double shows 18-19 and forces to game, the pass intended to hang them more if human's gib bids more above 3H)

pass(opp's gib's value is very limited) Double( lied again, double shows exactly 2 hearts) pass 3N(3N played from my side looks good, it turns out to be a very good contract after H or D lead. )

Double(opp's gib now got hooked) 4S(! fine, gib doesn't like playing 3NT, and rebid 4S with a bad suit) pass pass(slam is remote)

Double(opp's gib try to penalize it) p p (Redouble, trying to maximize the profit)

p 5C(Cuebid! this is nonsense, if 4S redouble made 6, the score is higher)

 

After a 1NT psyche, gib made so many bidding mistakes in just one hand. There are some basic logical bugs in the code for sure. Also, after the psyche, it just makes no sense to do any simulations, all simulations are based on the wrong assumptions. The correct strategy is to only use

the input from partner and opp's gib as reliable constraints to do the simulation, not opp's psyche. And the pull of the redouble is really really

very very bad.

 

As I said again and again, the programmer should really play more bridge with gib to understand and then solve all those problems. Gib in many cases

is free to deviate one or even two cards from its system description if the flawed simulation suggests something else. The flawed system and simulation then in turn opens a million profitable psyche possibilities.

[mgoetze]

the programmer should really play more bridge with gib to understand and then solve all those problems.

 

The OP should spend more time with the forum software to understand how to present hands in a format anyone could be bothered to read.

[cloa513]

I had no problem reading that but here it is in forum software

1.

[hv=pc=n&s=sa2hat95432dkj54c&n=skq543hk6dq86ck87&d=n&v=0&b=1&a=1sp2h3c3h4c4n5cd]266|200|Di d you bid 6H[/hv]

 

 

My comment: 3♥ is a reasonable practical bid but the DOPI lie is unforgivable.

 

2.

 

[hv=pc=n&s=sakj2hqj5dk84c654&n=sq87654h2d765caqt&d=w&v=0&b=8&a=1h1s1n(psche)d3dd(shows%202D)3h(has%204H%20and%202J%27s)p(forcing)pdp3nd(Penalty)4s(bad%20hand)ppdpprp5c(cue%3F)p5s]266|200|X- 18-19HCP[/hv]

 

North and West are GIBs

Some rediculous bids made by your GIB.

 

comment: rediculous bids, the limitation of the first double to 18-19 is absurd but you show allow for it especially as overcaller can be weak- you have spades so they will run anyway so don't waste your time with the double. 3D 12+ is not rediculous with lots of diamonds and hearts. The cue statement is unforgivable.

[Bbradley62]

The OP should spend more time with the forum software to understand how to present hands in a format anyone could be bothered to read.

.. or should report hands that are findable in myhands, so we can see the actual explanations, not second-hand interpretations.

[xxhong]

Money bridge hands are not findable in myhands. Those interpretations are mostly accurate. The purpose of this post is to point out how buggy gib's bidding is and how much it can deviate from its own system.

In my impression, bbo's programmer is busy adding new features to the bidding and lazy to solve old existing problems. Many problems existed many years ago when gib first landed at BBO.

Now the gib system has too many conventions and many of the conventions are rather badly programmed with a lot of flaws. And many hand types are impossible to show (For example, a strong hand after opps opened and responded. Take-out double would show exactly the correct shape, so strong hands without correct shape would be out of bids. Also, Robot tends not to make penalty doubles in many obvious situations. All these problems created a lot of psyche possibilities to destabilize the whole system more. Also bbo tends to encourage such kind of behaviors, instead of solving those existing problems. Fred Gitelman once said something like you can win playing with gib if you manipulate the bot well. IMO, the correct attitude should be if you play bridge well, you can win, not if you know how to induce more bidding mistakes from opp's bot, you can win. ) A robust system should also be firstly built on some very simple logic with as few conventions as possible. Gib's bidding logic is so bad in some cases, like the pull of the redoubled contract, or the DOPI mentioned in the post. Of course, what I mention happens rarely in robot tournaments, so they are probably of low priorities to solve, since money bridge doesn't attract a lot of players. Among all the players who frequently play 1 cent or 2 cents, I am probably the only one who regularly post in this forum.

 

Another impression is that the code is quite badly written without good designs in early days. Therefore, it is very difficult to add some very simple features like summoning gibs in the middle of the play to finish the hand, or allow the human player to declare when in dummy.

 

.. or should report hands that are findable in myhands, so we can see the actual explanations, not second-hand interpretations.

[xxhong]

Thank you. I did bid 6 and the trump is 4-0.

I had no problem reading that but here it is in forum software

1.

[hv=pc=n&s=sa2hat95432dkj54c&n=skq543hk6dq86ck87&d=n&v=0&b=1&a=1sp2h3c3h4c4n5cd]266|200|Di d you bid 6H[/hv]

 

 

My comment: 3♥ is a reasonable practical bid but the DOPI lie is unforgivable.

 

2.

 

[hv=pc=n&s=sakj2hqj5dk84c654&n=sq87654h2d765caqt&d=w&v=0&b=8&a=1h1s1n(psche)d3dd(shows%202D)3h(has%204H%20and%202J%27s)p(forcing)pdp3nd(Penalty)4s(bad%20hand)ppdpprp5c(cue%3F)p5s]266|200|X- 18-19HCP[/hv]

 

North and West are GIBs

Some rediculous bids made by your GIB.

 

comment: rediculous bids, the limitation of the first double to 18-19 is absurd but you show allow for it especially as overcaller can be weak- you have spades so they will run anyway so don't waste your time with the double. 3D 12+ is not rediculous with lots of diamonds and hearts. The cue statement is unforgivable.

[Bbradley62]

Let's try to break this down into something the programmers can use to improve GIB. The most useful threads are the ones that say "in this circumstance, GIB did x when he should have done y". Then, the programmers can make the appropriate modifications without having to spend so much time analyzing whatever else is going on in the thread.

 

The first question is what to do with: KQxxx, Kx, Qxx, Kxx

when the auction goes: 1♠-(p)-2♥-3♣-?

 

I think GIB should pass, with a minimal opening that doesn't want to rebid spades, raise hearts, introduce diamonds, bid 3NT or double. If this hand doesn't pass, what hand would?

[Rain]

Thanks all for highlighting this.

 

DOPI is a bug. It won't make it to this soon to be released patch, but very likely next patch.

 

KX heart raise is possible. GIB has always valued honor holdings like these more. This is evaluation and judgement, not a bug.

 

There are plenty, perhaps an unlimited number of GIB issues that can be worked on. We have to triage to get the most value out of our fixes.

That means the smallest amount of time that can improve the most problems. Some bugs are easier than others (like this DOPI one is probably easy, though I'm no developer) so even if there are more blatant errors, they get fixed faster.

Most are not as easy. I understand GIB's initial code was really arcane.

 

 

----------------

People deliberately misleading GIB in MB$ games and developers being "lazy" to solve old problems:

 

First, MB$ is currently the only robot "tournament" with human/bot vs human/bot. (GIB rental is usually 1 human 3 robots or 2 humans partnering against 2 robots, much less competitive.) A change to what is suggested would only benefit this group right now.

So although it's not impossible that such a tweak would occur soon, it's really unlikely given that important things like doubles sequences still need so much work.

 

Secondly, players like xxhong and most of the contributers in this GIB subforum play much more than usual. You have therefore more experience and more opportunity to see undefined or weird sequences. Judging the robot based on 1 bad sequence is inaccurate.

 

Thirdly, we HAVE already begun on the next GIB patch. That one is very likely to include some fixes to GIB logic so it doesn't go crazy when people lie to it. I do not know what will be included, but it's highly unlikely to be anything specific to human/bot vs human/bot. We'll know when next patch release is near.

 

Finally, how many developers do you think BBO has? They're trying the best they can. It's almost a full time job just reading through all the bug reports from various sources AND separating the actual problems (bugs, evaluation/judgement, logic) from unhappy QQ. (Not trying to diss anyone, BBF currently is the higher quality source for bug reports).