Join Bridge Winners
All comments by Sam Dinkin
1 2 3 4 ... 21 22 23 24
You are ignoring the author of this comment. Click to temporarily show the comment.
I would be playing my same old conventions as on my current convention card which I made in Excel. LaTeX is a typesetting language.
Aug. 9
You are ignoring the author of this comment. Click to temporarily show the comment.
Yes, it seems best to award at least as many masterpoints to the team that beat the evening's winner in a 4-way in the afternoon as to the evening's winner.

I also agree that both afternoon winners should count as having survived a round.
Aug. 7
You are ignoring the author of this comment. Click to temporarily show the comment.
Je ne lis pas le français.
July 14
You are ignoring the author of this comment. Click to temporarily show the comment.
Computer bridge has a long way to go before it gets to the level of Chess, Go or Backgammon. Take GIB for instance. GIB needs to work on their defense to common bidding methods. E.g., it's defense to 1 Precision is pretty poor. Second, they need to work on their competitive bidding. Their responses and rebids after a top-and-bottom cue bid is not so good. Third, they need to learn more about scouting particular players with regard to how well they adhere to their announced system and carding agreements.

Not knowing opponents' defensive carding agreements, bidding agreements and respective styles will lead to inferior card play and defense, too. The negative inferences available to intermediate-level human players are not available.

GIB also needs to learn to adjust its play style in the face of state of match concerns, form of scoring and recognition that its methods are being countered effectively. That is, GIB needs to learn to adjust its style if its opponents are playing certain methods or just happen to have above average skill at defense. This may mean more sound bidding if their assumptions about likelihood of making a contract at particular strength levels are revealed to be off. They also need to change gears in the other direction to attempt to run up the score against weaker defenders.

They need to learn some face recognition, timing analysis and table feel. They need to sometimes delay to make sure that they don't have a “tell” about what they're thinking about (or not). They need to become expert at voice recognition and psychology to figure out what common explanations mean about the likelihood of it being true or the happiness or unhappiness of the explainer. They would need to learn some about human sociology (smoke breaks) and physiology (late-night play) to temper its view of whether a player is bidding to be dummy or is too sleepy to make the usual play.

GIB needs to be able to learn to call the director and handle misexplanations and hesitations. GIB needs to learn enough lawyering to argue a case before a committee. If GIB wants to play with human teammates, it needs to learn to be a good partner and a good teammate and not be completely mute during the comparison and the break.

This set of tasks looks more like training a really smart generalist program to figure things out.

GIB, you've come a long way, but you're still a baby.
July 13
Sam Dinkin edited this comment July 13
You are ignoring the author of this comment. Click to temporarily show the comment.
They withdrew the assertion that computer bridge players are better than humans.
July 13
You are ignoring the author of this comment. Click to temporarily show the comment.
Any system design book with the Fibonacci sequence is probably worth a read, especially if you're designing a relay system.
July 12
You are ignoring the author of this comment. Click to temporarily show the comment.
Positioning points go back further than deep-seeding positioning points. Seeding points may be better, but wasn't speaking to that.
July 12
You are ignoring the author of this comment. Click to temporarily show the comment.
The best information we have on seeding is the regular positioning points. Once a team has qualified for a deep bye, I believe it's best to use this best information for accurate seeding within the round.
July 10
You are ignoring the author of this comment. Click to temporarily show the comment.
If the three GIBs together are taking a total of less than 1 min per board, then 5 min per board in the GIB individual would more generous than an 8 min/board live event. And I would expect that would still leave more break time in the GIB individual than the scheduled extra breaks in 4-person play. If everyone paused for a total of four minutes in 4-person play, that would take 16 minutes+ per board.
July 6
You are ignoring the author of this comment. Click to temporarily show the comment.
Yes, it would be best that the purposeful misdeal to give you the best hand is UI if the GIBs you were playing against were participants (since it's not only unauthorized to them, but they're programmed to be unaware of it), but they're non-participants.
July 5
Sam Dinkin edited this comment July 5
You are ignoring the author of this comment. Click to temporarily show the comment.
GIB will be happy to double based on values that you've shown. Even after you had a chance to double for penalties yourself.
July 4
You are ignoring the author of this comment. Click to temporarily show the comment.
I can't tell for sure, but I suspect GIB counts extra points for shape–including shortness in partner's suit.
July 4
You are ignoring the author of this comment. Click to temporarily show the comment.
Yes, it doesn't seem to have the intuition that this is a holding that will generally win because declarer will finesse (e.g. stiff king behind longer stronger holding and leader not having shown values). I understand there are tough tradeoffs with the current main solution technique.
July 3
You are ignoring the author of this comment. Click to temporarily show the comment.
Yes. GIB will often lead short honors. Even stiff honors in trump. GIB doesn't like GIB's partner crashing honors. It will also lead small singletons in trump.
July 3
You are ignoring the author of this comment. Click to temporarily show the comment.
Yeah, I had GIB cue bid 3 to show a stopper without a stopper this week. Didn't work that time.
July 3
You are ignoring the author of this comment. Click to temporarily show the comment.
Yes. 2NT-3-3-4-4NT is to play too. Sometimes you just have to read the tool tip. (And heaven help you if it's blank.) When GIB first started playing passed-hand unusual 4NT over 4M, GIB would pass it rather than pick a minor.
July 2
Sam Dinkin edited this comment July 2
You are ignoring the author of this comment. Click to temporarily show the comment.
Arguably GIB's bidding code should be required to be fully disclosed by the Laws of Duplicate Bridge. It's a fine line regarding GIB defensive signalling code which is arguably a defensive agreement and it's choice on how to defend which is not. Similarly, if GIB takes into account GIB's rules of opening leads (which its simulation of what holdings GIB would lead from may do this implicitly), then opening lead code is also part of defensive agreements. Given that GIB is used in GIB tournaments, releasing the entire code (commented hopefully), would allow wonks to identify flaws in GIB's game that could be corrected.
July 2
You are ignoring the author of this comment. Click to temporarily show the comment.
>>It's a different skill to track GIB tendencies and builds some bad habits that cost when playing against top pairs.

>And when playing with a real partner, I would think.

Yes, defending and bidding as if your partner is a GIB would definitely qualify as a bad habit.
July 2
You are ignoring the author of this comment. Click to temporarily show the comment.
I noticed that it's the advanced ‘bots for the tournament. I did a little bit of comparison of the cheap bots versus the expensive bots. The expensive ’bots think longer. They look at more possible deals to try to figure out what hand is consistent with your bidding and play. If you have what you said you had, this means the expensive ‘bots will be more likely to peg your hand and defend or declare better. If you don’t have what you said you had, then the cheaper ‘bots will tend to not notice as much and not do something brilliant or stupid (or both) so the payoff to bidding light or heavy or having an extra card or one fewer card is much bigger versus the expensive bots. The expensive ’bots will rarely cater to a 2NT opener or a 2, 2NT rebid hand with five card major. They will switch more often when they have 9 cards in a suit led against NT. In short, in many situations they are too clever by half. They will “end play” you for down 1 or down 2 and end up taking two fewer tricks because you have an impossible holding.
July 1
You are ignoring the author of this comment. Click to temporarily show the comment.
It's a different skill to track GIB tendencies and builds some bad habits that cost when playing against top pairs. But playing 1000 hands in a weekend against the bots can improve declarer play and help with 2/1 mastery and permit research into holes in 2/1 that also work against “top pairs” playing 2/1.

I didn't follow the ACBL official thread, but I would argue that knowing the ins and outs of 2/1 and GIB's take on it is similar to knowing enough about all the systems you're likely to face in a live tournament. Since there's only one system, you can get into tremendous depth, kind of like having written defenses if you work at it long enough. Furthermore the GIBs rarely adjust. That is, knowing GIB tendencies in bidding is just a special case of good scouting against the general tendencies of pairs that play the system you face and specific tendencies of specific pairs and players.

And knowing the GIBs' play and defense style is like knowing what sorts of plays work against weak opponents and strong opponents, how to falsecard against pairs playing standard count or upside down count or when to swing in a team match or trying to win an open pairs. So again, it's a special case of getting on the same page on defense as your partner and scouting the defensive leaks and declarer-play leaks of your likely opponents. That you play all 48 boards against the same opponents means there's more payoff to the investment–so more like scouting for a KO.
July 1
1 2 3 4 ... 21 22 23 24
.

Bottom Home Top