[diana_eva]

Hello BBO Community!

We're pleased to announce an innovative step in the evolution of our online bridge experience: meet Ben (the Bridge ENgine), our newest AI-powered bridge robot. Developed by the talented Lorand Dali, Ben started as an open-source project and has now found a home here at Bridge Base Online (BBO). Click here to read all about Ben on BBO and stay up to date with its latest improvements.

What's Happening?
Lorand Dali, a name many of you might recognize, has joined the BBO family. Over the past few months, he's been diving deep into BBO's vast data reservoirs to train various models of Ben. Blending machine learning with traditional robot coding, he is pushing the boundaries of how bridge robots play and learn.

Innovation Week (January 2024): Initial release of "Ben on BBO"
Mark your calendars! From January 8th to 14th, we're rolling out a special event on BBO. It's your chance to "preview" and play alongside one of the "Ben on BBO" models. This particular version of Ben has been trained on a whopping 100 million hands played in ACBL pair games right here on BBO. Click here to read all about it.

Ben & Friends Daylong: A Daily Bridge Adventure
Each day, Lorand is hosting the "Ben & Friends" daylong game - a unique opportunity for BBO staff and friends to interact with different versions of Ben. If you'd like to join in, drop Lorand a line (he is lorserker on BBO) and he'll add you to the list!

Let's Discuss and Develop Together
This thread is your go-to place for all things Ben. Share your experiences, discuss notable hands, provide feedback on the games featuring Ben, and stay updated on the latest developments in our AI and robotics endeavors. It's also a direct line to Lorand and the BBO team, as we continually work to enhance your bridge-playing experience with AI technology.


Let the games and discussions begin!


P.S.: In case this message seemed unusually witty, thank ChatGPT, our AI assistant. It seems that Ben isn't the only one learning new tricks at BBO. We're embracing AI, one post at a time!

This post has been edited by diana_eva: 2024-March-15, 13:32
Reason for edit: Added link to "About Ben on BBO" changelog

[pilowsky]

diana_eva, on 2024-January-05, 08:15, said:

P.S.: In case this message seemed unusually witty, thank ChatGPT, our AI assistant. It seems that Ben isn't the only one learning new tricks at BBO. We're embracing AI, one post at a time!

The end of the world is nigh.

[pescetom]

pilowsky, on 2024-January-05, 15:39, said:

The end of the world is nigh.

When Chatgpt fixes it's first Gib programming bug I will start to be impressed.

[helene_t]

http://tinyurl.com/yro9mhxe

So Ben has void-AJTxxx-x-AQTxxx and bids
1♥-(1♠)-2♥-(2♠)
4♥-(4♠)-5♥-(pass)
6♥

According to the hover info, the 4♥ bid is 18-21 HCP so presumably we are in a FP (especially as we are red/white). 5♥ is just 6-10 and 3+, doesn't sound like a slam try. So 6♥ is not consistent with 4♥ and it looks like Ben should have made a slam try instead of 4♥.

Maybe it's not a big deal as it could be that LHO's 4♠ bid made the sims more optimistic. But I wonder if the 5♥ hover info is accurate, or if no appropriate slam try was available on the previous round.

Sorry for posting such a small issue but I haven't run into any big issues so far (other than the transfer mishap in the other threat).

[helene_t]

http://tinyurl.com/ysng3r88


A lead-directing double on a small doubleton.

Partner didn't lead the suit anyway so maybe it's the hover info that is wrong but lead-directing would be the normal meaning here.

[helene_t]

http://tinyurl.com/ynxlmnk4

Ben East leads ♠4 from KT4 at trick 3, with 87xxx in dummy on its right hand. This blocks the suit and allows declarer to make. I think the layout should be unsurprising, so maybe Ben is forced to make the "systematic" lead? Or was it just unlucky with the sims?

[pescetom]

helene_t, on 2024-January-06, 07:34, said:

lead-directing would be the normal meaning here.

Really?
I would have thought punishment or some Zia-like choice, North is unlimited and EW have admitted they are bidding on LoTT and not much else.

[helene_t]

pescetom, on 2024-January-06, 11:47, said:

Really?
I would have thought punishment or some Zia-like choice, North is unlimited and EW have admitted they are bidding on LoTT and not much else.

Yeah I see your point. That's not compatible with the actual hand, though. The only think I can think of compatible with the actual hand is "I suggest we sac". I doubt that's the robot system, though.

[kgr]

What play agreements is Ben using? leads, discarding,...
Is it using more sophisticated signals than Gib?

[lorserker]

kgr, on 2024-January-08, 10:37, said:

What play agreements is Ben using? leads, discarding,...
Is it using more sophisticated signals than Gib?

leads top of sequence, top of doubleton, top of nothing.
when leading from length it tends to lead 4th best.
doesn't signal.

in the future we'll probably add something more specific.

[kgr]

lorserker, on 2024-January-08, 11:21, said:

leads top of sequence, top of doubleton, top of nothing.
when leading from length it tends to lead 4th best.
doesn't signal.

in the future we'll probably add something more specific.

Tx!
(I think Gib is often going wrong because it is not signaling. Adding that would improve Ben)

[helene_t]

Yes but on BW there was a thread about WB5 being exploitable because it always gives honest count, even if it can only help declarer. It may not be so easy to program when it should give honest count and when it should just play random spots.

[smerriman]

Is Ben capable of improving on GIB's system, or does that fact that it learnt from GIB mean it is stuck with the same system limitations?

---

I wouldn't open 1NT normally, just playing around - though doubling to show stoppers without any at all and then following up with 2NT feels like a bad habit it picked up from GIB

[smerriman]

Also, I'll post the hand after the tourney is complete (you'll probably see it yourself), but on the latest tourney, Ben responded to Blackwood to show an even number of keycards, and a void.

Unfortunately it didn't have either of those things

[fuzzyquack]

http://tinyurl.com/yqskmzjc
Raising M opening to 3 after 2/1 action (1S-2H/2S-3S) on xx is way too creative

[fuzzyquack]

http://tinyurl.com/yqskmzjc
Raising M opening to 3 after 2/1 action (1S-2H/2S-3S) on xx was quite creative

[lorserker]

helene_t, on 2024-January-08, 21:34, said:

Yes but on BW there was a thread about WB5 being exploitable because it always gives honest count, even if it can only help declarer. It may not be so easy to program when it should give honest count and when it should just play random spots.

I was thinking to start a thread to collect your thoughts about how ben should signal when i start working on it.
But if you have ideas, I am listening already now.

[lorserker]

smerriman, on 2024-January-09, 01:35, said:

Is Ben capable of improving on GIB's system, or does that fact that it learnt from GIB mean it is stuck with the same system limitations?

---

I wouldn't open 1NT normally, just playing around - though doubling to show stoppers without any at all and then following up with 2NT feels like a bad habit it picked up from GIB

Ben learns the bidding from gib, but can interpret it differently afterwards.
Learning the bidding from gib is not, in principle, a cap on the potential. True that starting from something better will probably result in something better.

According to some experiments I did, when judging the bidding alone*, ben is better than gib basic by almost 1 imp/board and is also slighty better than gib advanced. This results I have confirmed also by training ben to bid like argine and then comparing it with argine.

*the way i am comparing only the bidding is by playing a match between a team of bens and a team of gibs.
after the bidding, the opening lead is made single-dummy. the rest of the card play is skipped and the result is determined by the double dummy score of the hand. (we do this because different bots have different ability in card play and we want to isolate the bidding).
if you have an idea of a better setup to test only the bidding, i am happy to listen.

[lorserker]

fuzzyquack, on 2024-January-09, 02:14, said:

http://tinyurl.com/yqskmzjc
Raising M opening to 3 after 2/1 action (1S-2H/2S-3S) on xx is way too creative

It seems quite clear that North should bid NT instead of 3S. Thanks for raising.

[kgr]

This 2S seems not optimal to me.
http://tinyurl.com/yrf6btz7
Pass if no DBL seems better, and RDBL if DBL.