Ah hah, I didn’t take my own advice to step away and get out of tilt. Immediately after four “known” bad boards and publishing the post, I got back on the wagon of blunders. But it was reasonably fun – for those who are on the fence, come join the fun. You will probably do better than me, and you’ll be doing your part in self-quarantining.
Pressure Is Off
Good news, with each passing online NABC, my score has gotten worse and worse. But at least I finish them in a reasonable time. Not joking, somehow the first time I ever played the online NABC, I got a 61% but took a straight 3 hours to do 24 bot boards. Maybe I should slow down?
Bot Training Thoughts?
So we can’t talk about any specific hands until the day is done – but I swear there is a tooltip bug in the bots. Either that, or the bid is counter-intuitive enough that almost all humans assume a different interpretation, so thousands of past experiences have trained the bots to also expect a different hand from their partner.
For example, and this is completely unrelated to the “bug” I think I found. Imagine the bots were coded to play that opener after Jacoby 2NT always shows shortness, even without extras. However, enough humans bid fast arrival, to avoid the bots taking them too high after a light opening. Over time, do the bots start to build slightly higher expectations for partner’s hand in the shortness showing case?
As a gamer and software engineer, I would be curious to see how much bot bidding is lookup table versus constrained simulations. In the simple case, the bot just follows rules. In the second case, the bot generates thousands of hands based on what everyone has promised in the bidding, then figures out how best to bid.
You’ll do better tomorrow.