| |
BGonline.org Forums
A new PR analysis mode that considers the 'difficulty' of each roll/position
Posted By: Frank Berger In Response To: A new PR analysis mode that considers the 'difficulty' of each roll/position (sebalotek)
Date: Monday, 4 September 2023, at 1:28 p.m.
AI was one area of interest for me since the late 80s, so it is not that I'm sceptical "to all those new stuff", I just have a differentiated look. I prefer to think than to outsource that. ChatGPT&Co gains sometimes impressive results and sometimes rubbish, but always well formulated.
E.g. the timing aspect is irrelevant and a platitude. Naturally it is more difficult to have a position when you have only 2 seconds on the clock. But is it a reasonable measure of difficulty of a position? Oh this position is difficult, but obly if you have less than 5 seconds on the clock?
A position is quite difficult because you have many moves to choose from. At least that is true to a certain degree (you might overlook a move with a 44 etc more easily. There is a whole book on playing doubles), but in the recent Vogl book (highly recommended) roughly half the Genius failed for some positions when only 2 or 3 moves where possible. So that seems not to be the a real good candidate.
I discussed this with Tim some month ago. IIRC Tim suggested to analyse a lot of matches and see empirical where people err and train a net on that.
Another idea was to look at the heatmap of the resulting position and it might be difficult if the standard deviation of the resulting moves is very different (this thought is very unfinished or I don't remember correctly)
But keypoints remain - difficult for whom bot/human - difficult for Beginner or Expert
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.