lichess.org
Donate

A new idea for a puzzle book

I was wondering whether engine could be used to identify positions in databases that would offer different type of puzzles, based on engine parameters and their evaluation of move lines flowing from those positions.

Then different classes of engines, like some self-learning engines (the whole emerging ecosystem of them, once people figure out what parameters do, besides affecting ELO ratings of bots).

Focusing perhaps on positions where two classes of engine differ (human learning: A/B search, versus machine learning: some training set prior experience or dependence). Then I wonder the kind of teaching that this would provide for the humans. What kind of play style would be taught one way or the other. Of course, the evaluation should be up to a given tolerance criterion in variation as depth or node cardinality increases, so as not to do infinite searches, yet use the full strength regardless of compute efficiency of the two approaches.

Are there such positions BTW? Anybody tries to find such contrasting positions... I think there is a "filon" of puzzle making there. sorry if off topic. that is the closest I could find, there is even a bot post... if already done or discussed, I welcome pointers.

This topic has been archived and can no longer be replied to.