If it's not a rated game then I don't think 'cheating' matters too much against a Bot.
On FICS (free internet chess server), you can run an engine if you don't log in as a registered user, so there is a sort of "anything goes" approach to unregistered users there. There's a certain similarity between unregistered and unrated games.
If someone is running a Leela Bot on here there is a good chance that they are using it for Leela to learn to play chess as well as possible so consultation of opening books, use of engines, use of endgame table bases, etc are only like to help improve Leela rather than hinder.
Worse than cheating would be to constantly play like a Patzer and teach Leela that bringing her queen out early will likely lead to a win. Neural networks are trained to learn what works most of the time and I believe that seeing too many patzer games would be counter-productive.
I heard a story of a neural network that was trained with pictures of American and Russian tanks in order to tell the difference. Unfortunately the pictures of one type were taken on a cloudy day and the others on a clear day. When they tried out the neural network on pictures it hadn't seen before it didn't do well. When they analyzed what had gone wrong, they realized they had programmed a neural network to tell whether it was a nice day!
If it's not a rated game then I don't think 'cheating' matters too much against a Bot.
On FICS (free internet chess server), you can run an engine if you don't log in as a registered user, so there is a sort of "anything goes" approach to unregistered users there. There's a certain similarity between unregistered and unrated games.
If someone is running a Leela Bot on here there is a good chance that they are using it for Leela to learn to play chess as well as possible so consultation of opening books, use of engines, use of endgame table bases, etc are only like to help improve Leela rather than hinder.
Worse than cheating would be to constantly play like a Patzer and teach Leela that bringing her queen out early will likely lead to a win. Neural networks are trained to learn what works most of the time and I believe that seeing too many patzer games would be counter-productive.
I heard a story of a neural network that was trained with pictures of American and Russian tanks in order to tell the difference. Unfortunately the pictures of one type were taken on a cloudy day and the others on a clear day. When they tried out the neural network on pictures it hadn't seen before it didn't do well. When they analyzed what had gone wrong, they realized they had programmed a neural network to tell whether it was a nice day!