Chess Chats #3 | Custom AI Bots on the Chessnut EVO - My Take!
#chesschats #chessnut #chessnutevo
Welcome to another episode of Chess Chats, where we discuss all things about chess!
Today, I wanted to talk about the Chessnut EVO feature where you can train chess neural networks based on game data, and create a customised AI-based bot, using Leela Chess Zero. I have a few personal thoughts about this that I wanted to share with you, separate to the Chessnut EVO prototype review and demonstration video of the feature.
Firstly, I think that it's enormously impressive that it's working at all. As I showed in my video, after training on a small number of my games, two hundred, the vitualis bot played my usual openings for both White and Black, and these openings are unusual enough that it's unlikely that the engine just played them as a fluke. Engines won't usually play gambit lines unless explicitly programmed to do so.
Chessnut needs to leverage this amazing system beyond individuals submitting small numbers of games for custom bots. As Chessnut owns and controls the backend, they should create a suite of custom bots that Chessnut EVO users can download and try.
For instance, imagine the Chess World Champion series of bots! Individual bots trained from the game repositories of the first world chess champion, Steinitz, to the new champion Ding Liren! This would be amazing!
Or opening specific bots! If I wanted to learn to play against, let's say, the French Defense, I could load up a French Defense playing bot! Or, against the Open Sicilian! The opportunities are almost endless!
Or imagine, user submitted bot battles! Users can submit a bot that can only be trained by their own games and these flawed and far from perfect bots, battle to see who would win!
However, I am concerned that Chessnut's planned subscription model for people to access training is going to undermine this amazing feature. I am concerned that they are poisoning the well. Simply, unless there is a critical mass of users and an ecosystem around these custom AI models and bots, I highly suspect that most user won't pay for access as they will not see the value. My view is that the ability to train custom AI models is by far the most innovative feature of the Chessnut EVO. All the other features of the EVO - being able to play online games, being able to play in-built bots, being able to play over-the-board and record the moves, being able to use the board for learning and training, squares coloured to show the best moves - have existed before, though potentially with a different technical solution.
My suggestion to Chessnut? Think carefully about your planned subscription model as I think most users won't subscribe. Consider using the monetisation model that some of the image creation AI services use. For instance, all users can download the AI models, but perhaps some premium models are for subscribers. All users can train new AI models, but perhaps in a more limited fashion (e.g., limited number of times per month with more limited options and features). Consider upgrading the Chessnut App for the Chessnut Air and Chessnut PRO so that the custom AI bots can be used to increase the user base for this service.
Anyway, what do you think about the customised AI-based bot feature on the Chessnut EVO? Please put down your thoughts in the comments. I hope you enjoyed this video and thanks for watching!
