MereGurudev

Yes, they're distinguished: the pieces you place there manually ("fixed pieces" are opaque and thus should have the same color, off-white or dark-grey, no matter the background square.

The pieces the AI adds when you press generate, are translucent (and as such will be affected by the background color). Then you can press "Lock in" to convert the generated pieces to fixed pieces, in which case they will become opaque.

I realize that for anyone less used to the color scheme than me, this distinction is problematic since there will be no less than six different piece colors (white, black, translucent-white-on-light, translucent-white-on-dark, translucent-black-on-light, translucent-black-on-dark)

Thanks for the feedback! It's a simple modification to guarantee things like this – for this prototype I thought it interesting though to expose the "raw" output of the neural network, including crazy things like sometimes placing two kings of the same color. I even force the neural network to place more pieces than it intends, if you put the piece slider on max, to see what happens (and often the result is very bad)

Thanks! Ah, sorry for that. Just to clarify, do you mean for the semi-transparent pieces (the generated ones without "locking in") or even the manually placed pieces?

Hi, thanks for trying Noctie!

If you did the rating test, Noctie tries to adapt to your strength while you're playing, so if you play at 2574 level for a while, eventually Noctie will also play at that level. Since Noctie has no idea about your rating when the game starts, it might be that it took some time for the rating to adapt and therefore you found that the AI played much weaker.

The max strength of the AI is about 2700–2800 FIDE (level "Queen 4" inside the app if you have an account).

Per this site, https://chessgoals.com/rating-comparison/#lichessotb, 2809 Lichess is equal to around 2550 FIDE so if that's your lichess rating (wow btw!) maybe Noctie wasn't so far off. (EDIT: Ah I see, that's not your rating, sorry)

Obviously, you might get a different result next time – one game is very little information to make an accurate estimation off of, especially when the AI has to adapt it's playing strength as we go.

I don't use accuracy for the rating estimation BTW, I use custom neural networks that observe patterns in how humans at various rating levels play chess.

Hey, you're right, adjust piece count is a local modification, so opening the same dreamId in multiple windows would let you view the position with different piece counts (and the link to Noctie uses the local state)

Thanks! Good idea, I could add FEN / PGN export in addition to lichess. I haven't tried conditioning the network on opening but given my training data, that should be possible. For now, you can simulate openings somewhat, by trying to fix pawns and pieces in their typical positions, but it would be better if I could do it like you suggested "QGA, 15–20 moves" etc. I might try this out!

Hey, that's really cool, will check it out! The backend was experiencing some issues leading to laggy play by Noctie, I hope it's working better now.

Thanks! Yes, the traffic from here exposed a few issues & bottlenecks in the backend infrastructure. I have resolved most of them and it seems to be running OK now (?).

To answer your questions: Noctie takes every move into account. Especially in short games or very one-sided games, a lot of the rating estimate will come from your opening play. If you play a bad opening, Noctie adapts and you will get an easier game, which may lead to simpler positions where you have less risk of making a mistake that would lower your rating. So what opening you play definitely has a big impact when trying to judge the rating from one game only.

If you create an account and log in, Noctie will accumulate information from several games to give a more balanced difficulty and more accurate rating estimate (and show how it changes over time).

Noctie's involvement isn't strictly required for the rating estimate although it creates good conditions for it by making the game balanced, not dominated by time trouble or trying to flag the opponent, etc.

Good suggestions, thanks.

Thanks for the valuable feedback!

If the opening is rated as sufficiently bad, it would adjust its play accordingly and perhaps not challenge you sufficiently to be able to change its rating estimate. I.e. if you get a very easy, tactical game after that, the game might not be testing your abilities enough to revise the estimate from one game.

I think you were just unlucky with the opening! When I'm testing it's about 60/40 1...e5 vs 1...c5. The first opening moves should be quite close to what a 1600 FIDE player plays in online blitz.

Sorry for the bad experience, Noctie got stuck for a while due to a bottleneck in my cloud setup

Sorry for the bad experience, I had a service outage for a while

Perfect break entertainment

It seems to be working again. It looks like something wasn't working out with my cloud autoscaling setup..

Sorry, it seems to be very slow right now due to some quota issue :( I'm debugging right now.

It seems one of my services has hit a quota, trying to find out which..

Thanks!

Thanks! For board representation and legal move generation I use chess.js in the frontend, and shakmaty in the Rust engine.

OK, sorry for that, the part with back/forward is very useful info for debugging though. Do you mind sharing what browser / setup you're using?

Thanks for sharing your experience! It's a very interesting point about how you adapt your play based on the preconception about how computers play. The AI isn't programmed to blunder per se, it's more that I have tried to emulate a human intuition & thinking process – which is prone to blunders. So it never intentionally makes a mistake to test you, or to even the game out, it just tries to emulate your level of play.

Thanks a lot, and happy to hear about your experience! Yes, the review mode is quite basic at this point but I hope to add more features soon, including some that are unique when compared to chess.com.

On the tech side, what Maia is doing I think is similar to our NN component. It turns out to get realistic play (include randomness, avoid too many one-move blunders, same level of tactical play as positional play, realistic endgame play, etc.) you have to do quite a lot on top of that – much more than I initially thought. I have tested the AI over the last year with friends of different strengths to try to make the gameplay more realistic and engaging. I also think the automatic adaptation to your level of play, and predicting the rating, is unique. Bad gyal AFAIK uses a different training method where Stockfish evaluations is involved, but I haven't really tested it enough to know how it stands vs. Maia.

Hey, sorry to hear that, the moves should usually be played very fast in the opening, and only up to 5–10 seconds in tactical situations, so I think you might have encountered a bug. Noctie does almost no calculation, just like a human, BTW.

Was this in the rate-me page or as a logged in user? You can mail me with the e-mail in my profile if you'd like to help debug!