3000+ rated bot

@LeelaChessOfficial
no way to beat this bot

That's because it played against Penguin (who now has 3200 bullet by the way).

yeh

Leela is currently(as far as I know) the strongest existing bot and it(she?)'s only getting better as the time progresses because it has more time to train against itself. The design is different from normal engines- standard engines are algorithms that was written by a human and is easily understood- they analyzes possible lines of play and selects the ones that gives the best end result(check out www.chessprogramming.org/ if you want to make an engine yourself). Leela on the other hand is a neural network, which means that we don't really understand how it decides which moves to pick. It learns by actually playing chess and after playing a lot of games against itself(230 000 000 right now and still going) it is the best chess player to ever exist. Neural networks are really cool and they can be used for a lot of different things- for example lichess uses it to detect cheaters by feeding a neural network a lot of data like the time it takes a player to make a move or their mouse movements. The model was trained by "showing" it how people already caught cheating acted. It learns from that data and can accurately predict if a person is a cheater or not

Part of the reason Leela is rated so high is because they started at 3000 rating.

@LeelaChess is supposed to be "Leela Chess Zero playing full strength." and its rating is "only" 1900 in ultrabullet, 2400 in bullet, 2500 in rapid & blitz and 2200 in classical...

@ZeroSocialSkills there's no neural net involved in Lichess cheat detection, it's just some hand-coded heuristics plus manual review by moderators. See github.com/ornicar/lila/blob/master/modules/evaluation/src/main/Assessible.scala#L79

@polylogarithmique, @LeelaChess's relatively low rating is interesting. This is mostly speculation, but looking over its game history, it seems there are two different things going on here:

1. It's lost a lot of games, even to weak players, by timeout. This is probably a bug in the engine, one that I've encountered myself while testing it locally.

2. The remainder of its losses are almost exclusively to other bots. If you have a bunch of bots that only ever play against other bots, then they're all going to have artificially low ratings since they all start out at 1500 and one bot's gain is always another's loss. So when @LeelaChess loses to one of these it's going to get dragged down as well.

The term full strength isn't meaningful in this context. The strength of Leela depends primarily on how fast the GPU is. Judging by the ratings of this bot, it isn't running on fast hardware. Stronger Leela bots have played on this site before.

Hardware does of course matter but it should be playing better than 2400 regardless. There's only about one order of magnitude difference between the top of the line (Titan RTX) and what you get in a low-end laptop.