Ah this is a great find, the bullet chess clock. It's partly inference time before the intentional delay, and partly the animation of the move counts towards the bot's time. Should be fixed now though.
This approximates 2800 level bullet better now but overall things are actually worse. It's impossible to flag even the easiest bot, you can easily get it to exactly 0.2 seconds and then it will just play instantly and never lose time again. Even if you have 50 seconds you can't flag it. It has to lose a nonzero amount of time per move on average for it to be fun and realistic.
http://chessbook.com/training-games is the direct link, for usual Chessbook users it's still gated behind an early-access flag since I'm still working on the UX of playing the bot.
> Determining who has control of a square in chess is a pretty interesting problem. Is it the person with the most material aiming at that square? [...] The formula I decided on here was that each piece contributes the inverse of it’s piece value, squared.
This is the most interesting part for me in part because it seems arbitrary.
Back to the recent post:
> shapes the trunk; never feeds the policy logits
Fixed the premove thing too, I think.