They are playing different games -- worstfish is aiming for the quickest way to lose _assuming the opponent is trying to win_. For example worstfish would always make a move that gives its opponent the option for checkmate in 1, even if that same move lets the opponent force worstfish to win in 3. A human could never beat an AI truly trained to try and lose.
You prolly get this a lot but, your parents named you Magnus, they must've consulted some senior monk in the Himalayas or something to forsee your destiny fr
It would be truly impossible to lose to worstfish if it did to the player what you did to it. Force the player to make moves wich eventually checkmate itself.
I agree that would be the best method, though the depth required to realise that is not there yet, maybe a learning algorithm would be better than one that just finds the quickest self-mate.
You've not noticed that Ng7 is covering also e8. You could just go Ng7 much much earlier (at 5:55 it was mate in 2, I'm wrong, Rook had to disappear first but N on c7 is not needed) when King was on f8.
I think the problem is that the optimal strategy to lose a chess game should be to dominate the opponent and then switch gears and force them to checkmate you. The bot is only programmed to find the quickest way to let the opponent checkmate themselves rather than force it, and the opponent does not have to oblige. But having a goal to checkmate the enemy is more simple and the engine is better at doing that, so normal stockfish cannot be beaten by a human.
@@MagnusChessUK so does that mean that there can be a better version of worstfish which forces checkmate by forcing the opponent to checkmate it rather than just finding fastest checkmate