LCZero starts with an opening disadvantage..................

[Chessqueen]

Check out the last game that Lc0 beat SF in at TCEC ... incredible opening play. Totally destroyed SF by giving up a bishop and then going down 3 pawns so that it could keep the rook on h8 out of play for enough time to create a winning advantage. I have never ever seen an engine that can play the opening part of the game so well. Usually move like Bg6! are novelties that are discovered after hours and days of patient analysis by strong GM's guiding their engines as they go through the possibility of it being playable or not. Lc0 played this over the board.

What program has a worse bug LCZero or Stockfish, in game 29 out of 100 games Stockfhish calculate checkmate in 22 moves, and Leela saw the game as winning, take a look at the result and tell me what program has a worse Bug ?

Sorry, but what are you talking about? If you have a look at the TCEC archive and the eval graph of game 29 you will notice that both engines see Leela winning from at least move 32 on. In the video the chess puzzler mentions that SF calculates a mate in 22 (for Leela of course!) while Leela just knows that she's winning but that is not surprising.

The missed win of Leela in game 65 is another story. That may be called a bug. Or maybe not, I don't know enough about the NN stuff to make a judgment.

./Andreas

Correct both programs see that Leela is winning, but none predict in how many moves, according to Stockfish it was mate in 22 which did not happened in 22 moves, but a win is a win for LCZeroI suppose. The reason why Stockfish saw a mate in 22 moves and Leela did NOT is because Stockfish has a 7men TB and Leela only has a 6menTB, but the person that was narrating saw it as a bug from Leela part of course both programs saw Leela winning.