on Toshareproject.it - curated by Bruce Sterling
*ChatGPT was not designed to play chess, so it’s impresive that it can even pretend to do that. You won’t learn much about chess by watching this, but it’s an interesting insight into ChatGPT.
“It lies and cheats all the time!!” Well, yes, it does, but not to its own benefit; it’s just building statistical bridges across problem spaces.
*By contrast, by watching AlphaZero play chess, I actually learned something about chess. I learned that human players tend to concentrate hard on knotted problem areas of the chessboard, whereas you can actually win handily by considering the entire territory of the board at all times.
*Also, there’s no reason to win convincingly, dramatically and crushingly, except to impress other humans. Subtly winning with an inevitable one-pawn advantage, that’s just fine.