GPT-4.5 Can Play Losing Chess — LessWrong