Algorithmic Progress in Six Domains — LessWrong