Exploring Two-Level Visual Mnemonic Compression — LessWrong