DeepMind’s SCoRe shows LLMs can use their internal knowledge to correct their mistakes