Could we use LLMs to give a readability score to codes?

If a dumb LLM can understand a piece of code, then maybe that's one way to check if it's a sensible code.