If you use LLM stupidly, sure. You can get from the LLM pseudo-probabilities of next symbol and use e.g Bayes rule to combine the information of how well it ma

If you use LLM stupidly, sure. You can get from the LLM pseudo-probabilities of next symbol and use e.g Bayes rule to combine the information of how well it matches the page. You can also report the total uncertainty at the end. Done properly, this should strictly improve the results.