Why offer a solution then? Seems fairly easy for google to avoid giving the final number?
That's not how LLMs work. These systems take a stream of tokens, do some linear algebra to find a stream of tokens within the parameters of their vector similarity, and spit out the result.