Before or after punching out an encyclopedia?

:D

It's not a million monkeys, it's an infinite number of monkeys, and that infinite is necessary because the odds of a monkey typing out anything legible for a free sentences (let alone works of Shakespeare) are way lower than one in a million.

The saying itself also includes an infinite amount of time, though that isn't necessary, as an infinite number of monkeys will produce every possible combination in the time it takes to type that combination.

But (and it's a big but), even then you just have a relatively tiny number of works to sort out from the massive amount of random garbage. You can be sure that there are groundbreaking papers out there for every subject that would advance technology by millennia, but they are all somewhere among those infinite stacks of paper (from the monkeys that randomly decided to stack each page as it finished, instead of eating it or ripping it up or just continuing typing on nothing, fighting their neighbour, and every other possible behaviour... Those stacks of paper to review themselves will be rare).

Another big but, even if you can figure out a way to automatically sort the readable from the garbage, you'll still have to somehow figure out which one is "Advice for people of 2026 to quickly catch up to technology of 10,000 (as if society never collapses)" or "A practical joke: tricking the people of 2026 into ruining their future even more by claiming they can advance 8000 years of technology (lmao)". There will also be an infinite number of mixes between the two.

The saying is more about the concept of infinity combined with probabilities. With an infinite number of attempts, every single possibility will be seen, no matter how low the odds.

LLMs aren't like infinite monkeys, they use statistics to correlate words and basically go from an average of all the words they've been trained on to a specific case by predicting the most likely next word given the preceeding ones (and statistics are good at talking about the big picture, not so much when trying to predict specific samples, which is why they produce slop instead of descent results like the star trek ship computer, which is based on human writers writing what an AI they imagined would do).