what prevents LLM data from being poisoned by sheer quantity of garbage?
if they're crawling the internet for data to be fed into the LLMs doesn't that mean that data that appears more will have more significance in the data, instead of data that is better?
i.e. what is the "pagerank" of LLMs?