LLMs are trained on reddit/quora corpuse as well (correct me if wrong). So do the ‘number of upvotes’ on a reply/answer was considered as a parameter or feature during the training?

Also it’s not just about just reddit/quora but is answer realibility factor which is most case are its upvotes are considered?

Or being a “language model” it does evaluate and find similarity itself and chose what to retrive, might be the reason we see hallucinations.