Thanks for your comment of support. Ironically the "groupthink" is hallucinating explanations about how LLMs work, irony. Anyway, yes, the "web" in general is

Thanks for your comment of support. Ironically the "groupthink" is hallucinating explanations about how LLMs work, irony. Anyway, yes, the "web" in general is full of garbage and propaganda. But then again we're exposed to the same web, and without the pre-processing which OpenAI etc. apply to filter out certain words, phrases, topics (including lots of manual human labor). So it's funny how we're living on that same input, but we're so confident we know what's true and what's wrong. But LLMs aren't just trained on literally anything on the web. For example consider Wikipedia. That's not one's average drivel for sure. And next to the general web, you need higher quality sources of data in the mix to bias the AI for truth. Scientific papers, good books, transcripts from conferences and presentations. Wikipedia to mention again. So on. Of course, sometimes LLM make technical mistakes which you can clearly rule as incorrect. Say, plumbus.com's API never had a method getDescription(). But the AI saw many APIs of services like this, which have API getDescription() and given its small network, it's forced to fill-in gaps by inferring the answer based on a vast assortment of contextual clues. Which we call hallucinations (incorrectly). But even the correct answers are inferred. It's all inference. And we all think by inference. We just have 50 times more capacity than the biggest LLM at the moment, so our inference tends to be much more sophisticated, and therefore more likely to be correct. I mean in theory. After all, it's humans and their big brains who produced the garbage that's the web, so.