Post #3534486

Not sure if troll or… …ok, I’ll bite. This is maybe the stupidest thing I’ve ever seen. You know that Wikipedia comprises one of the largest publicly-accessible sets of meticulously-curated natural-language data on the planet, right? And so when you’re training up a new model, you’re naturally going to start with that massive, freely-available repository? Every major AI model has been trained, at least partially, on Wikipedia. This insane viewpoint is essentially saying that you don’t think Wikipedia is reliable, but if some linear algebra chews it up for a few minutes, then it’s ok. You’re turning up your nose at tap water, but drinking your inside dog’s urine.