That makes sense but the principle of "more data = more better" suggests that maybe training an LLM on all the possible data and then fine tuning it to only spi

That makes sense but the principle of "more data = more better" suggests that maybe training an LLM on all the possible data and then fine tuning it to only spit out the best answers might be better than only training it on only the best data to begin with.