Ask HN discussion explores whether an LLM trained on scientific content alone would outperform one trained on a broad corpus including novels and non-fiction for answering scientific questions.

The question posed asks whether omitting novels and non-fiction from an LLM's training data would result in better scientific question-answering compared to an LLM trained on a broader corpus.

The inquiry frames this as testing whether an LLM trained specifically like a scientist would be a better 'scientific' LLM.

The discussion centers on the relationship between training corpus composition and model performance on domain-specific tasks.