I've tried using an LLM to vet training data as "factual or fictional" or "from a informed and reliable source" and didn't find the magic enough setup get get

I've tried using an LLM to vet training data as "factual or fictional" or "from a informed and reliable source" and didn't find the magic enough setup get get results that seemed like they could be used for boosting model quality.