I’d be curious to use them on larger data sets. Log files, for instance. High volume, might be low value per line but not much cost per token if you already have a gpu to interpret them using a LLM.