It feels like you could create a cool workflow from low temperature creative association models feeding large numbers of tokens into higher temperature critical reasoning models and finishing with gramatical editing models. The slickborns will make the final judgement.