Interesting, is the implication that the sampler makes a big effect on both prose style and coding abilities? Hadn't really thought about that, I wonder if eg. selecting different samplers for different use cases could be a viable feature?
In terms of raw prose quality, I'm not convinced GPT-5 sounds "less like AI" or "more like a friend". Just count the number of em-dashes. It's become something of a LLM shibboleth.