OpenWebUI and similar ChatGPT-lite interfaces are a bad UI/UX for LLMs. "Normies" should not be using local models and those who want local models should be exposed to the settings/complexity that projects like oobabooga/sillytavern/comfyUI force a user to educate themselves about (i.e. LLM sampler settings, chat templates, etc). It's telling that the best AI generated content is made by people using these prosumer style projects, and not using simplified trash like OpenWebUI.

The proliferation of bad UI in consumption of AI systems has done unfathomable damage to discoverability of LLM features, especially new innovations in LLMs like far better sampling algorithms (i.e. min_p). Users are massively harmed by the features they don't know about and wish they were using but don't/can't because they are shepherded into the most brain-dead simple UI possible.