totally get your point, until the language model is giving us outputs for input prompts, it’s do-able.