Output Medium Transformation
A completion is text by default, but transforming it into the right output medium — speech, a UI event, a diff — is part of closing the application loop.
A model's completion is a blob of text. Sometimes that's the end product directly (a simple chat reply), but closing the LLM application loop — transforming the completion back into the user's domain — often needs further work:
- Structured extraction — asking for output in a specific format (tabular data, a fixed schema) and parsing it out of the raw completion text.
- Function calling — the generated text represents a function call rather than a user-facing answer; the application executes the real action (look up flights, or something with real-world effect like purchasing tickets) and returns the result to the user in their domain. See function calling for the mechanics.
- Medium changes — text-to-speech for phone-based assistants; UI events for applications with a complex interface; or even a presentation change within text itself, such as showing a code completion as a grayed-out inline suggestion accepted with a keystroke versus showing a code change as a colored diff.
Which transformation is right depends on where in the user's actual workflow the output needs to land — text is the model's native output, but it is rarely the application's actual deliverable.
Sources
- Prompt Engineering for LLMs (John Berryman), ch. 4