UX for language user interfaces
What you’ll learn
Section titled “What you’ll learn”A language user interface (LUI) is a fundamentally different interaction surface from forms and buttons, and the conventional UX toolkit applies poorly. The source curriculum is the Full Stack Deep Learning LLM Bootcamp (Spring 2023), by Charles Frye, Sergey Karayev, and Josh Tobin, freely available at fullstackdeeplearning.com/llm-bootcamp with recorded lectures on the Full Stack Deep Learning YouTube channel.
You will explain why a LUI is a new interaction surface; apply the five core patterns (streaming with thinking indicator and Stop button; citations that turn claims into evidence; regeneration and optional branching; hedging when context is thin; recoverable failure with legible messages, recovery actions, and preserved input); identify and fix common LUI UX issues with the patterns and supporting details; explain why hedging trades single-answer authority for long-term trust; and recognize that recoverable failure is multi-modal (timeout vs retrieval miss vs refusal vs malformed output vs rate limit each need their own message and action).
§6 framing note: taught as interaction-design throughout. Content-policy questions (what the model should refuse), moderation, labeling AI-generated content, and similar policy debates are real but out of scope here; same technical-not-policy discipline as elsewhere in the fleet.
Where this fits
Section titled “Where this fits”This is lesson 6 of 11, the third lesson of Phase 2 (building production apps). It is the UX layer the lesson-5 walkthrough deliberately deferred, and it threads back to lesson 2 (latency) and lesson 4 (citations from source-carrying retrieval). The next lesson closes Phase 2 with LLMOps, the operational discipline that keeps all of this working in production.
Before you start
Section titled “Before you start”Prerequisites: lesson 5 of this track (the walkthrough that named what UX defers to this lesson). Familiarity with one chat-UI implementation (Streamlit, a React/Vercel AI SDK chat, or similar) helps but is not required; the patterns are framework-agnostic.
About the math
Section titled “About the math”None. This is an interaction-design lesson; the patterns are described by what they do for users and what they require in the UI, not by formulas.
By the end, you’ll be able to
Section titled “By the end, you’ll be able to”The single capability this lesson builds: design language-user-interface UX patterns (streaming, citations, regeneration, hedging, recoverable failure) that make LLM applications usable. Concretely, you will be able to:
- Explain why a LUI is a new interaction surface
- Apply the five core patterns (streaming, citations, regeneration, hedging, recoverable failure)
- Identify and fix common LUI UX issues with the patterns + supporting details
- Explain why hedging trades single-answer authority for long-term trust
- Recognize that recoverable failure is multi-modal (timeout vs miss vs refusal etc.)
Time and difficulty
Section titled “Time and difficulty”- Read time: about 12 minutes
- Practice time: about 10 minutes (critique a real-feeling LUI UX against the patterns, plus flashcards)
- Difficulty: standard (no math; the work is internalizing the five patterns and the multi-modal-failure mindset)