Two LLM UI Patterns That Aren't Chat (poyo.co)

A recent exploration in LLM user interface design introduces two innovative patterns that move beyond traditional chat formats, highlighting the need for structured context in specific tasks. The first pattern, designed for comparison, transforms user prompts into a dynamic table interface where items become rows and questions create columns. This hybrid chat/table model allows users to organize complex information seamlessly, resulting in a more coherent comparison experience without the disjointedness often seen in chat interactions. The app searches for relevant items, auto-generates columns from initial queries, and can update responses based on new insights, making it applicable in numerous contexts such as e-commerce and research. The second pattern focuses on exploration through a tree structure, which facilitates deep dives into sub-topics without confusing overlapping contexts. By allowing independent branching, users can thoroughly investigate various threads without diluting the focus. Each node features its own expansion prompts, providing concrete next steps, relevant questions, or mutually exclusive options, all while retaining essential context from parent nodes. These UI patterns emphasize the importance of tailored context management in LLM applications, suggesting a shift toward more structured interaction designs that improve usability and effectiveness in information retrieval and decision-making tasks within the AI/ML community.