Generative Model GUI Layout With State Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing generative models like LLMs have sub-optimal user interfaces that make it difficult to explore or refine multiple topics due to transient state history, leading to increased user interaction and computational costs.
Innovation Solution
A graphical user interface (GUI) for generative models that uses schema prompts to guide responses, dynamically updating based on user interactions and preserving state between sessions, reducing the need for re-prompting and improving organizational layout.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a linear chat-style interface is used for generative models, then the interface is simple and easy to implement, but the user experience deteriorates as previous dialog turns disappear off-screen and users must scroll back to view previous responses
Solution Approach 1:
The patent transitions from a linear one-dimensional chat interface to a two-dimensional grid layout where responses are organized in rows and columns. This dimensional change allows users to view multiple previous responses simultaneously without scrolling, improving ease of operation while maintaining manageable interface complexity through structured organization.
Solution Approach 2:
The interface segments the continuous chat dialog into discrete, organized blocks or tiles representing individual responses. Each response is separated into its own visible unit within the grid, allowing users to access specific responses independently without navigating through a continuous scrollable list.
2Ease of operation
If a chat-style interface is used, then the interface is simple, but it becomes difficult for users to find and refer back to particular LLM responses
Solution Approach 1:
By organizing responses in a two-dimensional grid rather than a linear sequence, users can visually locate and access specific responses more quickly. The grid layout provides spatial organization that enables faster retrieval of particular responses compared to scrolling through a linear chat history.
Solution Approach 2:
The interface incorporates search functionality and visual organization that provides feedback to users about the location and content of previous responses. Users can search for specific responses or navigate through organized grid structures, receiving immediate feedback about available responses without time-consuming scrolling.
3Adaptability or versatility
If users explore multiple items or topics using the LLM during a given session, then users can refine topics, but due to the transient nature of LLM state history, users find it difficult to return to unexplored items in later sessions
Solution Approach 1:
The system performs preliminary action by automatically capturing, organizing, and persisting the grid layout and response structure during each session. This preliminary organization of state information ensures that when users return in later sessions, the structured history is preserved and immediately available for continued exploration without requiring re-prompting.
Solution Approach 2:
The patent creates a persistent copy or representation of the LLM session state and conversation history. This copied state information is stored and maintained between sessions, allowing users to return to previous topics and responses as if the original session were still active, thereby maintaining reliability of state preservation.
4Loss of information
If users re-prompt the LLM to obtain previously generated information, then users can access historical data, but this causes the LLM to forget earlier information and unnecessarily incurs computational costs
Solution Approach 1:
The system performs preliminary action by preserving and organizing the complete conversation history and response structure in the grid layout before the user needs to access historical information. This pre-organized state information is immediately available for reference, eliminating the need for users to re-prompt the LLM and thereby avoiding unnecessary computational costs and potential information loss.
Data Source
AI summary
Implementations relate to graphical user interfaces (GUIs) for interacting with generative model(s). Processor(s) of a system can: receive user input associated with a user of a client device; process, using a generative model (GM), a GM input including the user input and a general schema prompt to generate a GM output; determine, based on the GM output, GUI elements and a specific schema prompt specific to the user input and based on the general schema prompt; cause the GUI elements to be rendered; store a specific schema that has been determined based on the GM output; receive additional user input; process, using the GM, an additional GM input including the additional user input and specific schema prompt to generate an additional GM output; determine, based on the additional GM output, updated GUI elements and an updated specific schema prompt; and cause the updated GUI elements to be rendered.