Verified Data-Driven Prompts for Guided Large Language Model Queries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
User interfaces lacking graphical components and relying solely on chatbots can confuse users unfamiliar with natural language interactions, leading to frustration and inefficient use of applications.
Innovation Solution
A custom prompt is generated using a large language model to predict actions on user data, verified to achieve a specific goal, and displayed as a data-driven query in a chatbot interface, allowing users to access initial and detailed results without executing the prompt.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a chat bot is used as the user interface without graphical components, then the application can provide natural language interaction, but users unfamiliar with chat bots may become frustrated and unable to effectively use the application
Solution Approach 1:
The system pre-generates custom prompts with data-driven queries and initial results before the user interacts with the chat bot. These prompts are tailored to the user's specific data and role, providing immediate value and guidance without requiring the user to know how to formulate effective natural language queries.
Solution Approach 2:
The system introduces an intermediary layer that translates user roles and data into customized prompts with pre-computed results. This intermediary acts as a bridge between the chat bot interface and the underlying data, providing users with relevant information without requiring them to understand complex query formulations.
2Productivity
If pre-configured responses are used in the chat bot, then the bot can respond to limited questions, but users may become frustrated when their questions fall outside the pre-configured set
Solution Approach 1:
The system dynamically changes the parameters of responses by generating custom prompts tailored to each user's specific data, role, and context. Instead of using fixed pre-configured responses, the system adapts the query parameters based on user characteristics and data availability, enabling versatile responses across different question types.
Solution Approach 2:
The system enables self-service by automatically generating customized prompts and initial results based on user data without requiring manual configuration. The chat bot autonomously formulates data-driven queries and provides initial results, allowing it to handle a wide range of questions adaptively while maintaining fast response times.
3Measurement precision
If the large language model executes the custom prompt immediately, then the most current results are provided, but the user experience may be delayed and resource-intensive
Solution Approach 1:
The system performs preliminary action by generating initial results from the custom prompt before the user actually selects or interacts with it. These pre-computed results are stored and can be immediately displayed when the user clicks on the custom prompt, eliminating the need for real-time execution and providing instant feedback.
Solution Approach 2:
The system performs partial action by generating only the initial results in advance rather than executing the full custom prompt with all its computational steps. This allows the system to provide immediate preliminary results while reserving the option to execute the complete query later if the user requests more detailed or updated information.
Data Source
AI summary
A custom prompt is generated for a user of an application that illustrates a first user experience of a generative artificial intelligent (AI) feature incorporated into a chat bot of the application's user interface. The custom prompt includes a data-driven query verified to produce results on the user data. The generation of the custom prompt uses one or more large language models to predict an action that can be performed by the large language model with the user's data. When the action is verified to produce results on the user data that are on point with a given goal, the action becomes a data-driven query of a custom prompt. A summarization of initial results of the custom prompt are displayed with a “click all results” option. When the “click all results” option is selected, updated results are generated in real time by the large language model