Electronic device and method for evaluating output of artificial intelligence model, and non-transitory computer-readable storage medium
The electronic device evaluates AI-generated responses using attribute classification and evaluation prompts to ensure they meet criteria, addressing the issue of inappropriate outputs and enhancing user interaction.
WO2026116748A1PCT designated stage Publication Date: 2026-06-04SAMSUNG ELECTRONICS CO LTD +1
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- SAMSUNG ELECTRONICS CO LTD
- Filing Date
- 2025-10-02
- Publication Date
- 2026-06-04
AI Technical Summary
Technical Problem
Existing AI models lack effective mechanisms to evaluate and filter responses generated by natural language generation technologies, leading to the potential display of inappropriate or unsatisfactory outputs.
Method used
An electronic device is equipped with a processor that executes instructions to classify user inputs, identify attributes, and apply evaluation prompts to assess the output of AI models, ensuring responses meet predefined criteria before display.
Benefits of technology
Enhances the quality of AI-generated responses by filtering out inappropriate content, thereby improving user interaction and satisfaction.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure KR2025015878_04062026_PF_FP_ABST
Abstract
A processor of this electronic device is configured to: acquire a user input while an assistant application is executed; identify an attribute corresponding to the user input from among a plurality of predetermined attributes for classifying user inputs; identify an evaluation prompt corresponding to the attribute from among a plurality of evaluation prompts for evaluating an output of an artificial intelligence model; identify whether a response to the user input, generated using the artificial intelligence model, satisfies one or more criteria identified on the basis of the evaluation prompt; display the response to the user input, generated using the artificial intelligence model, in response to the identification that the response satisfies the one or more criteria identified on the basis of the evaluation prompt; and refrain from displaying the response to the user input, generated using the artificial intelligence model, in response to the identification that the response does not satisfy at least one criterion from among the one or more criteria identified on the basis of the evaluation prompt.
Need to check novelty before this filing date? Find Prior Art