Information acquisition method, device, equipment and medium based on large language model

By gradually updating the input state and intermediate state in the large language model until the target generation content is generated, the problems of inaccurate output results and hallucinatory phenomena of large language model are solved, and higher output accuracy and controllability are achieved.

CN117609437BActive Publication Date: 2025-03-28BEIJING BAIDU NETCOM SCI & TECH CO LTD
1 Cites 0 Cited by

Patent Information

Application Number
CN202311340344.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-10-16
Publication Date
2025-03-28
Estimated Expiration
2043-10-16

AI Technical Summary

Technical Problem

In the process of information acquisition and reply, large language models have problems such as inaccurate output results and hallucinations, which affect their application effect.

Method used

By inputting natural language content into a large language model, the input state and intermediate state are gradually updated until the target generates content, thereby extracting the reply content of natural language content. Through the introduction of action marking and action description content, this method increases the controllability and interpretability of the reasoning process and reduces hallucination phenomena.

Benefits of technology

It improves the accuracy of the output results of the large language model, enhances the control of the inference process, alleviates the hallucination phenomenon, and improves the accuracy of the reply based on the large language model.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The present disclosure provides an information acquisition method, device, equipment and medium based on a large language model, which relates to the field of large models, and specifically to the fields of artificial intelligence, large language models and human-computer interaction. The specific implementation scheme is: obtaining natural language content; inputting the natural language content into the large language model to obtain the current first intermediate state; in response to determining that the first intermediate state does not include the target generated content, updating the current first input state according to the first intermediate state to obtain the second input state; inputting the natural language content and the second input state into the large language model, updating the first intermediate state to obtain the second intermediate state; in response to determining that the second intermediate state includes the target generated content, obtaining the reply content of the natural language content according to the second intermediate state. The embodiments of the present disclosure can improve the accuracy of the output results of the large language model.
Need to check novelty before this filing date? Find Prior Art