Generative AI Specialization Using Simulated Conversation History
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing generative AI models struggle with accurately identifying and classifying semantic features in documents due to varying language usage and require manual review, which is time-consuming and prone to inaccuracies.
Innovation Solution
A method involving a fabricated conversation history with simulated roles and a curated example set is used to interact with a language model, guiding it to accurately identify and classify semantic features by providing pre-prompted context.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual review is used to identify and classify semantic features, then accuracy can be maintained, but time consumption increases significantly
Solution Approach 1:
The patent creates a fabricated conversation history that copies and simulates the interaction pattern between a user and language model. This simulated conversation includes example prompts and responses that replicate genuine usage scenarios, allowing the language model to learn from these copied interactions and improve its ability to accurately identify and classify semantic features without requiring manual review of each case
Solution Approach 2:
The patent performs preliminary action by generating and providing a fabricated conversation history before the actual semantic feature identification task. This pre-generated context includes example prompts and responses that prepare the language model in advance, enabling it to perform accurate classification without time-consuming manual review during the actual task execution
2Productivity
If generative AI models are used to identify semantic features, then productivity increases, but response accuracy decreases due to hallucinations
Solution Approach 1:
The patent implements feedback by incorporating example responses within the fabricated conversation history that demonstrate accurate semantic feature identification. These example responses serve as feedback signals that guide the language model toward more accurate responses, reducing hallucinations while maintaining high productivity through automated processing
Solution Approach 2:
The fabricated conversation history acts as an intermediary between the language model and the document being analyzed. It provides contextual mediation through simulated previous interactions that include example prompts and responses, helping the model maintain accuracy by referencing these intermediary examples during its analysis
3Measurement precision
If a fabricated conversation history with example sets is provided to the language model, then response accuracy improves, but device complexity increases
Solution Approach 1:
The system performs self-service by automatically generating the fabricated conversation history based on the document content and semantic features being analyzed. The process uses the language model's own capabilities to create the simulated conversation history, eliminating the need for external manual intervention or complex pre-processing systems while maintaining high response accuracy
Data Source
AI summary
A method of interacting with a large language model to elicit a semantic feature of interest from a document under review includes electronically inputting, in an application program interface of a chat application, a first prompt assigned to a user, the first prompt yielding a plurality of possible responses from the language model based on content of the document under review; generating an example set comprising text from example documents representative of each of the plurality of possible responses; and electronically inputting, before the first prompt in an application program interface of a chat application, a fabricated history of a conversation between the user and the language model. The fabricated history includes the example set and a plurality of possible responses assigned to the language model.