LLM Task Solving with Dynamic Example Retrieval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing large language models (LLMs) face challenges in accurately solving tasks due to the instability of user-generated examples and the mismatch between provided examples and current tasks, leading to inefficiencies and failures in task completion.
Innovation Solution
An information processing apparatus and method that utilizes a memory to store successful task solutions as examples, allowing a large language model to learn from similar past experiences and solve tasks effectively through In-Context Learning (ICL).
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a determined (fixed) example is given to the LLM, then the system complexity is reduced, but the task solution accuracy deteriorates because the example may not match the current task
Solution Approach 1:
The patent implements a dynamic example selection mechanism where the system automatically retrieves and selects examples from a database based on the current task characteristics. This allows the example set to change dynamically according to task requirements, resolving the contradiction between system simplicity and solution accuracy by automating the adaptation process without requiring complex manual intervention
Solution Approach 2:
The system performs self-service by automatically selecting appropriate examples without user intervention. The automated example selection process evaluates task characteristics and retrieves relevant examples from the database, enabling the system to adapt to different tasks independently while maintaining high solution accuracy
2Adaptability or versatility
If a user generates examples at any time, then the adaptability to current tasks is improved, but the user load increases significantly
Solution Approach 1:
The system implements self-service by automatically selecting examples based on task characteristics without requiring user generation or selection. The automated retrieval process analyzes the current task and fetches relevant examples from the database, maintaining high adaptability while eliminating the burden of manual example creation
Solution Approach 2:
The patent introduces an intermediary automated selection mechanism between the user and the LLM. This intermediary process handles example selection by matching task characteristics with stored examples, preserving adaptability while shielding the user from the complexity of example generation and selection
3Device complexity
If an inappropriate example is given to the LLM, then the device complexity remains low, but the productivity decreases due to time consumption or task failure
Solution Approach 1:
The patent implements a feedback mechanism where the system evaluates task characteristics and selects examples based on this evaluation. This feedback-driven approach ensures that appropriate examples are provided to the LLM, improving task solving efficiency without complicating the system structure through manual intervention
Data Source
AI summary
An information processing apparatus including: an accessor that accesses a memory for storing, for each of a plurality of tasks, a successful process in solving the task as an example; and a solver that acquires an example similar to a solution process for a task to be solved from the memory, causes a large language model to perform learning of the example, and solves the task to be solved by using the large language model in which the learning has been performed.