LLM Task Processing with Service-Specific Commands
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing large language models (LLMs) struggle to effectively understand context for targeted requests and execute device-specific tasks, such as searching for files on a user's client device, and are limited in summarizing lengthy meetings or discussions without providing targeted information.
Innovation Solution
An AI assistant employs LLMs to generate commands for various services, such as search, summarization, and email generation, by accessing a data store of examples to transform user queries into executable commands, enabling the LLM to execute tasks on the user's device.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If LLMs are used to summarize lengthy meetings or discussions, then the summary is generated, but the summary may be lengthy and take significant time to review in detail
Solution Approach 1:
The patent extracts only the specific information relevant to the user's targeted request from the meeting transcript, rather than generating a comprehensive summary of the entire meeting. This selective extraction approach retrieves only necessary information while filtering out irrelevant content, thus reducing review time while maintaining information quality.
Solution Approach 2:
The system applies different processing quality to different parts of the meeting content based on relevance to the user's request. Highly relevant segments are extracted in detail, while less relevant segments are either summarized or omitted entirely, creating a non-uniform information distribution that optimizes for the specific information need.
2Adaptability or versatility
If LLMs are asked to perform device-specific tasks such as searching for files, then the LLM generates responses, but it cannot execute tasks on the user's client device
Solution Approach 1:
The patent introduces an intermediary component that acts as a bridge between the LLM and the client device. The LLM generates high-level instructions, which are then translated into executable commands by the intermediary system that has direct access to and control over the client device's file system and applications, enabling reliable task execution.
Solution Approach 2:
The system segments the task execution process into distinct components: the LLM handles natural language understanding and high-level task planning, while a separate execution engine handles the actual device-specific operations. This segmentation allows each component to specialize in its strength while working together to achieve the overall task.
3Loss of information
If LLMs generate comprehensive summaries of meetings, then all information is captured, but targeted information for specific topics or persons is difficult to locate
Solution Approach 1:
The system extracts and isolates specific information related to the user's targeted request from the meeting transcript, presenting it separately from the rest of the content. This extraction approach allows users to quickly locate and review only the information relevant to their specific topic or person of interest without wading through unrelated meeting content.
Solution Approach 2:
Instead of generating a comprehensive summary and then requiring users to search through it for specific information, the system inverts the approach by directly querying and extracting the specific information needed based on the user's targeted request. This reverses the traditional workflow from 'summarize then search' to 'search then present'.
Data Source
AI summary
One example method includes receiving, by an artificial intelligence (“AI”) assistant, a user query comprising one or more tasks; determining one or more services based on the user query; obtaining, for each of the one or more services, a plurality of examples, each example providing an example command suitable for execution by the respective service; generating one or more instructions based on the user query, the one or more services, and the one or more pluralities of examples; providing the one or more instructions to a trained large language model (“LLM”); receiving, from the LLM, one or more commands corresponding to the user query; for each command of the plurality of commands, issuing the respective command to a corresponding service of the one or more services; generating a response to the user query based on results of the plurality of commands; and outputting the response