Abstract
While large language models (LLMs) can be used to perform complex tasks, e.g., rating search results, training and maintaining LLMs runs into several challenges. These include the significant expense to obtain training data, difficulty in identifying weaknesses, lack of explainability, and the lack of transferability to new foundation models or versions. This disclosure describes the use of flowcharts for orchestration of large language models (LLMs) to perform complex tasks. A complex task is broken down into a predefined set of simpler tasks. These simpler tasks are represented in a traditional flowchart format. The LLM is prompted to make the individual decisions at different steps within the flowchart. A simple pre-programmed orchestrator is implemented that follows the flowchart to choose which decisions to pose to the LLM and the next state transitions. Training sets for individual tasks of the predefined set and performance of the LLM on the tasks can be measured and verified. Constructing the flowchart can be done by humans, with assistance from the LLM.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
NA, "Flowchart Orchestration for Complex Task Performance Using a Large Language Model", Technical Disclosure Commons, (December 04, 2024)
https://www.tdcommons.org/dpubs_series/7618