LLM Software Tool Planning via Best-First Search
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional computing systems face inefficiencies and inaccuracies in navigating large action spaces for software tool interactions due to unidirectional exploration and excessive resource consumption, often leading to locally optimal solutions and incorrect operations.
Innovation Solution
A software action planning system utilizing a best-first search model with a large language model to iteratively select and expand actions in a decision tree, generating cumulative and future cost scores to optimize action plans, reducing resource usage and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional systems rigidly explore expansive action spaces in a single direction, then the system structure is simple, but the action plan accuracy deteriorates due to error propagation and limited exploration
Solution Approach 1:
The patent segments the action space exploration into multiple independent directions simultaneously. Instead of following a single rigid path, the system divides the search into parallel trajectories that can independently explore different parts of the action space, reducing error propagation and improving overall exploration accuracy.
Solution Approach 2:
The patent transitions from unidirectional exploration to multi-directional exploration by adding dimensional diversity to the search process. This allows the system to navigate the action space from multiple angles simultaneously, preventing getting trapped in local optima and improving action plan accuracy.
2Manufacturing precision
If tree search-based models explore all possible actions in decision space, then the action plan accuracy improves through global optimization, but the resource consumption deteriorates due to significant time and processing requirements
Solution Approach 1:
The patent applies partial action by exploring only the most promising directions in the action space rather than exhaustively searching all possible actions. The system selectively expands nodes that show potential for leading to optimal solutions, achieving good action plan accuracy without the prohibitive resource cost of complete exploration.
Solution Approach 2:
The patent dynamically adjusts exploration parameters such as the number of parallel directions, expansion depth, and selection criteria based on computational resource availability. This allows the system to maintain high action plan accuracy while adapting resource consumption to practical constraints.
3Manufacturing precision
If tree search-based models explore all possible actions, then the action plan quality improves, but the time consumption deteriorates making the system impractical for limited time scenarios
Solution Approach 1:
The patent performs preliminary evaluation of action space nodes to identify promising directions before full exploration. By pre-assessing potential paths and prioritizing those with higher likelihood of leading to optimal solutions, the system reduces the time required to achieve high-quality action plans.
Solution Approach 2:
The patent employs periodic evaluation and pruning of exploration paths, systematically reviewing and eliminating unpromising directions at intervals during the search process. This periodic action maintains solution quality while preventing excessive time consumption on futile exploration paths.
Data Source
AI summary
Methods, systems, and non-transitory computer readable storage media are disclosed for generating action plans utilizing a large language model with a best-first search model. The disclosed system determines a request to utilize a large language model to generate an action plan via one or more software tools. The disclosed system generates the action plan by traversing a decision tree comprising an action space involving the one or more software tools by iteratively: selecting, utilizing a best-first search model, an action from a set of possible actions in the action space of the decision tree; and expanding, utilizing the best-first search model, the action space of the decision tree to include an additional set of possible actions. The disclosed system also executes the action plan via one or more interactions with the one or more software tools according to the action.