Answer-first summary for fast verification
Answer: Implement a framework like ReAct, which allows the LLM to generate reasoning traces and perform task-specific actions that leverage external tools if necessary.
The question specifically asks for an approach that enables an LLM to perform multi-stage reasoning by planning and adapting actions while utilizing external tools. Option C (ReAct framework) is the optimal choice because it explicitly combines reasoning and acting - the LLM generates reasoning traces to plan its approach and can perform task-specific actions that leverage external tools when needed. This allows for dynamic adaptation during complex reasoning tasks. Option B (Chain-of-Thought with manual tool input) requires human intervention for tool usage, which doesn't enable the LLM to autonomously plan and adapt actions. Option A relies solely on pre-trained knowledge without external tools, and Option D involves unstructured API calls without proper planning, which is inefficient and unreliable for complex reasoning tasks. The community discussion shows equal support for B and C, but C better matches the requirement for autonomous planning and adaptation with external tools.
Author: LeetQuiz Editorial Team
Ultimate access to all questions.
Which approach enables an LLM to perform multi-stage reasoning by planning and adapting its actions while utilizing external tools?
A
Train the LLM to generate a single, comprehensive response without interacting with any external tools, relying solely on its pre-trained knowledge.
B
Use a Chain-of-Thought (CoT) prompting technique to guide the LLM through a series of reasoning steps, then manually input the results from external tools for the final answer.
C
Implement a framework like ReAct, which allows the LLM to generate reasoning traces and perform task-specific actions that leverage external tools if necessary.
D
Encourage the LLM to make multiple API calls in sequence without planning or structuring the calls, allowing the LLM to decide when and how to use external tools spontaneously.
No comments yet.