This is a student lab report on applying Chain-of-Thought reasoning to multi-step problems. It compares direct-answer prompts with step-by-step reasoning prompts across three problem types: a mathematical calculation, a logical puzzle, and a business investment decision. The report includes the exact prompts used, the model outputs for each, and a comparative analysis of accuracy, logical…
This document is a student lab report titled 'Apply Chain-of-Thought Reasoning to Solve Multi-Step Problems' by Salsabilla Putri. It is intended for an academic or educational audience studying prompt engineering, AI reasoning, or cognitive science approaches to large language models.
The report begins by presenting three problem types: a mathematical problem (calculating change from a purchase), a logical puzzle (seating arrangement of three people), and a business decision problem (evaluating a marketing campaign investment). For each, it shows both a direct-answer prompt and a Chain-of-Thought prompt, along with the model outputs.
A section on CoT implementation explains how the prompts were structured as algorithmic guides, isolating calculation stages to prevent the model from confusing gross revenue with net profit or skipping intermediate steps. The report then provides a comparative analysis table rating each prompt style on accuracy, logical coherence, and transparency.
The accuracy improvement section highlights that CoT corrected the business decision output from an incorrect 'Yes' to a correct 'No' by forcing profit margin calculation. The transparency evaluation notes that CoT exposes the model's reasoning steps, enabling human audit of logic.
The report concludes with scenario identification (where CoT is most effective) and limitation awareness, including latency, cost, compounded hallucinations, and unnecessary overhead for simple factual questions.