How to Get LLMs to Solve Math Problems Accurately

It is commonly stated that large language models (LLMs) “can’t do math.” While this is partially true, it oversimplifies the reality.

LLMs can, in fact, handle mathematical tasks — but only when prompted with the appropriate structure and context.

Understanding the Limitation

LLMs such as ChatGPT are not designed to function as traditional calculators. Instead, they predict likely sequences of text based on extensive language data. When asked, for example, “What is 173 × 19?” the model generates an answer based on examples it has encountered, rather than performing an actual calculation.

This explains why LLMs may produce correct results in some cases and incorrect ones in others.

Three Prompting Methods That Improve Accuracy

LLMs follow patterns. Engineers understand systems. When structured appropriately, these models can reliably generate accurate, logical outputs. The following prompt styles are proven to enhance mathematical accuracy:

• Break Down the Problem Instead of asking a single-step question: “What is 173 × 19?”Reframe it as a multi-step prompt: “What is 173 × 10?”“What is 173 × 9?”“Add them together.”

• Provide Role Context Start with a directive: “You are solving a math problem. Think through each step carefully.”

• Ask for Verification Prompt the model to double-check its answer: “Can you verify that result using a different approach?”

These structured approaches reduce error rates and prompt more disciplined reasoning from the model.

Why Engineers Are Naturally Aligned

The structured thinking inherent to engineering workflows aligns well with how LLMs process and respond to inputs. By designing prompts with clear logic and sequential structure, engineers can guide LLMs toward consistently accurate outcomes.

This makes LLMs a powerful extension of engineering analysis — not a replacement, but a tool that amplifies human capability.

Learn to Prompt Like an Engineer

At Singularity, we teach engineers how to integrate AI into their workflows to improve precision, reduce redundancy, and accelerate problem-solving.

Receive:

• A prompt library designed for engineering use

• Examples from real technical projects

• Guidance on integrating LLMs into daily tasks

Subscribe to our newsletter: https://www.singularityengineering.ca/general-4