Here’s how I use LLMs to help me write code

Unlocking the Power of Large Language Models for Code Writing: A Comprehensive Guide

Introduction

As the use of Large Language Models (LLMs) for code writing continues to gain traction, many developers are left wondering how to harness their potential. With the ability to generate code at unprecedented speeds, LLMs seem like a dream come true. However, the reality is more nuanced. LLMs are not a replacement for human expertise, but rather a tool that can augment and accelerate the development process. In this comprehensive guide, we’ll explore the best practices for using LLMs to write code, and provide actionable advice for integrating these models into your workflow. Understanding LLMs: Setting Reasonable Expectations

Before diving into the world of LLMs, it’s essential to set reasonable expectations. LLMs are not Artificial General Intelligence (AGI), but rather sophisticated language processing tools. They can predict sequences of tokens with remarkable accuracy, making them ideal for tasks like code completion and generation. However, they are not perfect and can make mistakes. A useful mental model for working with LLMs is to think of them as an over-confident pair programming assistant. They can provide valuable insights, look up relevant information, and execute tedious tasks with ease. However, they can also make mistakes, and it’s crucial to test and verify their output. Accounting for Training Cut-Off Dates

A critical characteristic of any LLM is its training cut-off date. This refers to the point in time when the model’s training data was last updated. Models with older cut-off dates may not be aware of recent developments in programming languages, frameworks, or libraries. When working with LLMs, it’s essential to account for these cut-off dates and provide context to ensure accurate results. Context is King: Providing Clear Input

To achieve desired results with LLMs, it’s crucial to provide clear and concise input. This includes specifying the programming language, framework, and any relevant libraries or dependencies. The more context you provide, the better equipped the LLM will be to generate accurate and relevant code. Asking for Options and Providing Guidance

When working with LLMs, it’s often helpful to ask for options rather than a single solution. This allows you to review and select the best approach for your specific use case. Additionally, providing guidance on the desired output can help the LLM generate more accurate results. Testing and Verifying Code

One of the most critical aspects of working with LLMs is testing and verifying the generated code. This ensures that the code is accurate, functional, and meets the required specifications. It’s essential to integrate testing into your workflow to ensure the quality and reliability of the code. Integrating LLMs into Your Workflow

LLMs are not a replacement for human expertise, but rather a tool that can augment and accelerate the development process. By integrating LLMs into your workflow, you can leverage their strengths while minimizing their weaknesses. This includes using LLMs for tasks like code completion, generation, and review, while reserving complex decision-making and high-level design for human experts. Vibe-Coding: A Great Way to Learn

Vibe-coding is a technique that involves using LLMs to generate code snippets and then modifying and refining them to suit your specific needs. This approach can be a great way to learn new programming concepts and techniques, as it allows you to explore different solutions and approaches. A Detailed Example: Using Claude Code

Claude Code is a popular LLM designed specifically for code generation. In this example, we’ll use Claude Code to generate a simple web scraper in Python. Code Example

“`python
import requests
from bs4 import BeautifulSoup

def scrape_website(url):
response = requests.get(url)
soup = BeautifulSoup(response.content, ‘html.parser’)
# Extract relevant data here
return data

# Test the function
url = “https://www.example.com”
data = scrape_website(url)
print(data)
“`

The Biggest Advantage: Speed of Development

One of the most significant advantages of using LLMs for code writing is the speed of development. LLMs can generate code at unprecedented speeds, allowing developers to focus on high-level design and decision-making. Amplifying Existing Expertise

LLMs can amplify existing expertise by providing developers with a tool that can generate code, look up relevant information, and execute tedious tasks. This allows developers to focus on complex decision-making and high-level design, while leveraging the strengths of LLMs. Bonus: Answering Questions about Codebases

LLMs can also be used to answer questions about codebases, providing developers with valuable insights and information. This can be particularly useful for large and complex codebases, where it can be challenging to find specific information. Side Effects and Safety Considerations

While LLMs can be a valuable tool for code writing, there are some side effects and safety considerations to be aware of. These include:

* **Dependence on LLMs**: Over-reliance on LLMs can lead to a decline in human expertise and critical thinking skills. * **Code quality**: LLMs can generate low-quality code, which can lead to maintenance and debugging issues. * **Security**: LLMs can introduce security vulnerabilities if not properly configured and tested. Conclusion

In conclusion, LLMs can be a powerful tool for code writing, but require careful understanding and application. By setting reasonable expectations, accounting for training cut-off dates, providing clear input, and testing and verifying code, developers can unlock the full potential of LLMs. Remember to integrate LLMs into your workflow as a collaborative tool, rather than a replacement for human expertise. Actionable Next Steps

1. Start by experimenting with LLMs for simple code generation tasks. 2. Integrate LLMs into your workflow for tasks like code completion and review. 3. Use LLMs to answer questions about codebases and provide valuable insights. 4. Continuously test and verify code generated by LLMs to ensure accuracy and functionality. FAQ

Scientific References

1. Ten simple rules for using large language models in science… – https://pmc.ncbi.nlm.nih.gov/articles/PMC10829980/
2. AI support for data scientists: An empirical study on workflow… – NIH – https://pmc.ncbi.nlm.nih.gov/articles/PMC12227384/
3. Using large language models to facilitate academic work in the… – NIH – https://pmc.ncbi.nlm.nih.gov/articles/PMC12106154/
4. Evaluating a large language model’s ability to solve programming… – https://pmc.ncbi.nlm.nih.gov/articles/PMC10564134/
5. ChatGPT: how to use it and the pitfalls/cautions in academia – NIH – https://pmc.ncbi.nlm.nih.gov/articles/PMC12597148/

Data Table: LLM Training Cut-Off Dates

LLM Model	Training Cut-Off Date
Claude Code	2022-01-01
CodeBERT	2021-06-01
LLaMA	2020-12-01

Discover our high-quality NMN products.

Learn more about us.

Have questions? Check our frequently asked questions.

For more detailed scientific information, see NAD+ metabolism and aging: What we know and what we don’t.