ChatGPT: first hand testing

ChatGPT is currently the hype of the moment. Having tested its abilities as both a writing tool and a coding assistant, I will focus on the latter in this post. As a new toy for developers, I was curious to see how its capabilities to produce code could be used in real-life scenarios. This is a no-nonsense evaluation of its usefulness in actual cases of code writing.

My main objective was to determine how much faster ChatGPT makes coding possible. The idea behind this is as follows: you have an app concept, you need to choose the most suitable libraries for development, and then assemble them to achieve the desired result.

After thorough testing, my conclusion is that ChatGPT does help a lot, but it is crucial to learn how to use it effectively. Its usefulness lies more in developer management than in simply doing Google searches for information.

My experience with ChatGPT is based on the free version of GPT3.5, which has limited access with occasional interruptions and a cap on the number of requests. However, this does not really hinder the job at hand. It’s important to note that ChatGPT cannot code an entire application with just the provided specs, but let’s be realistic, how many coders can do that? Instead, ChatGPT should be seen as a junior developer with very fast response times to requests or feedback.

The process to follow when working with ChatGPT is straightforward. First, you need to get an overview of the different steps required to develop your app. ChatGPT will provide you with a broad list of steps to follow, and you can ask for information on which libraries to use and how to combine them.

After that, you need to ask for more specific details on each point of the development process. However, it’s important to be careful not to get lost in the details, as ChatGPT can quickly spiral down deep, nonsensical holes, making it easy to lose sight of the big picture and how different parts of your code need to interact. In case you feel like you’re hitting a dead end, start a new chat and focus on your specific problem. This is a common problem even when managing human coders, so it’s not surprising to encounter it when working with a machine.

It’s crucial to understand what the code suggested by ChatGPT does. You will need to make corrections and adaptations, and suggest modifications as well. One difficulty that can arise is that ChatGPT may not be aware of the latest updates to libraries, leading to broken code. You will need to combine your own Google searches or Stack Overflow support to overcome these issues. However, this is a normal process for a software manager in charge of human coders.

In the end, coding with ChatGPT is like having a development team for free. It’s both exciting and frightening at the same time. I’m interested to see how software will improve in the future, and we should see significant progress in the capabilities and robustness of codes. This is provided that the use of ChatGPT stays under proper control.