Last Updated: 30 September 2023
Building a Web App with ChatGPT-4
Do you ever wonder, can an AI build a web application? The thought intrigued me, and I spent some time exploring the potential of OpenAI’s ChatGPT-4 in web development. Let’s unpack this journey together.
Here are some thoughts after using ChatGPT to build a free web based teleprompter app. I am both surprised and disappointed in the experience. If you’re new to teleprompters, here’s how you use the teleprompter app.
General Knowledge vs Domain-Specific Expertise
ChatGPT-4 is a fantastic conversation partner. It can provide insight into a wide array of topics, from history to pop culture, and even philosophy. But web development? That’s a different story. It’s a field that requires specific, detailed knowledge. Asking ChatGPT-4 to build a web application is akin to asking a generalist to perform a specialist’s job - the results are, predictably, a mixed bag.
Playing Roulette with Code Generation
When it comes to generating code, using ChatGPT-4 feels like playing roulette. Every so often, you might hit the jackpot and get a functional piece of code. More often, though, the output feels like a shot in the dark. It sometimes leads to syntax errors, incorrect logic, or non-functional results.
When attempting moderately complex tasks and asking for a suggestion to be integrated into a previous piece of code, it often leaves out other pieces of functionality. It is an extremely iterative process and in some cases it would be much faster to write the code myself.
Other times it overcomplicates things and when you suggest a simpler approach it agrees that it is a better way to do it. However, it is inconsistent in suggesting the better options.
Hallucinations
I’ve found ChatGPt helpful in converting some of my documentation and articles into markdown for use in static site generators, but you have to double check things. One article had some affiliate links to amazon products. ChatGPT decided to add links for all products and generated fake affiliate links that looked similar to the real ones.
Missing Edge Cases
Edge cases, those peculiar scenarios that only occur under specific conditions, pose a significant challenge for ChatGPT-4. It’s like trying to explain to an alien why we wear green on St. Patrick’s Day. The AI’s approach to handling edge cases is still maturing, and it’s not quite there yet.
ChatGPT-4’s Memory Limitations
A major limitation of ChatGPT-4 is its lack of memory. It doesn’t retain past interactions, making project continuity a significant challenge. In one instance it completely lost context while in the same chat session. Every interaction is a new chapter with no connection to the last.
Also due to the current limitations on memory allowed per message, it is challenging to work with entire code files. ChatGPT sometimes does well breaking things into code blocks and other times it can’t make up its mind. I’ve seen it take a chunk of code and spit out a partial code block followed by some regular text and then back to code block, completely mangling the code.
Debugging
When it comes to debugging and error handling, ChatGPT-4 is more of a bystander than an active participant. It’s like asking a sloth to catch a frisbee. It might stumble upon an error now and then, but don’t expect it to take the reins when things go south.
Sometimes it’s right on the money and helpful. Other times its solutions are irrelevant or from outdated versions of code. A lot of times it merely explains the obvious.
Collaboration
Collaborating with ChatGPT-4 is, at best, a one-sided conversation. It’s a bit like talking to a brick wall. Sometimes it gets it and other times it feels like talking to a two year old.
The Bright Side of ChatGPT-4
Despite these limitations, ChatGPT-4 isn’t all bad news. It can serve as a handy assistant, offering suggestions and ideas, playing a supporting role in your coding journey. It’s like a sous chef in your kitchen, helping you chop and prep, but not quite ready to take over the main course.
Looking Ahead: The Future of AI in Web Development
The potential for AI to overcome these hurdles and become an equal partner in web development is enticing. Will it be able to navigate the complexities of code, or will it remain a tool of suggestion and assistance?
In the meantime, we should continue to explore, innovate, and remember to find humor in the unpredictability of AI. After all, it’s not every day that an AI suggests a hot pink and neon green palette for a professional business website! So, stay curious, keep coding, and embrace the journey.