GPT-3 Programming Needs To Go, And It’s Not Even Here Yet.
“Build web apps lightning fast.” At least they didn’t claim to have cured scalability!
Another day, another way to ruin the programming scene. What makes programming so useful is having a specified instruction set that the computer will execute without any unknown-interpretations of what it should be doing. Such was broken when recently a website that goes by the name “debuild” showed off a unique feature that used OpenAi’s GPT-3 language model.
Before we go into the specifications of the website, let us talk about the GPT-3 model. What the model offers is an extreme performance for runtime text decoding by a neural network. GPT-3 is useful for “strong performance on many NLP datasets, including translation, question-answering, cloze tasks, and many other on-the-fly reasoning[s].” — Sourced from Cornell University.
This is extremely useful when we could be asking artificial intelligence a question that could have multiple interpretations. By having a language model that can infer based on context, you can get a powerful A.I. that could define the future of computer perception.
The issue isn’t with GPT-3 as a matter of fact, when implemented for its proper usage it’s amazing. The issue lies within trying to code components and websites in raw English language. We invented coding to have a known interpretation every time. Using English removes the certainty of the program. It is impressive that GPT-3 can morph a blob about how something should work and look into code, but it’s not impressive when it comes to long term scalability.
Shameem, wrote two samples for GPT-3 to learn off, then it proceeded to do everything else. He stated, “There were a few exceptions, like the JSX code for tables being larger than the 512 token limit.” He even mentioned that if he removed the JSX and turned it into plain HTML/CSS it would output plain HTML/CSS.
Shameem even generated Google’s home page.
Simply stated, debuild.co should not exist. Imagine if someone received a code base with the entire thing was in English. Writing an application like that could lead to a contextual nightmare. Code exists for a reason, to have an even playing field for the industry, and allow anyone to understand what is going on. Taking away critical thinking from the developer is going to lead to more people creating, and less people understanding the underlying application.
My Verdict?
It is a unique experience, fun to mess around with, and a good exercise to show what GPT-3 is capable of. But negligent to think it will be able to take over as a new meta for coding, nor should we want it too.
