The conversational chatbot from OpenAI, ChatGPT, has attracted the attention of users worldwide. However, the lesser-known tool called CODEX from OpenAI has quickly become a top favorite among developers. Codex currently powers the Copilot feature on GitHub.
For those, who are relatively new to the world of programming, GitHub is an open-source community where developers share the code for the software they have written for others to use. Microsoft acquired GitHub over four years ago. Working closely with OpenAI, Microsoft has gained access to ChatGPT and Codex, OpenAI's ChatGPT-like solution for code.
How does Open AI's Codex work?
Like ChatGPT, Codex also predicts what the user is about to say next and, in the case of developers, what they will write next in their code. Since, in addition to the core code, software engineers also spend a lot of time writing scaffolding code needed to make the software work, Codex can help users finish these bits faster.
GitHub's deployment of Codex has been branded as Copilot, which has gone through millions of pieces of code on the repository to see which pieces of code are standard and can help developers write this when they start writing their code. GitHub estimates that Copilot can write over 40 percent of code, improving productivity.
Software companies, big and small, have been using Copilot to get through the mundane bits of coding so that developers can focus on the significant bits. In some cases, users have found it spooky that artificial intelligence (AI) knew precisely what they were trying to code for.
Like ChatGPT, Copilot can also tap into previous interactions with developers to understand the type of projects undertaken and accordingly make its suggestions. Developers could also use it to create essential apps without doing much of the coding themselves. However, this does not mean we no longer need software engineers.
While Copilot might be great at suggesting pieces of code that solve developer's problems, the AI is also prone to 'hallucinations' - a term finding an increasing number of mentions are more users interact with AI. Users have found that the bot can give compelling answers with great confidence, only to find that certain tools it mentions do not exist.
Not just that, Copilot's suggestions are not always correct and can end up making the piece of code not work. What is worse is that the AI can't help you figure out the errors with the code either, leaving developers who intend to rely on it in the lurch.
So developers out there, here's what you need to know. AI may be able to help you a little bit in your work, but you can't rely on it to do the most difficult parts, either. That's where you stand out and are getting paid for.