Testing OpenAI’s ChatGPT for coding: first impressions

Like many developers this week, I've been experimenting with OpenAI's new ChatGPT service. Here, GPT stands for "Generative Pre-Trained Transformer".

ChatGPT is essentially a web-based conversational interface (thus: "chat") over a large language model (LLM), and it can answer questions, write text, and, quite surprisingly, write code!

So far, tools that say they "write code" usually fall apart after a few lines and are mechanical at best. But this time, ChatGPT is different: it understands the context, follows your instructions, and can explain what it is doing as it goes. And since it is a chat, you can even ask it to rewrite something you are not happy about. Google's search can't do that.

A few early impressions so far:

• It's fast (and good) for producing small code snippets at the function/method level. It also knows many languages, not just C# (which I tried first, as you might guess).
• It can summarize or rephrase technical text.
• It understands about PowerShell, SQL and even EDI (I tried).
• You can "talk to it" and guide it like a very patient helper.

What this means in the long term is hard to say, but even in these early days, I can already see lots of uses: generating templates, explaining tricky pieces of code, or helping quickly draft something you'll refine yourself.

This is an impressive technology preview. I'm curious to see where this goes in the coming months.