Can ChatGPT Code for You Yet?

September 01, 2023


I have been testing and trying ChatGPT as an aid for my coding and DevOps work. In a few fun instances I have been able to feed ChatGPT a bunch of existing code in JavaScript, ask it to insert a conditional, and it outputted the answer. Everything outside this has fallen short of the promise of what AI tools can execute for coding. AI remains mostly an enhanced search engine, that sometimes gets the correct answer.

Where it Shines

Many tools have documentation that is not easily discoverable, poorly written, or disorganized. ChatGPT cuts through this mess to deliver the thing you’re looking for more quickly.

Why it Flops

  • ChatGPT often lacks the full context on the project you’re working on.
  • It rarely is trained on or has access to any private libraries you use.
  • It fails to differentiate between versions of libraries on different operating systems.
  • The training data is not frequently refreshed, leaving out new features and changes.

What I am Using Instead

  • Kapeli’s Dash App makes it easy to search official document sets locally.
  • Kagi has a programming lens that makes it easy to search lots of programming related resources.
  • Kagi’s FastGPT is a middle ground and is AI backed by web results.
  • GitHub and StackOverflow’s search engines search what are the most comprehensive resources for most of my issues.

Want to get posts like this in your email?

This work by Matt Zagaja is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License.