How are you using LLMs for coding?

Whenever it gets in a cycle of fucking up, I ask it to break the code down into smaller functions and write a test suite for each individually.

Tell it what language or packages to use or it might make selections that add dependencies or require installing stuff on your computer.

Tell it how you want your code written or it will be an extra chore to accommodate linting requirements.

Basically just being very explicit.

I mostly write low-level code in Python and JS. When working with a new API, I would sometimes find it difficult to locate information to fix problems. With ChatGPT I simply rubber duck my problem. It usually only takes a couple of iterations to zoom into the core solution.

Copilot in visual studio to fix compiler errors and let it type all the boilerplate code by TAB-ing. Sometimes it even manages to spit out an entire - and correct - function.

Mistral Le Chat for more advanced questions and figuring out things that copilot can't.