I feel burned out just knowing how many people use the dumb things.

I’ve only used them a fraction of what’s described in the article and I get it. I’ve tried so many prompts, tactics, angles and techniques to get the llm to understand exactly what is required, but it always seems to miss something. I don’t mean it makes a mistake per se. More like it makes odd assumptions or suggests something and then ignores it.

So far the only reliable use it has had consistently is spotting bugs. Next up would be suggesting alternative ways to do things. Which is very useful on one hand, but can also add to my workload because now I have to checkout this new approach and see how it could be incorporated into whatever it is I am working on.

Asking it to write things from scratch, even within strict parameters and guardrails, with per unit self-testing and debugging, still doesn’t seem to result in code you don’t have to review and vet extensively to ensure it does what it is meant to do.

I guess this is a long winded way of saying that what is often gained from using an llm for coding, sometimes ends up taking away some of the magic of creating the code yourself.

If that makes any sense…