3 ms·
After using these tools, I feel like they are only good at producing code that is already simple to write if you know what you're doing. Basically if your probl
by spicyusername 2y ago
After using these tools, I feel like they are only good at producing code that is already simple to write if you know what you're doing. Basically if your problem fits neatly into something that would have appeared in documentation or directly in a Stack Overflow post that the training data vacuumed up, you might get something usable.
For anything even slightly non-trivial, I always spend more time fighting the prompt than I would have if I just sat down, read the docs, and thought a little about the problem.
And this doesn't even get into things like style, readability, etc.
Maybe this will change with later incarnations of these tools, but a cursory understanding of how these tools work make me suspicious. How can something that, from first principles, generates statistically random output, produce reliably functioning code for large non-trivial tasks?
That's a very very different requirement then producing something like a poem or summarizing meeting transcripts, where a little randomness is expected and nearly all the training data is in the right format of regular human dialogue?
There's no intelligence in these tools, just regurgitation. The exact opposite of what you need for solving software engineering problems.
- j7ake 2y agoSimple for you is not simple for a beginner. For example, let’s say you wanted to write code in a language you did not know. Then these simple tasks can be accelerated because you’re not blocked by syntax errors.
- IshKebab 2y agoI agree they're best at simple things and things that are talked about on the internet a lot (they're pretty great at parsers for example). But... a significant amount of code is simple. Even in complex projects there's still a lot of trivial code. And even when they produce something wrong, it's usually still faster than starting from scratch. At least for smallish bits of code. Though admittedly it can be a bit more frustrating on occasion because you're doing something more like code review and who doesn't hate code review?
- shombaboor 2y agototally agree. I use it for boilerplate functions and restapi templates. It saves me a lot of time for these specific things but it's been wrong many times for very library parameters.