3 ms·

Genuine question, as I didn't use LLMs much: Sure, they're good for quick prototypes or even launching production grade services. But what about maintenance?

• cloverich 3y ago
  Yesterday I asked chatgpt to write me some boilerplate and it did. I went through the exercise of using it, find some bugs, pointing them out, and fixing them (with help, or on my own). Today I asked a similar question and it gave me boilerplate. Then it ran the code. Then it noticed the error, then it fixed it on its own. All without me typing anything at all. I pointed that out to someone, and they did not appreciate that it has quickly gained the capacity to test if what it says is true, at least in some limited sense. I'm in awe. At any rate, I think yes, you plug in code and ask it questions and it can answer them, over time answering them better. Since i've been using it not only can it execute code, but can generate test cases and test _data_. I think wrt to legacy code, it could come up with tests and refactoring strategies. More generally though, AI is coming. And with it, the ability to write better code than even the best of us. That maybe a year or 50 out, I"m not sure. But its coming. And with each passing day, I move the "when" more into my lifetime.
• sharkjacobs 3y ago
  From the article: > You can’t just say to the A.I., “Solve my problem.” That day may come, but for now it is more like an instrument you must learn to play. You have to specify what you want carefully, as though talking to a beginner. It’s easier to explain “solve this common problem you’re already familiar with” or “create a prototype with pretty flexible specifications” but if you can break down and explain a problem to a rubber duck[1], you can explain it to GPT4, and it’s a better conversationalist than the duck [1] https://en.wikipedia.org/wiki/Rubber_duck_debugging https://en.wikipedia.org/wiki/Rubber_duck_debugging