| > I don't understand what use this is. It takes 30 seconds versus 5 minutes to write myself? > So you run it on a GB of similar data (meaning not literally specifically a GB, but well more than you can reasonably verify by hand). The same way you would test code you wrote yourself. Firstly, I check over the code. Secondly, I run it on a small data set and make sure things look correct. Thirdly, there's always the option of writing tests into the prompt after or even as the initial prompt. > Then what? How do you know it works fine, or if it obviously crashes, what to do next? For this specific use in this language I've done this 100s of times. It's just letting me save my energy remembering things and avoiding the stress of forgetting things. 30s vs 5m is a very large difference as well. How many 30s tasks would you be confident in finishing a day vs 5m ones? For me there's a huge difference in energy used. > How do you know it works fine, or if it obviously crashes, what to do next? Well, you either: - give chatgpt the errors and see if it can fix them
- fix errors yourself and reprompt
- fix errors yourself Using this as a tool like this is a skill like anything else and only experience will tell you which direction is most effective. For me though, I typically only give it a few chances to get things right. If it doesn't, I just finish the task myself. Typically I can start with the scaffolding it gave me though. There's also usefulness for exploring things you don't know about, but I wouldn't have high confidence or use that for an important work task without adequate testing to build my confidence. |
But that's not what programmers do. A programmer that does the same thing hundreds of times is not programming.