from Hacker News

Ask HN: Is ChatGPT or GitHub Copilot increasing productivity?

by web99 on 4/23/23, 9:39 PM with 8 comments

Would love some real examples where these tools are helping you write less boilerplate or other code, debug faster, and ship faster in general. If they are, what's your sense of time saved or productivity boost?

  • by softwaredoug on 4/24/23, 12:50 AM

    I largely treat ChatGPT as my personal Stackoverflow. Except I get to break all the stackoverflow rules.

    I can ask for hyperspecific questions, ask one-line questions, ask it to generate code to do a very specific thing, and ask follow up questions. I can literally ask it to do my homework if I wanted to - though that is of limited use when you actually have to maintain the thing. At best "doing the work" is really just getting an example to get started.

    Like Stackoverflow, you get more value if you ask a good question. I need enough expertise to ask a question to get started.

    Does it always work? No. Does it completely get rid of Google? No. But it does work about 90% of the time. That's probably better than random stuff I find online :) We should carry skepticism at all times, regardless of the search.

    Where is Google better / more useful?

    * Getting to actual docs, straight from the horses mouth, with no hallucinations :)

    * Looking for authority or social proof (What Stackoverflow or Reddit answers are highly voted, linked to, considered authoritative)

    * Get the latest info - more up to date than the LLM's training data. ChatGPT will sometimes give me a version that worked several years ago, but doesn't work now.

    Google is better at retrieving to things-about-the-knowledge; ChatGPT is better at synthesizing and presenting knowledge

    All in all, it doesn't replace me, but creates a very seamless experience of researching, learning, and taking the next step.

  • by swah on 4/24/23, 10:56 AM

    The other day it wrote the first one or two versions of a python script to connect to a database and do a specific change to a column. I only had to slight changes because my prompt was not super complete and the changes (duplicating some lines and renaming some words) were faster than re-prompting...

  • by frje1400 on 4/24/23, 8:49 AM

    I find it amazing for turning data into markdown tables (which are tedious to write by hand), for documentation.

  • by tkiolp4 on 4/24/23, 8:53 PM

    I can only use ChatGPT with generic code questions, because I don’t want to upload proprietary code to OpenAI’s servers. In that regard, ChatGPT is slightly faster than Googling/SO. I don’t see any productivity boost as long as I cannot upload proprietary code.

  • by satvikpendem on 4/24/23, 12:51 AM

    Sure, I'm building an application where had I done it manually, it would have taken me twice, or even thrice as long. I had ChatGPT generate a lot of the boilerplate and talk me through the general approach when I was stuck, and it was a lot more effective than Stack Overflow.

  • by verdverm on 4/23/23, 10:41 PM

    Yes, it is very task dependent.

    I used it the other day to help me with graph drawing in a Jupyter notebook the other day. It certainly felt faster than searching with google or sifting through docs just to find the right functions to call.

    I also see it fail miserably on other programming tasks.

  • by maxilevi on 4/23/23, 10:39 PM

    Did you try it and couldn't find any boosts in productivity?