Hacker News new | past | comments | ask | show | jobs | submit login

> Trying to get some non-trivial code out of GPT that isn't a mindless boilerplate

For some of us, the boilerplate code generated by ChatGPT is incredibly valuable. I have been utilizing it to create snippets of code for languages that I am not familiar with in terms of syntax. I understand that I may be criticized for this, however, I am not an expert in all of the languages I am required to support. Thus, I often find myself struggling with the syntax while understanding the overall concept I am attempting to accomplish.

Example prompt: "Write me a proof of concept in language X that serializes JSON then POSTS to a HTTP API endpoint."




The thing is - how often do you end up needing such snippets? How much more useful is receiving it from ChatGPT than looking it up in a search engine?

Continuing with the theme of your example, at least in my experience most of the programmer's time is spent on figuring out which endpoint do you need, why and when does that endpoint sometimes return something you don't expect and how to deal with that happening. Writing JSON serializing in a language your head got rusty with may be faster with ChatGPT than it would be otherwise, but the overall benefit seems negligible, especially when you need to watch out at every step for convincingly-looking nonsense.


All in good faith, I believe that receiving assistance from ChatGPT is far more useful than looking up snippets on a search engine. Instead of delving through semi-appropriate snippets, I can concisely describe what I need in one or two sentences and ChatGPT typically provides an accurate result, unless the query is opinion-based, complex, or obscure. I am not expecting perfect code since I take it upon myself to polish the output.

I am now experimenting with ChatGPT parsing our own historical documentation, such as API references, and even basic markdown notes. When it comes to API endpoints I can get 80-90% accurate boilerplate. Admittedly, ChatGPT cannot provide assistance in understanding unexpected behavior, unless perhaps it is due to a misunderstanding of the language. And, though it is true that resolving API endpoint issues can be a time-consuming task, this does not negate the time savings that ChatGPT can provide in other situations.

I would urge you to try it out in your own workflow before calling shots on if the usefulness is negligible or not. And even then, I'd say not every tool is for everyone. This thing is really helping me out and I'm seeing long-term viability for it in my workflows.


> I would urge you to try it out in your own workflow before calling shots on if the usefulness is negligible or not.

What makes you think I didn't? I already stated that:

> It only seems useful for when you know exactly what to do, but you don't feel like typing it all out - which happens

It does pretty well in these kind of situations when the code in question is trivial, but in my experience the time savings coming from that are minor. In fact, its main value comes from my ADHD mind not being able to focus on boring stuff I already know how to do - in those cases, GPT gets me the boilerplate I can start operating on right away, which sure is helpful as otherwise I'd just end up in a web browser scrolling some stuff. Other than that? It seems like verifying ChatGPT's output is much more time consuming than doing it on your own as soon as you get out of "problems copy'n'pastable from StackOverflow" territory.


In good faith I really don't know what you're arguing for.

If you re-read our conversation I express that simple snippets of code from ChatGPT have been accurate enough, ultimately saving me time (paraphrasing). You then push back on me with the following:

> The thing is - how often do you end up needing such snippets? How much more useful is receiving it from ChatGPT than looking it up in a search engine?

> [...] but the overall benefit seems negligible, especially when you need to watch out at every step for convincingly-looking nonsense

I doubled down, doing my best to elaborate where it has added value to my workflow, when it breaks down for me, and addressed specific concerns in your comments. When you say "overall benefit seems negligible" - I read that very clearly as overall the benefit (of ChatGPT) seems negligible (regarding the context of our conversation). That context to me was that ChatGPT helps me with simple snippets and POCs. I did my best to politely rebut that with my personal and professional experiences regarding the tool.

Now you say:

> its main value comes from my ADHD mind not being able to focus on boring stuff I already know how to do - in those cases, GPT gets me the boilerplate I can start operating on right away, which sure is helpful

This is directly compatible and agreeing with what I feel I have been advocating for since my first message in this conversation - that using ChatGPT for simple boilerplate code can be valid (paraphrasing). But, I feel like this is contradictory to, and in conflict with what you previously said: "the overall benefit seems negligible".

---

Overall this conversation seems to contradict itself, and with that it's confusing and frustrating to me. Sorry.

Cheers


> that using ChatGPT for simple boilerplate code can be valid

I don't disagree. I just don't see it as "incredibly valuable", it's way too erratic for that. When it works well it does help, but it works well in such a small minority of software developer's tasks that its impact is pretty much negligible overall - and what I meant to point out is that this is an opinion coming from someone who does find it useful in certain cases.

It makes it a bit easier for me to not get distracted by my web browser, sometimes. It does not change my quality of life as a programmer at all. It would be probably more valuable to delegate these tasks to a junior dev in my team if it had one - the junior would learn from that experience and get closer to becoming a senior, ChatGPT won't.


> I don't disagree. I just don't see it as "incredibly valuable"

Lol - that is by definition disagreeing with the first sentence that I said in our conversation. You disagree that it is "incredibly valuable" and have spent every comment saying that the overall impact of ChatGPT for developers is "negligible overall."

You do disagree with me... and that's OK =)


Have you read what you're replying to? I said that I don't disagree "that using ChatGPT for simple boilerplate code can be valid". It's a completely different thing than what you're implying.

We're commenting under an article that talks about how ChatGPT supposedly makes most of human's skills worthless. Similar sentiment can be seen in many places when it comes to programming. It can't do that if all it helps with are some trivial snippets sometimes. It's helpful for getting boilerplate going, but getting help with boilerplate has negligible effect on programmer's work.


> Have you read what you're replying to?

Yes. I find it sorta offensive that you assume otherwise.

---

Let's boil this down. I said "the boilerplate ChatGPT provides me is incredibly valuable," and further elaborated that using the tool is "valid."

You disagree with "incredibly valuable" but "don't disagree" with the tool being "valid."

So - I'm saying. We disagree that boilerplate output from ChatGPT is "incredibly valuable" and frankly that's just gonna be subjective regardless. Let's agree to disagree here. I just find it funny that you say "I don't disagree" then outright disagree with the exact language I used in the first sentence of our conversation regarding the value of this tool.

---

> We're commenting under an article that talks about how ChatGPT supposedly makes most of human's skills worthless. Similar sentiment can be seen in many places when it comes to programming. It can't do that if all it helps with are some trivial snippets sometimes.

And honestly I feel like this is why our communication and conversation is frustrated here. I don't see our conversation as being a proxy for the infinitely complex and complicated "making human skills worthless" conversation... personally, I just wanted to say "I use this tool for X, and personally have found great value in it."




Consider applying for YC's Spring batch! Applications are open till Feb 11.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: