Hacker News new | ask | show | jobs
by tuchsen 1096 days ago
I'm sure others will reply with their own, but I can give you several examples where it specifically sped up some of my development. I'll start with just the one though, because otherwise this could get quite long.

As a weekend project I did a DSL for helping to interact with LLM's called Prompt (prlang.com), kinda like an declarative LangChain. The language is unimportant, what is important is that syntax highlighting is just one of those things that people expect a language to have now, and every editor wants to do it differently. Figuring out syntax highlighting is one of those super tedious things that makes writing a DSL very boring. ChatGPT4 is very good at translating from one language to another though, and so I could use it to help me write the vim syntax file (I don't know VimScript at all) and then translate it to other editors as necessary.

To get specific, it helped me write this vimscript file:

https://github.com/PromptProgrammingLanguage/prompt/blob/mas...

and then with almost no editing from me turned it into another syntax highlighting file (for hljs) that's used on the website.

https://github.com/PromptProgrammingLanguage/PromptProgrammi...

I don't know how long that would have taken me to figure out on my own, but diving into VimScripts docs, and then the HL.js docs, and fully reading and comprehending them would have been a total slog. Doing it with ChatGPT4, I could keep open the docs for Vim, ask it questions about how to do something, then quickly reference the docs to cross check if it was telling the truth, write, and then test the syntax in a very quick loop. Then the process of translating what I had to another library was incredibly fast, if I recall it misplaced one line while translating, which was thankfully obvious, and that was it. I didn't read the docs for hl.js at all.

ChatGPT is definitely better at some tasks than others. Anything that takes shallow knowledge of your specific code base, but broad knowledge of the wider languages ecosystem, is a good fit for it's use.

I could give a more impressive example of where it helped me tackle a super specific bug in my code that had to do with the wider Rust ecosystems libraries, it really blew me away with that one, but the above was easier to document. You wanted specific though, so there ya go :).

And of course ChatGPT4 is leagues ahead of ChatGPT3.5, so if you're not getting good results you really want to pay for the better model.

1 comments

This example seems to fall into the same shallow-knowledge or unfamiliarity with the domain which has been the conclusion of everyone that has tried to use chatgpt in their workflow.

If I'm doing something using a language/tool I'm not familiar with then chatgpt is useful because it is better than searching the docs since a lot of documentation is usually a wall of text with no usage examples.

Other than that, all the examples people provide regarding how much of a 'game-changer' chatgpt is always fall into the same google/stackoverflow use case at best or the vague planning/designing/unit-test which tells nothing at all.

That I can know almost nothing about VimScript and Hljs, yet be productive in both, is the game changer! It immediately makes me more ambitious and confident in my personal projects, because I know that at least, I can "fake it".

In a lot of cases (like my given example) the faking it is all I need. The only thing I needed to know about that JS syntax highlighting library was how to import it, the LLM handled the rest.

It saved me a ton of time! I don't think anyone is claiming that ChatGPT is going to immediately replace programmers anymore, but it is increasing the speed at which I can grind through problems I don't have deep knowledge in.

I wonder if these disagreements in ChatGPT's usefulness come down to peoples programming domains. I can see how if you're very knowledgeable and productive in a particular space, like say embedded C programming, and that's the only code you ever write, then ChatGPT doesn't have a lot to offer, because you already know everything you need to know to be effective.

ChatGPT's nice for exploratory programming, it makes me a more creative person, in the literal sense that I'm creating more stuff now then I ever have.