Ask HN: Post-LLM's, do you create web applications much more quickly?

I was writing an SVG editor web app but didn't know anything about the fabricjs api. Here's a portion of my prompts with chatgpt (april 2023 version, the date at the top is the share date). The initial code came from a different chatgpt thread.

https://chat.openai.com/share/e630c5d4-d492-43cb-b7e5-214ff8...

I finished the app in 2 days, with a third day for css/visual styling. I previously might have hired someone to do this or tried to figure it out myself, which would have taken about a month.

At the end of this thread chatgpt kind of goes off the rails a bit and fails to center the uploaded image. I think this is because it can't execute the code and see its results, and can only take blind guesses at what the problem might be.

I still had to write about 10% of the code myself, but it's about 10x faster for me to use chatgpt. I think I'd use chatgpt even if it were slower, because I prefer "thinking in natural language" vs "thinking in code"

here is the final app in production: https://tinyurl.com/368w3a9y

Development time in the end didn't change much - I code all the time, most software I develop isn't or will get done. But the quality and the codebase size I've being able to create and manage is much bigger now than six months ago. I've being coding using chatgpt almost every day since then, many hours a day.

In fact, chatgpt might have made the development time bigger as I'm refactoring it more often and even recreating things from scratch with a much better quality after validating some PoC's. I'm really enjoying this workflow for a side-project.

This may be a bold claim and it's hard to quantify, but based on weekly output I think it has sped up my development by 50%-80%, between researching technologies with chatgpt4, coming up with designs and code, copilot filling in the blanks, and gpt4 helping with troubleshooting. These are things that used to take me hours of scouring the web and docs, time spent on proofs of concepts, and additional trial and error- which has largely disappeared.

With the amount of time and effort saved I generally feel more confident about being able to create scripts or programs to solve problems or create art as a second nature, rather than having to treat each idea as a large project.

Similarly I have seen my brother go from someone who has never programmed to doing beautiful data analysis in python thanks to gpt4, for his own side projects. The speed in which he's learned has really impressed me. Makes me think that this technology will introduce many new programmers quickly.

I only used it once to create a toy browser extension, but I believe it extrapolates.

The key thing it allowed me to do was to skip the initial “how to, even” learning curve. It wrote a manifest and html+js, from which I got that there are permissions. Then I asked again about an extension with a specific permission.

This skipped an hour of learning what a manifest is, how to use it and so on. I’d learn it faster, but with modern google search you have to skim through and close a few bunches of “tutorials”, before you find one that is less full of bs than usual. I couldn’t search directly for a manifest format because I didn’t know it’s a thing.

This is a great value for adjacent areas I’m not specializing in.

I found that Copilot sped up the trivial things (boilerplate, simple APIs, popular algorithms) and routinely messed up the harder tasks (user flows, composing several functions meant to work together, etc.). It was a net negative for me and I stopped using it after a month or two of experimentation.

I did use ChatGpt to replace stackoverflow a few times, and for that it was super helpful. Stack is so full of junior modding closing useful questions left and right with no clear links to the answers. ChatGpt just steals all that content then summarizes the answer, lol.

Overall no faster or slower. But definitely higher quality than humans being argumentative pedants.

Day-to-day coding on a platform that I already know, there not much of a difference. It might save me a few minutes here or there.

Where it does help is when I am learning something completely new. I can ask how to do something, get an example that may or may not work, but that often exposes me to a new way of doing things and gives me hints of what else to study to learn this new thing. That learning curve is significantly sped up.

Yes, though I haven't done any web development in a decade, so it's more of an enabling technology since I couldn't be bothered to relearn everything. I wrote most of an Anki replacement in a weekend recently and i don't want to even think about how long it would've taken me without chatgpt. Realistically I would've given up.

Nope. LLMs are marginally helpful. Often to be LLM worthy a problem has to be pretty desperate because of the high chance of getting a wrong answer. That said as a search engine Phind has been useful. I find it more useful than a LLM alone. But we are talking a 1% saving in time as most problems are better served reading the docs.

I find it immensely helpful for boilerplate code and ‘autofilling’ trivial things.

It’s like a next generation intellisense.

On average, instead of spending about 5 mins copying and modifying code, I can spend about 2-3 mins typing out a comment and letting the LLM generate the code.

It gets the code 100% correct about 90% of time. Sometimes it’s about 90% correct about 10%.

It has made the easy parts easier and faster but most of the time is spent on the harder parts. I think my stuff is more polished now thanks to LLMs since I don't have to write css myself anymore.

Yes. LLMs are basically just a more efficient Google to me a this point. Searches for what I need on the average are way quicker.