Hacker News Reader

Comment by HenryBemis | Hacker News Reader

Nope it cannot.

I have been using 3.5 to write code. I wrote a 'simple' app (200-300 lines) that although 'simple' it took A LOT of back and forth with the 3.5. In the end I had to spend a couple of days combing through it to add some key features (i.e. set values to zero so you don't see old data, etc.). Basically crude (but smart?) coding (I am not a professional developer).

When I finished the code and the app was performing EXACTLY as I wanted (v1.0), I thought to start planning for the v1.1 and when I pasted my code to the 3.5 and asked it to 'work on it on the new features' it was a shitstorm :)

Perhaps it was me feeding the requirements 'wrongly', but it was like pulling teeth 'guiding' it to correct even the simplest mistakes.

View on HN

This is far too flippantly confident about a piece of tech that's barely a year old, and the inferior version of that.

I'm surprised by the inability of people in tech to extrapolate the advancement of this particular tech.

They replied to this:

> Even chatGPT 3.5 can be better than them.

I maintain that ChatGPT 3.5 can be better than entry level Infosys engineers.

It's like showing up at Kitty Hawk in 1903 and, "You mean this thing can't even fly me to New York? Useless!'

I think there is a difference between mastering fluid dynamics to the point where you can have sustained power flight, and building a machine that arranges words next to other words based based problematically in training data.

But even if you want to torture this analogy, LLMs are a godsend to techncial service Companies. It's like saying "Oh, planes are faster so I guess people won't be spending as much time traveling and there will be less of it now".

I think you'll find the most succinct requirements for a program is the code itself - you can add more and more logic to guide GPT to generate something but eventually you're just writing code yourself anyway.

This is also how it is with offshoring.

GPT can sew snippets of code and see patterns but it has no understanding of coding and such. But even without this ability it could work amazingly well if you know what to ask for and accept its limitations.

Someone who's not a professional developer was able to take the weakest model OpenAI offers today and get exactly what they wanted for a V1.

Have you actually worked with a consultancy? Have you priced out what the back and forth you had with 3.5 would have cost with a real human with a similar level of depth?

I personally don't think AI's effect on offshoring will be so cut and dry, since it's not like offshore developers can't level up with it too, but your example is showing how badly people are underestimating the impact AI will have on programming.

Try 4.0. I use it a lot for tedious glue code, and I couldn't be happier to have my time back to focus on more important things.

Reply on news.ycombinator.com