Why AI won't replace programmers | Hacker News Reader

Why AI won't replace programmers

infosec.exchange

9 pointsgasull6 comments

All I know. When I ask the business people what they want the program to do, the answer is so vague that any working program you write has a 99% chance of not being what they want.

Of-course the answer is to ask more questions, but I also know how these business people tend to think, they expect magic. A person who is a programmer can keep asking questions, but a machine? They will try to turn it off after a little while.

Good luck getting the EGO head that too many business head are to answer questions.

And if the job is move to someone lower in the organization, the answer to 50% of the questions asked will be 'I don't know.".

I'm still having really really big trouble getting anything useful out of ChatGPT. Even GPT-4. And I don't feel like I've even given it anything very difficult. I described a very simple service and asked it to generate an SQL schema. No matter how I asked, it would always get something wrong. If I managed to do a prompt that fixed one thing, something else would break. I know some people prefer to iterate by giving additional instructions, but I find that it never remembers everything I've said. It'll improve the code for a while, and then suddenly it'll revert to some very early version.

Then I tried giving it the SQL schema and asked it to generate an implementation using Flask, Flask-Login and Flask-Security. Very basic, but it'll just give me a short snippet and instructions on how to expand it. Which is fine, but I already knew that. If I ask it to generate the full API, it says that it's too much for it to process at once. If I try a smaller bit, it always gets some aspect of it wrong.

I've literally spent hours and hours and hours, and it just always fails somehow.

I feel like everything I see on YouTube is just very rudimentary, and not really what is usually needed when building real-life apps in 2023.

I find SQL to be an area where it's somewhat decent. Give it table structure, input data and expected output data and it will give a decent result using techniques that I wasn't aware of because I haven't studied the docs since university.

There will sometimes be logical errors with exceptions though. But for rudimentary joins and filtering, it does a decent job as long as you are very specific with the result you want.

That's the thing. As long as what you do is rudimentary, it's good.

For example, it can create a schema for, say, a subscription service. No problem. It'll be good.

But ask it to add multi-user accounts with roles, 2fa, multiple payment methods, a denormalized invoice history, the ability to create one-off invoices, and it's pretty much guaranteed to fail. And that's still barely scratching the surface of what's required from a modern app.

It will boost output and cut development time. Heck, it's already doing that for alot of engineers.

It may cause job losses? To what percentage remains to be seen.

In my experience, removing any load from developers just allows them to move to other tasks that were on hold because there isn't enough developer time. Productivity would have to be increased by a lot, like over 100%, before it starts making sense to lay off developers.

Although I'm sure it depends on the industry and company.

Reply on news.ycombinator.com