Rembrandt
the artist
It's not live yet and do you have gpt+? Only available through that
Y'all heard about ChatGPT yet? AI instantly generates question answers, entire essays etc.
- Thread starter RamsayBolton
- Start date
It's not live yet and do you have gpt+? Only available through that
IIVI
Superstar
Yeah I have GPT plus, but can't copy/paste or an option to upload. I guess need to wait a little longer.
yes just keep logging out and in, if you are a pro user eventually you'll see the update
IIVI
Superstar
I've been one of the ones here who have been advocating for how insane ChatGPT is for us developers.Yet
Netflix started with shipping dvds and look where they are now
Buh buh they need me all this code
They needed the milk man at one time to
It'll still be awhile, until that point though people will still have their jobs.
It still ain't perfect though and still can't put everything together. GPT-4 is great, but it still falls short from some testing. It's an awesome idea generator though.
Like I just tested a code rewrite/small refactor and it failed.
Copy and paste code without knowing what it's doing and that's a potentially devastating bug in the system
.It got it right after some more iterations though and some specific guiding.
Like I said too, a CEO don't want to put everything on an A.I.
They'd be absolutely insane to because once those bugs hit, that A.I is detecting no errors and a lack of good programmers = BANKRUPT company.
Last edited:
Yeah obviously but I could see coming to a point where instead of say like 5 people on a team it's only 1
That makes sense, thanks for explaining. So how high do you think this could go in terms of the things it can do?
IIVI
Superstar
Yeah, I think honestly it'll be like that 80/20 rule. Using A.I/ChatGPT/etc. a company will only need the top 20% to run everything.
shyt, you may only need the CEO, CFO, CTO, etc. if everybody is that intelligent at what they do.
yup go to your local mcdonalds just a few years ago you'd have like 5 people on cash now when I go there is just one and everyone orders through those touch screens. But you still need that cashier for the few ol school folks or if the machine breaks down etc.
The scary thing is it still being in the infancy. Right now its like iphone 1 just dropped with 8gb storage space
Sauce Mane
Superstar
I can’t imagine what ai will be capable in 10 years
IIVI
Superstar
Right now it's great for small projects, scripts and building in small chunks. If you can orchestrate that, then you can have a real nicely maintained system. ChatGPT can't scan an entire codebase, but you can keep posting code for it to analyze, rewrite, etc. and piece it together. Right now it's basically incredible for "trees for the forest" use cases.
If you want more of an integration test (put multiple parts together) you can make multiple posts to kind of prime it for later questions, something like:
The bigger the codebase, more requirements, etc. that's when you'll basically run into some problems.
You do have some tools like Copilot that can look at more areas of your specific codebase, but it still kind of misses context and some real world requirements.
In practice there is code out there that has clean, straightforward logic and is works as intended. However, when in action it'll do exactly what someone doesn't want.
For example, I know of this "bug" at a company/app where somebody could download a pre-loaded version of the unreleased program (many gigabytes) to do all the downloaded upfront, then the app will download a little more on release date to trigger an unlock. That way you can get ahead of everybody downloading on release day. However, there was an decision made that if there was a corruption or error it was designed to delete the stored data.
Now because some connection errors occurred when people were downloading on release day, the system saw that as an error and did what it was programmed to do: delete everything. So everyone's pre-loaded apps got deleted because of bad network connection (release date they got slammed) and they had to redownload, but because download speeds were slow af customers were pissed it the company ended up losing a lot of business and dishing out many refunds.
Nobody could find the bug, because there wasn't a bug: it worked as intended (delete the app on any error). An A.I may tell you at that point it's all working fine (all the automated tests passed): the code is working as intended and you're trippin. This is of course as people get mad and money starts pouring out. The issue was found because someone basically said "Hold up, you wanted it that way didn't you?!" Issues like this are actually quite common and many "bugs" are basically because the feature is working correctly, it just missing some context and edge cases. My friend worked at this company, said the amount of finger-pointing chasing that issue down was legendary.
An A.I can only interact with the code itself, but the "side-effects" in the real world are something it has no context of, and that's a limitation for right now.
If you want more of an integration test (put multiple parts together) you can make multiple posts to kind of prime it for later questions, something like:
The bigger the codebase, more requirements, etc. that's when you'll basically run into some problems.
You do have some tools like Copilot that can look at more areas of your specific codebase, but it still kind of misses context and some real world requirements.
In practice there is code out there that has clean, straightforward logic and is works as intended. However, when in action it'll do exactly what someone doesn't want.
For example, I know of this "bug" at a company/app where somebody could download a pre-loaded version of the unreleased program (many gigabytes) to do all the downloaded upfront, then the app will download a little more on release date to trigger an unlock. That way you can get ahead of everybody downloading on release day. However, there was an decision made that if there was a corruption or error it was designed to delete the stored data.
Now because some connection errors occurred when people were downloading on release day, the system saw that as an error and did what it was programmed to do: delete everything. So everyone's pre-loaded apps got deleted because of bad network connection (release date they got slammed) and they had to redownload, but because download speeds were slow af customers were pissed it the company ended up losing a lot of business and dishing out many refunds.
Nobody could find the bug, because there wasn't a bug: it worked as intended (delete the app on any error). An A.I may tell you at that point it's all working fine (all the automated tests passed): the code is working as intended and you're trippin. This is of course as people get mad and money starts pouring out. The issue was found because someone basically said "Hold up, you wanted it that way didn't you?!" Issues like this are actually quite common and many "bugs" are basically because the feature is working correctly, it just missing some context and edge cases. My friend worked at this company, said the amount of finger-pointing chasing that issue down was legendary.
An A.I can only interact with the code itself, but the "side-effects" in the real world are something it has no context of, and that's a limitation for right now.
So mainly, the TL;DR:
Non-nuanced coding small projects/snippets: insane. Absolutely a monster.
Non-nuanced coding bigger projects: insane with somebody guiding it.
Left by itself to do nuanced coding: potential liability.
If you're new to programming, it's probably one of the best tools you can use to teach yourself how to code and find out what's wrong with your code. Much better than stackoverflow imo.
Last edited:
no one can..
