Is it worth the 20$ a month? will it be able to make me simple python scripts to boost productivity?
for instance, if I want to fetch a certain dataset using a REST API and dump it to a folder in a formatted CSV, would it be worth asking chatGPT to do something like this for me?
how capable is it?
 |
 Ape Out Shirt $21.68 |
|
llama 3 70B can probably do all of these and you can use it on huggingface chat or nvidia chat page freely without registration
yeah llama3 is amazing all should use it
>Am I stupid-
yes
I'm paying $9 to Sourcegraph and nothing more. I have access to everything from OpenAI, Anthropic, plus Mistral. It's quite good.
>boost productivity?
This is why you eat lunch alone
Im a NEET of 4 years and Im actually trying to speed run a beefy github portfolio to get a career started
Good luck, bro. Go for React. It's shitty, but there's plenty of jobs. Also do this: fullstackopen.com.
same but 10y
same but 30y and the money my mom left me when she died is about to run out so I have to do something, but tech companies don't want to hire someone that hasn't had a job in 30 years and doesn't have anything they can show on their resume.
i think we just need to network and build a portoflio. I know many people who did it and don't even have a degree
Just start your own business bro. Jobs are shit anyway.
using AI to code is like asking somebody to stroke your dick while you frick so it takes less time
True, I want to learn how to code and I have frequently used AI to generate code for me. I have learned significantly less than I would have if I had done it on my own.
same goes for IDEs
it's only good if you want a very quick intro
when i learn a language i dont want to slog through 100pg of documentation just to know how to create a tiny program
so i'd rather see a tutorial or an AI example of a fun little project
Got my degree with it plus side projects lol. In a few months I'd have a pretty impressive portfolio and get started as a freelancer
I sometimes use it to come up with test cases. I don't trust it for anything else
>he pays for inferior Black person tier software that cucks paypigs all the time
Do not fricking pay a single cent to OpenBlack folk because you will get fricked by their shitty load balancing implementation. I lost counts on how many times that they will give you server errors and I wasn’t able to get shit done. Unironically for consistency sake it’s better to leech off Bing Copilot or use a local llama as a freetard or subscribe to Claude 3 for their really nice Opus model.
TLDR do not use ChatGPT until the moronic hello world homosexuals at OpenAI figured out how to watch an intro to aws class
chatGPT sucks, takes too much time. But copilot is great at autocomplete.
I use the local models + free versions. Claude has the best coding output for my needs.
Just look at the the workflow:
Chat:
>write some code
>switch to chat window
>type a explanation to the AI, what you are trying to do, what env (and what versions) you use
>paste in the code you have
>get response
>if it's wrong, explain why (Probably wont work, because it's last incorrect response is in context)
>copy paste the code back to editor
>repeat
VS. with in-editor completion:
>write some code
>it autocompletes what you wanted to write
>if it's wrong, type a character or two (For example, you wanted file.open(), then write "f")
The later is so much faster. The only time when it's worth using chat (as opposed to using nothing, or a search engine) is when you're completely clueless. If you can code, it's a waste of time. The problem is, when you're clueless, you'll get stuck when the AI messes up, which is often.
Not only that, but the same model will be more intelligent and more capable when it only has to complete code, rather than also "talk", the hints you can give it is much more powerful than telling it what to do and it doesn't have incorrect answers in context.
chat is nice for a higher level design where I just have to fill in the code myself
You're supposed to get whole functionalities from the chat code. If you have a good design, it's really easy to get proper code.
Often when it doesn't work, you just copy/paste the error message or find out yourself that it had a minor oversight and get it fixed in your next prompt. At this point if you can't get anything working from GPT4 it's clearly a skill issue.
I am getting increasingly annoyed with it. The first few things I asked it to do it seemed fine. Extremely simple stuff of well defined problems, like convert a number to roman numerals. Asking it to pack three floats into an uint10_n2, it gives me something different every time, none of which worked. It even seems well defined but apparently it is not. I was hoping to use it to sort of go into autopilot for these sorts of problems, but being completely focused is the best way. The problem was that my vectors could be negative, yet the type I was packing was unsigned; so I had to convert it on one side and when unpacking convert it again. It was unable to see all the intricacies of the problem.
It's not even a hard problem though. if I had just focused it would have been done in minutes. I wanted to be chill though and let the ai do it. End up wasting a lot of time.
I just use brave Leo for free. It's handy for simple stuff