Join devRant
Do all the things like
++ or -- rants, post your own rants, comment on others' rants and build your customized dev avatar
Sign Up
Pipeless API
From the creators of devRant, Pipeless lets you power real-time personalized recommendations and activity feeds using a simple API
Learn More
-
@retoor I must admit you got a point, unfortunately, I am involved in some huge projects and companies don't like their code to travel around AI companies.
OpenAI is only allowed because they have a collaboration.
Occasionally for some of my private project I tried copilot with all the experimental models and it produced what I said in the post.
Might want to give codeium a shot but I'm not too fan of paying, therefore I think I will try some nice extensions I saw around for vscode, and run deepseek r1 or qwen locally with ollama and use those in place. Will try also to see if free codeium tier (if they got one) suits my needs.
Sigh, too much stuff around to try
Thanks for the other POV I needed -
I agree with the notion of AI. The next model is either so smart it steals everyone's job, i. e. an AGI, not only from programmers or programmers will be safe.
And still it'll steal our jobs.
A substantial amount of dev work is pumped into bullshit like web3.
This money is running behind trends and for a while it will not be on the market to hire developers with but fund GPU mega server farms.
When you wanted to know deep learning immediately
That's not AI.
AI is dumb and is not going to rob your work as a programmer.
Expanding on this:
https://devrant.com/rants/12459112/...
Don't know about the others, but programming and IT is mostly safe unless you're a secretary answering to mails pressing 1 keystroke at time with index finger.
Bullshit.
I’ve tried EVERYTHING. As a developer, I know exactly what instructions to give and how to explain them. I tried this stuff for years.
I abandoned the idea to give Ai a full blown workspace to vscode with copilot, even with experimental LLMS (Claude 3.5, Gpt4o, o1, as per my linked post, copilot is dumb as a rock), because it fucks up every fucking time so bad.
I tried getting an AI to build a simple project, something that has plenty of samples of code around, something that I was sure it could have been in its training dataset. A copy of Arkanoid, in HTML/CSS/JS, even reformulating the prompts over and over with different LLMs that claim to have reasoning abilities. I provided detailed feedback step by step, pointed out the errors, improvements, and problems in-depth to: o3, o1, 4o, deepseekv3+R1, and Qwen 2.5 Ultra. I even activated web search and suggested scanning GitHub repos when necessary. I gave examples of code after several failed attempts.
And guess what? Nothing. A total mess. Half the time, the game didn’t even run, and when it did, everything was wrong—bricks overlapping, barely anything working the way I asked. Even though the internet is full of similar code, and I gave it part of the solution myself when it couldn’t figure it out.
Don’t worry, AI isn’t going to steal your job—it’s just a broken toy. Fine for repetitive, simple tasks, but nothing more.
It's years that they make hyped up bold statements that the next model will revolutionize everything and it's years that I get delusional results.
It's just good at replacing some junior bovine work like mapping some classes or writing some loops with not too much variables and logics involved.
Sigh. My error was getting too comfortable using it and trusting/hoping that this ramp up in AI developement would have brought an easier life to dev.
Silly mistake.
devrant
llm
ai