Interesting. Did not expect it to be this low. Wow.
Discussion
Simple things, e.g., hello world, are fine. But if you get in a little more complex architecture, it’s slop all the way down. What can be done in 5 lines of code will balloon into 50-100 lines of slop with bunch of repetition and lack of any architectural understanding
Do you think it’s possible to prompt it down after a functionality is complete?
Absolutely, if you know how it should look and what is required
I don’t… but I know how less lines of code look like 😂😂😂
🤣😂
You can "massage" the LLMs output if you know what you're doing.
You can also prevent a lot of stupid mistakes before it does them by being extremely careful (and technical) with your wording.
Having a large, well written code base that's mostly written by humans and always cleaned up, and carefully reviewed by humans also helps.
But doing all of that gets it to the 5/10 that fishcake was mentioning.
And the more code the LLM writes the closer it gets to that 0 mark, no matter how good you are at prompting.
For an app you don't care about (for example one just for you, or your friends), vibe code your way through it (just please, if I can make ONE request is that you NEVER prompt your way through anything security related).
For a prototype to test out an idea, vibe it up, but then expect to have to throw away 100% of the code and hire a real developer to build it.
And the models are getting better, as is the tooling. But I think it will be a long time before you can look at LLM slop without being able to tell.