There was a time - prior to the 1990s? - when computers were very expensive compared to corporate drone salaries. A lot of techniques were developed to design and implement large-scale systems.
At some point, people will rediscover this literature and realize that the LLM can take the place of the corporate drones and that you can design systems that are arbitrarily large yet at any particular layer and component, the necessary context is small enough that an LLM can succeed.
Of course, at that point the question is going to be one of self-disciple and not falling asleep from the boredom.
bediger4000 · 3h ago
The LLM generated code I've seen is extremely mediocre, novice tier. This reconciles with LLMs having ingested astronomical amounts of GitHub code. Let's be honest, that code is virtually 100% novice tier. LLM written code is therefore patterned after garbage.
Will we be satisfied with garbage code everywhere?
stockresearcher · 1h ago
Fair enough :)
However, I do think that we will see that project size is going to end up not being a constraint to producing shippable LLM-generated code.
For better or for worse.
pavel_lishin · 1h ago
> Will we be satisfied with garbage code everywhere?
I've worked with some code bases where that seemed to be the case :)
techpineapple · 4h ago
If this isn't fake I'll eat my hat.
fcpguru · 4h ago
eh, i think you just need to write a better prompt to fix what's wrong. Over and over. Repeat. You do need some dev skills with how to write the prompt well. How to even know what it's doing wrong. It's not that you can't get prod ready code via vibe. You just need tip-top dev knowledge and prompt writing skills. So yeah, you kinda do need a "dev" but you can learn these new AI dev skills is a lot of ways.
pavel_lishin · 3h ago
Isn't this like telling someone who did their own plumbing in their house to just keep cutting away and replacing pipes, despite the fact that their basement is flooded?
fcpguru · 2h ago
no, they do need some knowledge of plumbing to talk to their robot AI plumber and explain what to do. Think about luke talking with R2D2 to fix something complex on the ship. Is luke a "dev"?
zzzeek · 4h ago
huh? it's in the name! vibes, dude. vibes are awesome
At some point, people will rediscover this literature and realize that the LLM can take the place of the corporate drones and that you can design systems that are arbitrarily large yet at any particular layer and component, the necessary context is small enough that an LLM can succeed.
Of course, at that point the question is going to be one of self-disciple and not falling asleep from the boredom.
Will we be satisfied with garbage code everywhere?
However, I do think that we will see that project size is going to end up not being a constraint to producing shippable LLM-generated code.
For better or for worse.
I've worked with some code bases where that seemed to be the case :)