Conversation
I reached the stage of complexity in this project where the LLM declares victory too early to avoid work
3
0
5

is that better or worse than it getting stuck a loop burning tokens doing nothing?
@sun

1
0
0
@p this burns know-nothing vibe-coding "founders" but I just immediately detect it and tell it to keep working because I know what I'm looking at.
1
0
1
@p I'll show it off soon because I'm getting close to something workable.
0
0
2
@sun Does the llm ever make utterly wrong or bad assumptions and theg end up propagating?

I see this a lot when using them as a information tool

Code is kinda a best case scenario for them
1
0
0
@meeper I usually catch it before it does, but that's a you don't know what you don't know scenario too. straight up, on this project I have not manually reviwed a single line of code and it works great.
1
0
2
@sun Kinda surprised that some invisible errors don't propagate.

I've seen some extremely disgusting hallucinations in even modern models, so I feel they could be making major technical debt at least in enterprise stuff ig
1
0
0
@meeper local models do this, the saas models are getting pretty advanced
0
0
2
>LLM declares victory too early to avoid work

They are learning fast
0
1
1