Y
Hacker News
new
|
ask
|
show
|
jobs
by
olig15
149 days ago
Surely because LLM generated code is part of the training data for the model, so code/patterns it can work with is closer to its training data.