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.