Reaction
A critique of LLM-generated code's heap allocation habits
A user argues that pretraining on mediocre code leaves habits that post-training cannot entirely erase.
TLDR
A user argues that LLMs retain coding habits from mediocre pretraining examples despite post-training. They say they run extra critique-loop iterations to clean up heap allocations they consider unnecessary.
Combined views
—
1 Source, first seen ago
