Comment by namaria

> ... which explains why some models are better at code than others.

No. It explains why models seem better at code in given situations. When your prompt mapped to diffs in the training data that are useful to you they seem great.