GPT 5.6 has similar language quirks that makes its comments nearly unusable.

I wonder if this is a side effect of MoE models — they can write excellent prose, but not simultaneously with writing code.