Corey Ward @coreyward.dev · Jan 27
1 / 3

First, the heavy hitters: the reasoning models. ChatGPT o1 and o1-mini rewrote it to use `concat` without changing the inefficiencies. The readability is worse IMO, but fine. DeepSeek R1 used `slice` instead of array expansion, but that's it. None of these are better at all.

0 likes 1 replies

?

Replies

Corey Ward · Jan 27

Next, ChatGPT 4o, Claude 3.5 Sonnet, and Amazon Nova Pro all arrived at the same solution that changes the behavior by reversing the whole array without reversing the order of the inputs. This introduces a bug and would depend on either a careful PR review or a test suite to catch it.