But that’s not what the review says:

“It’s actually worse at writing than GPT-4.5, and I think even 4o”

So the review is not consistent with the PR, hence the commenter expressing preference for outside sources.