I think the reason for the misconception is mostly sample bias. People don't see most of the cases that get silently fixed and/or don't realize that we fixed them. Only the ones that seem wrong stand out.
"The only way we can tell if a newly created title is accurate is to read the article, and we're not about to read every article submitted to HN. The only option is to revert to the original title, which is at least what the author intended." (https://news.ycombinator.com/item?id=6572466)
After that, it seemed like people kind of gave up the fight and just started leaving the original title more often.
(I don't mean to argue with you here. I gladly accept that it is not the way that post seemed to indicate. I'm just trying to identify the source of the disconnect.)
The only thing we changed about titles since PG handed off moderation duty is to emphasize that it's ok to take other language from the article, such as a subtitle, a photo caption, or a representative sentence, if the formal title doesn't represent the article accurately. It's nearly always possible to do this, so it strikes the right balance between wanting informative titles and not letting submitters editorialize.
But even that is barely a change. PG mentioned subtitles in the post you quoted.