This comment was the last straw that made me fire off an email to dang regarding the recent decline in comment quality with a few structural changes that I think may help; let’s see what comes out of it.

Your account was created in 2023 and you talk as if you have been on this site since the beginning of time.

To be fair, I read HN regularly since 2016 but only created an account since the reddit fiasco looking to engage more actively. Could be a similar situation.

I’ve been a reader of this site since 2018, but only have I recently created an account.

Then you should know that people have been frequently mentioning Reddit since 2009, which is why it's even mentioned in the guidelines.

The Redditization is real this time though. Just because something wasn’t true previously doesn’t mean it continues to be that way.

"Please don't post comments saying that HN is turning into Reddit. It's a semi-noob illusion, as old as the hills."

Read the guidelines.

what changes do you propose?

We train a LLM classifier using a HN dataset and a Reddit dataset to automatically flag the Reddit like comments.

People can then decide to hide them in their settings.

I’m not sure when to put the cut-off date for the HN dataset. The eternal September started in 1993 and HN in 2007, we need at least a few comments.

There’s no need for a cut-off date; just take highly upvoted comments off Reddit and find similar comments on HN as determined by word vectorization, and then train a model off of that. Or maybe even skip the similarity step; after all the goal is to find Reddit like comments.

Primarily caps karma earning limits and voting limits, which should (at least I think) nudge people towards more thoughtful voting patterns and comments.

The unintended effect may be that high-contribution users are disincentivized from participating.

Be patient. Submissions pick up dumb comments early, because they're quick and easy to write, and better comments later.

It's true; I won't start reading a thread less than a day old, if I'm planning to read-only.