Which feature would Paul Graham most like to change on Hacker News?

Sometimes it's tempting to up vote a comment just because it aligns with my view point. I think it's to do with that.

The default amount of Karma for posting or recommending a link shouldn't be added to your 'score' until someone else actually votes it up.

Downvotes should cost the downvoter karma (perhaps not 1:1).

"Me too!" Follow on comments shouldn't get lots of karma.

"No, your broad generalization fails in one particular case" follow on comments shouldn't get lots of karma. Sometimes I feel like Hacker News is full of anal retentives, or aspies just waiting to pounce on anyone and massively upvote the most minor 'correction' - even when it turns out that the 'correction' is in fact incorrect.

Consider this - theoretically a post by PG could sit at 5 karma, and the reply to that post (from some unknown and unqualified freak from the far ends of cyberspace (such as myself)) could hit hundreds or even thousands of karma.

Anyway, given the above, I suggest that in addition to not adding default karma to karma scores (which encourages certain kinds of spammy behaviour) upvotes for replies should be capped to some multiple of the upvotes for the original point. A 2:1 multiplier for the cap would be interesting, particularly in the case of replying to a post that already had negative karma...

pg: I believe one of forrst.com's main value is its high quality contributions and faces a similar issue as it grows. They just recently released a few short guidelines to ensure that posts continue to be of quality. Maybe HN would benefit from a similar approach for comments/posts?

forrst guidelines:

1. Are you seeking critique and feedback from the community?

2. Does this teach the community something they may not have known before?

3. Will this foster constructive discussion?

4. Is this something development- or design-related the community will find interesting?

Ever looked at how Opinion Space works? (http://www.state.gov/opinionspace/) They build a very simple profile that roughly reflects a given user's belief system.

This can give much more insight into a user's upvote - if they're upvoting the comment of a user on the opposite side of the "belief space", this is probably a more significant signal than upvoting the comment of somebody who's in the same neighborhood of the "belief space".

I think it would be interesting to see how it could work in less-political HN instead of the intentionally-political Opinion Space.

I think there could be a solution.

When a article is first submitted, the focus and the criteria for voting is based on the quality of the article (well actually, just the article name). If the article is viewed and looked upon favorably, it garners a vote.

Over time, two things increase as the submission gets more popular: points and comments. Number of points perpetuate the quality of the article. Number of comments demonstrate the quality of the discussion. Well, actually it doesn't. When comments are off-topic, stupid, or mean, they still add to the comment count and portray the increasing quantity/quality of discussion. So should these comments be included in the comment count?

Another issue with the degradation of comments is the display order of comments and their related threads. When a new comment is submitted, it is favored by being displayed in its own area and not tabbed in. Replies and subsequent threads are included under the original comment. This is pretty much standard practice. However, the issue is that new comments and ideas are unnecessarily displayed below the fold in favor of lower quality replies to more popular comments. This is favoring lower quality discussions to newer ideas that may never be displayed to the first-time viewer of the article discussion. So should children of parent comments be displayed automatically by default given this issue?

If there are certain times of day when comment quality is especially bad, then increasing the delay for thread replies might help.

Another thing that might help is to put the flag button next to each comment instead of having to click on link and then flag.

Stop signups and throw out bad users over time. That will stop the problem.

Wouldn't that be like controlling how people should comment? Smells quora to me.

Why not? Can a small change not end up having a huge impact? Of course it can. It's simply a matter of figuring out what that change should be.

I wonder if a more fundamental problem is that you simply don't have the time, energy, or motivation to further develop Hacker News. Sorry to get personal, but I think saying, "I don't want to do it." would be more productive and honest than saying, "I doubt it can be done."

Paul, thanks very much for answering my question!

I think that the gradual degradation of comments is a fascinating problem to think about.

The existing disincentives are pretty clear (over time: topics no longer link to the first page, new comments have a harder time standing out). There's the usual ways to deal with these issues (hot topics section, more clearly distinguishing between old and new comments, ordering discussions by time and not by points) but I'm not sure any of those necessarily promote quality; they just promote the number of comments.

One idea is to make the site deliberately slower.

As for myself, there is so much content on the internet that if a link is slow, I'll just close the window and move on to something else. When the work network is laggy, I close the browser alltogether. And get back to work in my IDE.

This implies two things. Since a slow site would making participation (viewing links, posting comments, reading comments) much more difficult, people would only have an incentive to do so when they really felt it was worthwhile.

The second is that those of us at work would be more productive. Either that or open a tab for Slashdot, but all you can control is the amount of time wasted specifically as a result of HN.

Of course, people will complain that the site is too slow, but you could be "smart" about performance degradation. At what rate has a user opened links?

Often if I'm cruising away clicking articles and then the network slows, I'll stop and think "Man, I have 6 tabs open. Maybe I should just close the browser and get back to work." Or "Posting this comment can wait. I'll just do it later" and never get back to it unless I really felt the need to contribute.

This is consistent with what you yourself have posted about with respect to internet-vs-non-internet computers. Rolling your chair across the room (45 seconds) is equivalent to a 45-second network delay - possibly enough to get your head back into the game.

I'd personally like it if you could visually see who was part of YC quickly (different colour nick for example).