I think that in some situations individual settings may help, but generally global settings are more useful.
Imagine a site with 100 users where 1 obvious troll comes and starts posting. Which option is better: (a) the first five users downvote the troll's comments so they become invisible for the rest of the users, or (b) each of the 100 users sees the troll's comments and must remove them individually? Now imagine dozen trolls.
I believe the former is better, because it requires 5% of the work to achieve the same result. And we need to get a good work-to-result ratio, especially if the site becomes more popular, it will attract more of the worst kind of users. There are people ready to post thousands of stupid comments. There are people ready to make a new user account every few weeks (perhaps using the same name and appending a new number) to get out of everyone's personal killfiles. There are people ready to use proxy sites to register dozens of user accounts. I saw them on other websites. The more popular a site is, the more it is exposed to them. So it would be good to have mechanisms to deal with them automatically, globally.
For example, current LW would be vulnerable to the following kind of attack: someone (one person using proxy servers, or an organized group -- just imagine that we attracted an attention of a larger mindkilled group and we seriously pissed them off) registers hundreds of user accounts, posts some comments, upvotes each other, accumulates a ton of karma, downvotes everyone else. A complete site takeover, possible to be done by a simple script.
There is a simple mechanism that would prevent that: Don't give new users rights to upvote. A new user may only post comments, until they get for example 20 karma from the existing users; and only then they are allowed to upvote others. For more safety, introduce a time limit: you have to get 20 karma points and then wait another week, and only then you are allowed to upvote. -- A similar strategy is used by Stack Exchange: users get rights gradually, so there is a limited damage new users can do. You pay for your rights by contributing the content. And it seems to work.
EDIT: And if you want to invite someone important, who wouldn't have the patience with the rules, you (website admin) can simply create an exception for them, so they can post their article immediately.
Is Less Wrong, despite its flaws, the highest-quality relatively-general-interest forum on the web? It seems to me that, to find reliably higher-quality discussion, I must turn to more narrowly focused sites, e.g. MathOverflow and the GiveWell blog.
Many people smarter than myself have reported the same impression. But if you know of any comparably high-quality relatively-general-interest forums, please link me to them!
In the meantime: suppose it's true that Less Wrong is the highest-quality relatively-general-interest forum on the web. In that case, we're sitting on a big opportunity to grow Less Wrong into the "standard" general-interest discussion hub for people with high intelligence and high metacognition (shorthand: "intellectual elites").
Earlier, Jonah Sinick lamented the scarcity of elites on the web. How can we get more intellectual elites to engage on the web, and in particular at Less Wrong?
Some projects to improve the situation are extremely costly:
Code changes, however, could be significantly less costly. New features or site structure elements could increase engagement by intellectual elites. (To avoid priming and contamination, I'll hold back from naming specific examples here.)
To help us figure out which code changes are most likely to increase engagement on Less Wrong by intellectual elites, specific MIRI volunteers will be interviewing intellectual elites who (1) are familiar enough with Less Wrong to be able to simulate which code changes might cause them to engage more, but who (2) mostly just lurk, currently.
In the meantime, I figured I'd throw these ideas to the community for feedback and suggestions.