Another variant would be to have the weight be voted on. Weights only matter for a fraction of submissions, so I wouldn't want to complicate submission by making everyone (especially new users) come up with a weight for what they're submitting.
Maybe it would even be a good idea to have the votes on weights be public, so no one would be tempted to use that as a way of censoring stories.
Maybe have the HN software go pull the page that is submitted, and assign it a weight based on number of tokens on the page. Obviously you'd remove stop words and HTML tags.
This would assign lower weights to shorter fluff, and lower weights to articles that are split up over a lot of pages (which in my experience tend to be fluff too, with a 4:1 ad to content ratio). It'd be kind of like Bayesian filtering for post importance.
This was actually one of my ideas for submitting to YC, but I rather like HN, so maybe you could experiment with it here.
Maybe restrict users below a certain karma to X number of upvotes per day? That way, you need to consider carefully which stories you upvote. Comment upvotes could stay infinite.
An ideal solution, although I'm not saying it's easy to get it right, would be to measure the weight of the story based on what portion of clicks are long term clicks, i.e., clicks after which people spend considerable amount of time (presumably reading the article) before coming back to perusing other stories on news.yc.
Yes, both of those seem excellent. Could we give it a try and see how it turns out? You could define the acceleration of a downvote as (exp (- (* numdownvotes 0.5))) or some such, to discourage cabal behavior.
What if you did a word count of the linked article, and used that for the weight? It might be good to strip out links from the count, but other than that it seems a pretty sold way of judging complexity.
Maybe it would even be a good idea to have the votes on weights be public, so no one would be tempted to use that as a way of censoring stories.