Typically 100-500 per page released on our new website this year. The data isn’t great: two complications are that people can link to pdfs directly (e.g. we have no viewing data for the movement growth paper), and some of the articles have been cross-posted elsewhere in their entirety.
A couple of recent posts have gone rather higher: around 1,500 so far for the flowchart, mostly from people sharing it on social media, and around 1,800 for AI safety calculator, of which half were referrals from Slate Star Codex.
The importance of page views for our thinking depends a lot on the piece. In some cases the major route for impact we see from the work is influencing future research directions—in that case we care more about opinions from engaged researchers and people referring back to or citing the work than we do about page views. In other cases we are trying more directly to inform the thinking of a larger community. In this case page views seems very relevant.
Some places that page view numbers have fed into my thinking:
When we want lots of people to interact with the research tools we produce, we need to lower the activation energy. This post had quite a few upvotes but not many people contributing or visibly using the tool. The AI safety calculator was based on similar underlying theory but is our most-viewed post—probably because it was easier for people to get stuck in.
The flowchart has been quite widely viewed, but we’ve had repeated feedback that it could probably be shared wider still if in an easier-to-interact-with form. The high viewing figures actually make this more important to act on, because it increases the likelihood that if done correctly it could get a very large number of views.
Our posts which give a small update on the organisation are among the least-viewed, so we should mostly regard these as updating our core audience (exception being ones which advertise positions or internships with us; our post advertising internships had over 1,000 views).
When we want to make a point widely, we need to make that point up-front rather than leave it to the reader to draw implications from the theory. This post on factoring cost-effectiveness rather flopped, but I had a positive reaction to a different presentation of similar material at EA Global.
Typically 100-500 per page released on our new website this year. The data isn’t great: two complications are that people can link to pdfs directly (e.g. we have no viewing data for the movement growth paper), and some of the articles have been cross-posted elsewhere in their entirety.
A couple of recent posts have gone rather higher: around 1,500 so far for the flowchart, mostly from people sharing it on social media, and around 1,800 for AI safety calculator, of which half were referrals from Slate Star Codex.
The importance of page views for our thinking depends a lot on the piece. In some cases the major route for impact we see from the work is influencing future research directions—in that case we care more about opinions from engaged researchers and people referring back to or citing the work than we do about page views. In other cases we are trying more directly to inform the thinking of a larger community. In this case page views seems very relevant.
Some places that page view numbers have fed into my thinking:
When we want lots of people to interact with the research tools we produce, we need to lower the activation energy. This post had quite a few upvotes but not many people contributing or visibly using the tool. The AI safety calculator was based on similar underlying theory but is our most-viewed post—probably because it was easier for people to get stuck in.
The flowchart has been quite widely viewed, but we’ve had repeated feedback that it could probably be shared wider still if in an easier-to-interact-with form. The high viewing figures actually make this more important to act on, because it increases the likelihood that if done correctly it could get a very large number of views.
Our posts which give a small update on the organisation are among the least-viewed, so we should mostly regard these as updating our core audience (exception being ones which advertise positions or internships with us; our post advertising internships had over 1,000 views).
When we want to make a point widely, we need to make that point up-front rather than leave it to the reader to draw implications from the theory. This post on factoring cost-effectiveness rather flopped, but I had a positive reaction to a different presentation of similar material at EA Global.