Hacker News Reader

Hacker News

Comment by dang | Hacker News Reader

Full thread

dang

Recent related threads:

Brave disables Chromium FLoC features - https://news.ycombinator.com/item?id=26765084 - April 2021 (335 comments)

Am I FLoCed? - https://news.ycombinator.com/item?id=26755313 - April 2021 (33 comments)

Google’s FLoC Is a Terrible Idea - https://news.ycombinator.com/item?id=26344013 - March 2021 (348 comments)

I know there have been others - if anyone finds them, I'll add to the above list.

View on HN

There's also the EFF website, "Am I FLoCed?" https://news.ycombinator.com/item?id=26755313 - April 2021 (33 comments, 144 points)

Thanks! Added.

I want to write software to let the community build these lists. Doing things like this is good practice for figuring out how the software can work.

That's a really cool idea!

A bit off topic but I'd love to see a sort of ML/AI feature where "related discussions" can be added automatically for added context, I appreciate you do this manually and I find it very useful to get more of an idea of the relevance/importance of a topic!

(Maybe even a "HN search found 3 possibly related threads")

Why would that need ML? PageRank has worked for ages for this problem.

I'm not sure what you think PageRank is, but it doesn't produce links between documents as output; it requires them as input.

I would consider PageRank as a quite successful example of machine learning. Am I wrong?

It started out that way, anyway.

The definition of what counts as ML/AI can be fuzzy, but I don't think the original PageRank algorithm would qualify as ML/AI under any definition: http://infolab.stanford.edu/~backrub/google.html

The goal with those 'related' lists is to get something more meaningful and interesting than HN Search results (which are pretty good already). That's a high enough bar that it takes judgment and taste to know what to include.

For sure we want to get beyond "mods do it manually" as the strategy. I'm not averse to trying out an ML/AI approach—the nice thing is that all the relevant data is public, so anyone could build it and we could all try it out. But my personal inclination is to build something that would rely more on the community.

Reply on news.ycombinator.com