Hacker News new | ask | show | jobs
by halflings 2880 days ago
First, one thing: the people working on these projects are not dumb. They are not going to launch a feature that simply does not work. These teams run experiments and confirm that the feature actually does what it's supposed to do (e.g improve battery life) before launching them in production.

It sounds like this feature is exactly made for cases like the one you describe: if you don't use Snapchat throughout the day, it won't stay in memory like it currently does (which seems to be ~ a LRU cache), instead staying in memory only when you're predicted to use the app over and over. (so not Snapchat in your case)

2 comments

> the people working on these projects are not dumb. They are not going to launch a feature that simply does not work

I've seen no evidence of this. Plenty of features that do not work or work poorly are launched every day.

Especially machine learning stuff, which is hard to test for normal (and abnormal) users and debug the recommendations. See e.g. the Nest. I know people who got one and hated it since it never get the hang of their schedules and would make bizarre decisions. Others love them.

Yet I get ads for something I bought weeks ago, I get ads saying "sign up for Facebook" even though I'm logged in on the very same phone, and Facebook's "advertising profile" is completely backwards for my political views. When I accidentally click a certain type of YouTube video my recommend fills up with that type of video even if I watched it for 5 seconds. Same with Amazon recommended.

In my experience, "personalized recommendations" are universally trash

Personalized recommendations make or break certain businesses like media and e-commerce. Not unusual to see an increase in revenue between 10% and 30% after adding recommendations. They're not "universally trash", since the alternative is showing you truly random items, or whatever item is the most popular.

Check YouTube's "trending" tab and come back here to say if you'd rather have that instead of the personalized recommendations.

YouTube's trending tab is algorithmically generated, not the actually most popular. That's why is called something vague like "trending" and not "most popular".

It is trash, but that's what you get when you use data based recommendations.