Vibecoded this after seeing models do amazing things but still drift on simple recursive steps; tracks exact match, answer accuracy, prefix correctness. Feedback welcome.
Linux page cache + PHP-FPM OPcache already keep hot PHP in RAM (no per-request disk hits after warm-up), and if your dataset fits in memory you size innodb_buffer_pool you don’t move the whole MariaDB datadir to tmpfs and throw away durability.
Definitely helps. MacOS was using about 10 GB of RAM by default, and with rust-analyzer and other projects running the system now swaps much less — so my SSD should last longer.
That folder is 5gb on my Air, at 300-400mb per video.
I have a category selected that says it's 22 possible videos, and out of those only 18 are currently cached on my machine. It looks like that directory is kept pruned at least occasionally.
Nonetheless, just an indication of the potential size on disk would keep users aware when they pick these screensavers. An explicit "Screensaver videos" category in the disk usage graph is also a good solution.
I mean, how else did you expect it to play a random assortment of high quality videos as a screen saver? It's good these aren't bundled, and very few people have metered internet.
Thanks, works well but slowly on a Mac Air M3 with 24gb. Will have to try it again after freeing up more ram as it was doing a bit of swapping with Chrome running too.
(later). It did nicely for the default example text but just made weird sounds for a "hello all" prompt. And took longer?!
Looks like they used it to scrape the DMs for the @doge account, as they are calling for people to report "fraud" there. Crazy that the US government is allowing a third party such as X to have such control.
