Everything, Everywhere, All At Once would definitely also be my recommendation. Because a kung fu comedy that’s also a sci fi thriller, a bilingual family drama that can switch languages multiple times per sentence, a Ratatouille parody, and a 10 minute silent shot of two rocks in a desert, that just SHOULD NOT WORK. The fact it does, and does brilliantly, with ten new directing ideas every minute and a climax that leaves me in tears every time, borders on genius.
- 0 Posts
- 9 Comments
Neat! Thanks for the explanation. :)
CapyReader (synced to FreshRSS)
Oooh, what does that look like in practice?
Balinares@pawb.socialto
Ask Lemmy@lemmy.world•Do you know any privacy-friendly DNS resolvers?
2·1 month agoI mean, that’d be a major GDPR breach, be hard to extract any signal from because queries will usually be coming from a relay or from behind a NAT so you can’t tell who the query even originates from, and DNS is cached heavily too so you only get a small fraction of the queries anyway. I’m not seeing a way the calculus work in favor, basically.
OTOH the question of why they’d even run a public DNS is interesting, yeah. Running a public DNS is cheap and helps the Internet work better, and they make more money when the Internet works better since that adds up to more page views. Less charitably, though, it’s possibly just a thing from back when they were an engineering company first and foremost and did that kind of stuff, and now they can’t turn it off without breaking a lot of things and causing a lot of costly anger.
Balinares@pawb.socialto
Ask Lemmy@lemmy.world•Do you know any privacy-friendly DNS resolvers?
4·1 month agoI’ll admit I’m not sure what the threat model is with 8.8.8.8.
I’m mildly offended that I can consistently be reduced to a blubbering mess by a show with literal fart-based plot points.
This is abhorrent, yet I’m intrigued.


They invented a hybrid attention design that drastically reduces the amount of memory needed for the KV cache at inference time. Like, dividing it by 10. And memory is a large part of the cost of inference.