←back to thread

Let's talk about AI and end-to-end encryption

(blog.cryptographyengineering.com)
174 points chmaynard | 2 comments | 17 Jan 25 05:50 UTC | HN request time: 0.45s | source
Show context
username223[dead post] ◴[17 Jan 25 16:54 UTC] No.42740207[source]▶
[flagged]
airstrike ◴[17 Jan 25 17:25 UTC] No.42740744[source]▶
> tl;dr -- Most AI stuff won't run at an acceptable speed on your computer

...today. There's no real reason we can't get acceptable speeds in the near future.

replies(1): >>42744405 #
1. deathanatos ◴[17 Jan 25 23:37 UTC] No.42744405[source]▶
Yes, but also no: surface area is a proxy for compute power, because it's a proxy to heat dissipation. Phones are also only passively air-cooled — which is like the absolute worst cooling mechanism, too¹. So yes, processors do get better, but watts are watts, and there's a hard upper limit to how many watts a phone can consume, and you have to get better within that limit.

Or … IDK, maybe you don't. Recent Android phones are now capable of prompting the user to actively cool them off because the ~bloat~ software is consuming too much compute/power. What basically translates to "help I'm dying" is an amusing message, but also a depressing state of affairs.

¹compare to a laptop, or worse for the phone, a desktop. The comparative cooling surface area is multiple times larger. And both have fans, desktops can be liquid cooled — it's not mere chance that they're faster, and phone form factor literally presents challenges. Even with an infinite battery, or a USB capable, you can't dissipate 60 W into a human hand.

replies(1): >>42744563 #
2. airstrike ◴[18 Jan 25 00:07 UTC] No.42744563[source]▶
I agree but I was replying to the comment about computers specifically, meaning desktop and laptop form factors rather than mobile phones, especially when I say "near future"