kline 🏴󠁧󠁢󠁳󠁣󠁴󠁿 @kline@cmpwn.com

Feeling pretty pessimistic about AI just now. Not because it's evil or anything though.

Neural networks seem like an evolutionary dead end that deliver results basically as a brute force if you have lots of data, compute, and money.

Even in the mid 00s you could put a bounding box around a head for nothing on a shit laptop at 30FPS. Now I'm looking at a "state of the art" pose estimator that needs a $1500 GPU and gigs of ram and manages about 20FPS, of which half the time is spent on putting a bounding box around a person.

And this is where we're throwing all the money, making the brute force slightly faster and fiddling with unexplainable, inscrutable models, rather than pushing to find brainy rather than brawny solutions.

As an aside, this is 150x the largest fine ever applied under the GDPR recorded on wikipedia.

If you want to see what effective legislative punishment looks like, you should always look to VW: https://www.bbc.co.uk/news/business-52795376

€30 billion in consequences already for dodgy software, and executives in the frame, despite trying to put developers under the bus.

Now is also the time to look at licensing software engineering: the developers at the time knew this was illegal, but didn't have the tools to push back. Licensed engineers know the buck stops with them and that the consequences are real, and have a much better history of being able to push back and know they wont be undercut.

web development with lots of routes for different actions - instead of just a blob of JS - feels like writing a really shit choose your own adventure novel.

Watching another license debate and I don't think I've ever made my own position on the most reasonable setup:

* For applications: copyleft licenses such as the (A)GPL.

* For libraries: permissive, such as the MIT license.

Why? For applications, it's important that your application doesn't get picked up wholesale, extended, and your original userbase extinguished, without reaping any of the benefits a commercial enterprise might stick on.

For libraries, it's important to build a base of high quality software on which applications can be built. As more and more applications (proprietary or not) build on these quality, free software blocks, the fewer bugs we can hope to see, and the easier it will be to interoperate with other applications or build alternatives.

Why does Kuenssberg look so damn tired?

Keybase, the company that asks you to upload your private keys to their servers, has just been acquired by Zoom, an essentially Chinese company notorious for having terrible concepts on how encryption should be implemented.

Even if you gave Keybase the benefit of the doubt beforehand, this is corporate suicide at it's most graphic. Delete your Keybase keys. Close your account. Rotate everything that Keybase touched, be that password or cryptomaterial.

https://blog.zoom.us/wordpress/2020/05/07/zoom-acquires-keybase-and-announces-goal-of-developing-the-most-broadly-used-enterprise-end-to-end-encryption-offering/

Very disappointed that Chromium devs consider even the last bastion in basic automation to not be worth un-breaking in their software.

https://bugs.chromium.org/p/chromium/issues/detail?id=393145#c15

man, why is polybar so buggy