As an aside, using an LLM results in a post that's all black text, except a couple of colorful emojis that jump out:

> Speak with confidence. “I think that I can possibly consider agreeing” . Don’t share weak decisions or opinions. Either agree, or disagree.

When I see those I mentally downgrade the whole text.

I'm surprised their 1GB file wasn't cached entirely in RAM during the attack, eliminating the SSD from any timing. Do people keep their machines that heavily loaded that a file being constantly read from doesn't stay in the cache?

Or mine where the entire browser profile is on a ramdisk.

If you compressed it with H.264, it wouldn't make much sense to send it remotely to be encoded with a better codec.

Why not? If h.264 is the best you can do with minimal resources, you can give it 5x the final bitrate and send it to a specialized/beefy encoding system to become something better.

The context was remotely encoding the video, which would require sending the uncompressed stream.

I think the context was intended to be "encoded in some fashion on the upload, just it not as AV2 until after the remote end does all of the transcoded variations". I.e. upload as 2x target bitrate AV1 once and distribute as 1x target bitrate AV2 1,000 times and you'll get the same quality without having to encode AV2 locally.

I've actually done a version this for some multi-system live AV at an event before. Between the main software mixer workstations at various fields in the event it was a dumb but simple encoding they could do in hardware at a high bitrate and then in the machine compositing for the livestream out it did AV1 software encoding to upload to the streaming site to minimize bandwidth requirement from the venue and maximize quality on the streaming site. We've since upgraded to hardware with AV1 encode though.

The practical downside is AV2 is only providing a 30% advantage over AV1. For the streaming providers their bandwidth costs are pretty cheap compared to revamping the transcoding infrastructure, so it'd probably only make financial sense once the remove end can do the most complex and quality encoding used and the rest are all simpler.

I'm reading that L1 takes 4-5 cycles to read on modern CPUs, whereas it was just one cycle in the late 1980s.

It kind of reminds me of the equality of opportunity people versus the equality of outcome people. One sets the starting conditions for developers, the other the ending conditions for users.

You forgot the shareholders, who are not users but have less freedom under the GPL.

BTW there's a summary today that has bullet-point markup and it just wraps around without proper formatting. Title is "Show HN: Tiny-vLLM – high performance LLM inference engine in C++ and CUDA".

I declined things like rm -rf because the path was relative and it wasn't showing me the current directory. How would I know what project it was in?

The individual franchises can always get the clue and drop being a franchise, just be a local store.