My favorite detail is that the switching was actually too fast, so they added animation to make it easier to understand. Such a great example of how good UX isn't always about making things faster.
You’re right I’m overly dismissive. But there are dozens of papers like this showing some more efficient alternative to attention. There is plenty of evidence pointing to this paper going nowhere.
Efficient alternatives have been coming out for 10 years now. Flash attention is the only one that has really moved the needle - because it’s not an algorithmic compromise. It’s just a smarter way to put the parentheses around the linear algebra operations.
My understanding is that the original M-Disc pitch mattered much more for DVD because it replaced the organic dye layer, while BD-R already uses an inorganic recording layer
For something explicitly meant to be recovered in a worst-case scenario 10–20 years from now, optimizing for boring, widely available readers seems more valuable than squeezing another 50GB onto each disc
I can see this making a lot more sense for photo/video. The Holocron approach is neat when your truly irreplaceable data fits on a handful of discs, but I would not want to burn and manage dozens of them every year
Yeah I'm tending to shoot around 1TB of data per wedding I cover now that video is in fashion for it. Thankfully it's exactly the sorta thing tape is suitable for - once I'm done I can just archive a couple of weddings off per tape. It's generally very unlikely I get requests for that data later (once they've ordered prints) so doesn't make much sense keeping it on spinning disks in my NAS.
So the mitochondrial part of the "reset" is less like rejuvenating an old population and more like starting from a maternally inherited population that was largely kept out of the metabolic action in the first place
reply