What about “Attention Is A You Need”, the white paper that kick-started the whole generative AI and LLM thing? OpenAI’s GPT models are built off of that architecture, like pretty much everything that came after it.
Like sure, modern Google is notoriously bad at building the actual end products, but it publishes a lot of research and some of it pushes others forward quite a lot.
Some generally older software has buggy multi threading implementation, and forcing its threads to execute “interleaved, but never simultaneously” often improves stability or even completely hides the issue.
Also I know some games used to have big performance issues if they had too many hardware threads available, but I don’t know why exactly
My Pixel 7 is only getting 4 years of updates
Or if you don’t count the initial year, it’s just 6 years for 8+
Is it real that Arch users are rich and have mid level PCs and is always up to date?
I do have a powerful desktop running Arch (or rather EndeavourOS with the EndeavourOS packages slowly removed over time), but also a 15 year old netbook. There’s plenty of up-to-date lightweight software in Arch repos, you don’t need to limit yourself to old software.
Also it’s not like a new version of GIMP is suddenly going to use double the memory if you use the same tools in it.
The biggest issue for the netbook is web browsing, but that’s just about web pages being resource hogs nowadays. uBlock Origin and NoScript help massively. Trying to use an old web browser is IMHO a terrible idea - you’ll end up with broken pages and a massive pile of security vulnerabilities.
It didn’t have to be this way
Yup, I know exactly what this video is going to be about, definitely not a click bait title, lol.
The topic is interesting, the video apparently not so much (if it needs that title to do well). So I think AI is acceptable here, given that no one else bothered to provide a summary so people don’t need to waste time watching it.
This is seriously cool. Although the current specs are a bit funny - if you take the largest possible disc size (360 TB), then it would take a million hours to fully write it at 10 MB/s (more if the storage unit is actually TiB). That’s over a hundred years.
Also if the goal is datacenter archival, then I wonder what the plan for practical use is - many individual discs with separate write “heads” (basically a RAID 0) to bring the speed up? And then maybe the maximum size per disc is a theoretical limit for the technology once we get faster access rather than something practical they intend to build anytime soon.
Are you sure about that? DHL GLS (the shipping company Valve uses in the EU) has an app to move your delivery time or reroute the package, but it wouldn’t allow me to delay my SC package because “the sender” blocked it. So it sounds like a Valve problem to me here in EU.
A roguelike written in my lisp, where the magic system is a lisp.
This is so much better and more eye catching description (first line of readme)
Wayland does force clients to be able to cope with a compositor that doesn’t do SSD - CSD support is mandatory, SSD optional.
2017, actually. But the search engine (supposedly the last innovation according to the previous comment) has seen most of its leaps forward in late 90s to early 2000s, which is a lot earlier.
Also it’s pretty hard to find a more clear example of Google sharing well done research than the white paper that started the current AI bubble. But they do publish a lot of stuff.