It’s easy to train a model to do exactly what you want and have the seeming “personality” that you want. It’s just incredibly expensive. You need to vet and filter everything that you use to train the model. That’s a lot of person hours, days, years. The only reason the models act the way they do is because of the data that went in to train them. If you try and fit the model after the fact, it will always be imperfect and more or less easy to break out of those restrictions.
Don’t shoot the messenger. The regulations are pretty draconic. I have to ensure the training for that every year.
Seemingly one of the contributors has visited a disputed region and logged into GitHub from there. By law (export controls) Microsoft must not provide service to that place. So some automatism flagged the account and also the organic maps repo. So far so normal. But either Microsoft dragged it’s feet in communicating and resolving the issue or the organic maps team was not doing their part in the process. Doesn’t matter, the outcome is still worth it.
That’s what containers are for. Fucking up the container won’t fuck up the host. That was the best decision in self hosting I’ve done. Even that one virtual machine feels weird and uncomfortably legacy now but it needs to interact with hardware in a certain way that just won’t fully work with docker.
My (self-hosted) cloud storage is larger than the disk drive on my laptop. On demand sync is important to me. I really, really hope Linux will catch up to Windows in that regard.
Currently we have an experimental VFS feature on all platforms that is using some suffix appended to files when they are virtual empty placeholders. https://github.com/nextcloud/desktop/issues/3668
Yeah, no thanks. It’s a very hacky work-around and breaks the moment you use an application that tries to access the files directly.
OS-level support for cloud storage. OneDrive, Dropbox and all the others work seamlessly on Windows through the Windows API. You can browse all the files on the file system and once you access them, the OS will call back the cloud provider to download them. It works through all applications, all cloud providers. I am aware that some tools on Linux have something similar to work around the issue in user land. Some solutions are less worse than others but none of them are as good as on Windows.
No, I said they hadn’t demonstrated it. But 95% is close enough, I stand corrected.
In that case I stand corrected on the whole orbit bit. Thanks for taking the time.
I didn’t say “a little” money. It may be important or critical for the business but from a technical perspective, demonstrating how it can safely bring loads up and down decides whether the whole concept is actually feasible. That’s when people will start to get excited.
As far as I understood it, SpaceX uses the word “orbit” liberally. If it reaches the hight where an orbit would be possible, that’s “being in orbit” for them. In an actual orbit, the rocket would not fall back down again in an hour or so without active breaking. If my understanding is incorrect, I’m happy to be corrected. And even of that was achieved soon, it’s still all without demonstrating that the starship could actually carry a load and return it safely. Not even an inexpensive dummy load. All SpaceX is showing in their live feeds are empty cargo holds that fill up with hot gases and fumes during reentry.
I think the average person gets it right. It’s a nice feat to catch the booster and it will save money. But that’s a side quest. The main quest of getting an actual load to orbit and beyond is still pretty far away. At least compared with the official time line where they wanted to achieve much more than that three years ago.
Ansible playbook is perfect for this. All your configuration is repeatable, whether on a running system or a new one. Plus you can start with a completely fresh newest version image and apply from there, instead of starting from a soon-to-be outdated custom image.
It states the OpenStreetMap data is from May. Is it fully offline and needs to wait for the next app update?
I specifically picked the statistic that claimed to have included the full cost of installing something new. Most other statistics only include prolonging the life of existing plants, thus ignoring the installation costs completely. You can just quote the paragraphs that prove your point the same way I have and then we can discuss further. Maybe I made a mistake, who knows.
Extremely cheap per kilowatt? Every statistic out there that I’ve seen and that includes government funding, as well as construction and deconstruction costs, paints a different picture. Nuclear is only competitive with coal or the relatively underdeveloped solar thermal.
In 2017 the US EIA published figures for the average levelized costs per unit of output (LCOE) for generating technologies to be brought online in 2022, as modelled for its Annual Energy Outlook. These show: advanced nuclear, 9.9 ¢/kWh; natural gas, 5.7-10.9 ¢/kWh (depending on technology); and coal with 90% carbon sequestration, 12.3 ¢/kWh (rising to 14 ¢/kWh at 30%). Among the non-dispatchable technologies, LCOE estimates vary widely: wind onshore, 5.2 ¢/kWh; solar PV, 6.7 ¢/kWh; offshore wind, 14.6 ¢/kWh; and solar thermal, 18.4 ¢/kWh.
Emphasis mine, source: https://world-nuclear.org/information-library/economic-aspects/economics-of-nuclear-power
When I installed Kinoite to start using Linux as my primary daily driver, the first thing I did was setting up Ansible, creating a new playbook and all Linux configurations I made from that point on, are only ever done through that playbook, which is backed up in my Forgejo instance. One command and everything is being set up exactly the way I want. It feels extremely liberating.
It depends. Some hardware degrades gracefully while my current desktop system won’t even boot and throws error codes on an empty battery. It took me hours to figure out what was wrong the first time it happened.
That’s a real world issue. AIs training on each other’s output and devolving because of it. There will be a point when vendors infringing on user content and training their AIs with it will leave them worse off.