These are described as "subprocessors", so generally, this means that Persona is (potentially) sending any data they receive to these companies/platforms.
I'll probably get vote-murdered for this, because this is unfortunately not a popular opinion for a lot of very justified reasons that I actually mostly agree with, but I'm going to throw this out there anyway, and I hope people hear me out for long enough that you can decide for yourself instead of just kneejerk downvoting.
Imagine if someone created a statistical numerical model that was based on, and could therefore approximately reproduce something close to the cumulative total of all human knowledge ever recorded on the internet which probably represents exabytes of information, but this numerical model was only the size of a few movie files, and you could dump those numbers into a simulator that within some margin of statistical error, reproduced almost any of that information on currently available consumer-level hardware.
If you're not picking up what I'm putting down, I just described open weight LLMs that you can download and run yourself in ollama and other local programs.
They are not intelligences and they do not represent knowledge, because they don't know anything, can't make their own decisions and can never be assumed to be fully accurate representations of anything they have "learned" as they are simply greatly minimized and compressed statistical details about the information already on the internet, but they actually still contain a great deal of information, provided you understand what you're looking at and what it's telling you. The same way demographics can provide a great deal of information about the world without needing to individually review every census document by hand, but never tell the entire story perfectly.
While I agree with the suggestions to get a proper encyclopedia or just download Wikipedia, for a more reliable and trustworthy dataset, I think you're doing yourself a disservice if you dismiss the entire concept of LLMs and vision models just because a few horrific companies are hyping them and overselling them and using them to destroy the world and civilization in disgustingly idiotic ways. That's not the fault of the technologies themselves. They are a tool, a tool that is being widely misused and abused, but it's also a tool that you can use, and you get to decide whether you simply use it wisely, or abuse it, or don't use it at all. It's your call. It's already there. You decide what to do with it. I happen to think it's got some pretty cool features and can do some remarkable things. As long as I'm the only one in charge of deciding how and when it's used. I acknowledge it was plagiarized and collected illegally, and I respect that (as much as I respect any copyright) and I'm not planning to profit from it or use it to pass off other people's work as my own.
But as a hyper-efficient way to store "liberated" information to protect ourselves against the complete enshittification of content and civilization? I don't see the harm. Copyright is not going to matter at that point anyway, the large companies who control the data and the platforms for it have already proven they don't respect it and they're going to be the ones dictating it in the future. They won't even let us have access to our own data, nevermind being able to do anything to prevent them from taking it in the first place. We, the people and authors and artists and musicians and content creators it was designed to protect, now have to protect ourselves, from them, and if that means hiding some machine learning models under my bed for that rainy day, so be it.
Unlikely. Power supplies usually have internal protection, and as a result, if they become overloaded, they will trip off (and the whole computer either shuts down or reboots). Is it possible the internal protection is not working? Maybe. But it is far more likely the issue is with other hardware, or even more likely, with software/device driver issues. Try booting a LiveCD/LiveUSB with Linux on it or something and see if the problem goes away.
Because Microsoft is trying to use it to fracture and control the Minecraft community, divide and conquer. It's a lazy ripoff of real Java Minecraft, locked down with DRM and restricted to platforms Microsoft blesses, and the only significant improvement is more graphical sparkles (not as much as modded Java though) and new modes (not as many as modded Java though) and new creatures (not as many as modded Java though) and new blocks (not as many as modded Java though) and new dungeons (not as many as modded Java though) ... are you sensing a theme yet?
It also has all kinds of paid DLC and microtransactions and advertising shoved into it which they want you to believe are somehow "better" than the countless variety of mods that have been developed for free by the community (they're not)
I agree that quadlets are pretty ugly but I'm not sure that's the ini style's fault. In general I find yaml incredibly frustrating to understand, but toml/ini style is pretty fluent to me. Maybe just a preference, IDK.
Subtractive colors like paint create color by selectively removing some colors from existing light.
Additive colors like backlit or light-emitting displays create color by creating colors of light in various proportions that are then combined.
If you are in a dark room, all paint is black. Until you turn on something with RGB, because then you have some light for it to selectively absorb. However if your RGB is only displaying green light, and you shine it on red paint, it will look exactly the same as black paint (within a certain ballpark of imperfect materials, anyway). Green paint will look green, or white, depending on how your eye adapts, and green and white will be indistinguishable.
That's the difference between the two color models. Does it rely on other light sources (subtractive), or is it a light source (additive)?
How the brain actually perceives color is really, really wild, so this is all a bit... fluid when you start getting into the weird edge cases, but the general principles of additive=light emitting and subtractive=light absorbing are generally applicable.
I think ActivityPub is closer to the right answer than ATProto, and ActivityPub's issues (though many, as the author notes) are more manageable in the long run. I think the article makes a good analysis of the fundamental differences, but is a bit glib in referring to Piefed's topics and discussion merging as a "joyful mess". It's not a mess at all. It's making order out of the chaos, and it's the right way to build on top of ActivityPub into something that is actually fluid enough for users to actually use.
Mailing lists were built on top of federated email in much the same way, and they formed enduring, resilient, well-structured communities, some that continue to this day (the LKML being perhaps the most notorious)
I think ATProto makes creating enduring communities too difficult, and BlackSky illustrates that perfectly. The author's criticism of ActivityPub, on the other hand, seems to be that it makes creating communities too easy, and this results in a "mess". I disagree, I think the mess is a necessary and inevitable part of having community. Communities are messy. They fracture and schism, they rejoin and reshape themselves. That's normal. It is the responsibility of the software to make sense of the mess and make it presentable, and with ActivityPub, that is not only possible, it is happening. Piefed is the present example. I expect there will be more examples, and a wider variety of them, as the ecosystem continues to develop.
I think the biggest thing that ActivityPub still needs is better portability, for both users and communities, to allow moving servers more seamlessly. The "Personal Data Server" of Bluesky is not a bad concept, although I don't love their implementation. I think ActivityPub can find a way to handle portability even better, but it doesn't seem like it's been a priority, and that's fine. But it will need to happen eventually.
Americans have had business interests blowing smoke up their ass for so long they just think it's constantly foggy. Canada's not much better, but we do our best given our proximity to source of this madness.
Mistral is European? I had no idea. I just assumed it was from either China or US, since most of the cutting edge open models are. But Mistral/Ministral/Devstral have actually consistently been among my favorite open, local models. Now I feel even better about that.
Despite our governments (well, one of our governments) the Canadian and American people still care for and look out for one another.