Matrix is neither a good protocol, nor a replacement for Discord.

Why is it not a good protocol? Most problem stem from it not delivering messages, but synchronizing a DAG, including message delays, synchronization issues and the inability to reliably edit or remove messages. Sometimes rooms do not even reach the same state.

Why is it not a Discord replacement? Because Discord servers can be influenced a lot by bots in ways Matrix does not implement (and that would be hard to implement with Matrix' architecture). At the moment I think Matrix cannot even hold a voice channel connection when you change rooms.

The alternative may be Revolt, that actually tries to clone Discord. It doesn't federate, but mumble (or even teamspeak) don't either and still make good platforms for teamchats. And in the end, a server would be an actual server again and not a misnomer like it is with Discord.

Define your niche, find a good 7B-30B models and you outsmart GPT-3.5 (beginning with 70B sometimes even GPT-4).
Most claims to outsmart GPT-3.5 in all disciplines with a single model are false.

Who claims that the quality of sound is worse, shouldn't fear the AI, should they?
Also there is no need to steal voice. Future AI voices will be exactly what the musicians want them to sound like. Singers should be glad if someone needs their voice, because it cannot be finetuned and adapted for each song.

> Just put in a picture of some non-human object as the standard, and be done with it.

You can't do that when someone writes about issues regarding human perception as it may be important to show a face. Just let's think of how many JPEG artifacts you can add while people still recognize the image shows a human. It might be much harder for people to still recognize a table than to recognize a face as our visual system is well trained to recognize even partial images of faces. So there can be no drop-in replacement that works in all papers that use Lenna as reference.

Snaps do not have someone responsible for them. The package maintainers of Debian packages are developers trusted by the project. In snap and flathub each software has its own uploader who isn't vetted by anyone.

Package maintainers follow the development of the programs they are packaging and notice if something is fishy. How often did Debian package a malicious software?

The problem is, that Ubuntu tries to establish an Appstore instead of using a package repository and now they get the problems that Appstores have.

Most the time your phone will be turned on and then Apple can install updates anyway.

But for the binaries I got under GPL, I can still ask for the source. Even ten years later. It doesn't matter if later versions weren't GPLed anymore.

Yes. But it doesn't say it has to be distributed the same was as the binary as long as you get the source in a reasonable way. So they may, for example, e-mail you a zip with the source if you ask for it. They don't need to host it as long as there is a way to request it.

You still need to give people who already got the binary the corresponding source.

Google advanced AI quite a lot. What's bad about doing do? They both saw a trend and they helped progress. Now they are in a good position. What's the problem with that?

And I don't understand Microsoft either. While Google has good hardware and good models for, e.g., image recognition, it's generative AI are failures. Gemini sucks while MS does have a good partnership with OpenAI who create GenAI that actually work. Yeah, ChatGPT is getting worse (probably because they are optimizing costs), but Gemini isn't quite usable at all.

If you don't know, why does it matter how the artist you like to pay creates its art?

It won't get ready without distributions shipping it. How are you supposed to find bugs without having users?

Have a look at the KDE folks. They had to ship a completely broken 4.0 to get developers to fix their things to work with the cleaned up architecture. It got usable around 4.2, but many Linux distributions shipped 4.0, so they got their feedback. Now, plasma 6 is greater than ever, still building upon the cleanup that happened between 3.x and 4.0.

Apple is working toward this on the desktop. Try to run unsigned software on macOS. They make it more and more complicated. If nobody forbids it, you won't be able to run software that wasn't signed by apple on your mac in ten years.

StackOverflow had committed itself to give back cc licensed DB dumps. They stop providing them some time ago, justifying it with LLM training when asked and re-enabled them after they got a bit of a shitstorm for no longer giving back to the community. The site content itself is cc-licensed anyway.
This means, that if they continue to provide the dumps, people could just train on the dumps. If they do not continue, people can legally (given that attribution is provided) crawl the site itself.
And finally, one should think that one of the points in "giving back to the community" would be allowing the users to train LLM on their (own) data. The whole deal with that was users saying "We accept SO being a silo as long as the silo regularity provides us dumps of the data".