I get that, and I agree with it in general, but there’s literally no company on earth that would approach open source developers with the intent to pay them to work on a closed source product, or to buy out their open source work without having an NDA in place. Hell, even if Meta just wants to pay them to do open source work to support the community, there will still likely be an NDA covering what they can say to the public about the arrangement or anything they learn from having access to internal systems.

It’s like saying “Meta has security guards at the doors to their datacenters! They must be doing something illegal in there!”

Meta is evil and is very likely doing something bad with these developers, but the NDA isn’t the smoking gun evidence of evil… It’s Meta’s history in general

I don’t know why everyone is so upset about the NDA thing… It’s such a standard business practice. Whenever I (a mid tier infra engineer at a mid sized software company) needed to talk to a vendor, get a product demo/consultation, get support on a licensed application, etc… We either sent an NDA to that company or bad one on file already with them. Nobody discusses internal processes, policies or roadmaps with an outside contact without an NDA first. It’s literally just a standard business practice.

It could be nefarious, since it’s meta afterall, but I wouldn’t be shocked if there’s thousands of people/companies who have standing NDAs with meta just so they could come on campus and demo their product to some team

I’m guessing it’s based on the new Instagram twitter-alternative that’s rumored to be based on the mastodon code base. But I’ve also heard rumors about Instagram wanting to make it on a rust code base and have been investing internally into rust a lot more lately, so it wouldnt surprise me if they’ve been consulting Lemmy or other fediverse devs for their experience

It’s not the tech here. Postgres can scale both vertically and horizontally (yes there are others that can scale easier or in different factors of CAP).

The problem is how the data is being stored and accessed. Lemmy is doing some really inefficient data access and it’s causing bottlenecks under load.

Lemmy (unfortunately) just wasn’t ready for this level of primetime yet… It has a number of issues that are going to be quite tricky to fix now that it’s seen such wide adoption (database migrations are tricky on their own, doing so on a production site even harder, doing so on 8k+ independent production sites… Sounds like a nightmare)

Nobody trains on a GUI desktop though. Training is done on a cluster. And the kind of models that can be run on a consumer grade GPU… Nvidia doesn’t care about. They’re focused on selling 50k a pop cards to AI companies not fixing the Linux desktop for $600 card users.

It’s pretty clear that Linux users should buy AMD or Intel GPUs if you want to support even a semi open source world