Because that's all they know how to do.

Google search does a lot more complex processing (crawling, historical weighting) and it does it on unstructured pages. Discoveravility in a microblogging system doesn't have to be a lot more than indexing/collation of structured posts, and end-user clients can be designed to participate in that. "Retweets"/"boosts" are already a form of that.

You're presupposing that discoverability requires some degree of centralization.

Has there been research into some hierarchical attention model that has local attention at the scale of sentences and paragraphs that feeds embeddings up to longer range attention across documents?

But who gets john-smith?

I never did understand why this path/file structure was exposed.

What if there are two people named John Smith? Or two restaurants called Main Street Diner? Or two towns named Springfield?

Much of that 5M are hardware register definitions expanded into C headers. I am not sure how you'd consolidate that but it's not like that's all bespoke C code.

MS has deep ties into the state department and intelligence apparatus that few other companies do. Just as deep as the defense contractors who have a near monopoly-monopsony relationship with the federal government. You can argue about how exclusive they are in particular qualities but the scale and depth they operate at makes their relationship approximate the relationship Huawei does with the Chinese government. They're just what state-backed enterprises look like under liberal-ish capitalism.