Very interesting results so far, because the dominant impression I get from this sub is that a single AGI will take over everything.

I personally think multiple companies or groups will develop different AGI through different methods, and they'll all be valid in their own ways. I don't think there's any one route to AGI, and even our own brains vary wildly from one another. It would actually be nice if we had such variety, so maybe a particular cognitive architecture could be paired with an individual to best help them, either because they operate similarly or very differently depending on their needs.

As for the form it will take, that's hard to tell. I think at first it may take a small supercomputer to develop it, but by the time it's ready for public use, computers will have changed a lot, and maybe we'll have similar specs in a much smaller package. If it's little more than software, it should be able to adapt, and hopefully we'll be able to install it on just about anything that can support it.