I never understood why people use Tailscale

I use it for the NAT busting and direct connections. This means that my devices can talk directly to each other, even when there’s NAT and dynamic IPs sitting between the devices with no port forwarding. This is not possible with Wireguard alone; usually you end up with a hub and spoke network model.

As for them man-in-the-middling, the client is open source (for Android and Linux at least) and traffic is end-to-end encrypted. If you don’t want to trust them with distributing the keys (completely valid concern) then it’s possible to configure things such that you must sign the keys of clients yourself for your devices to trust them (see Tailnet Lock).

In my case, because I like self-hosting, I self-host an open-source coordination server called Headscale. So in at least my circumstance I really am only using my infrastructure and open-source code.

I think you missed the point of his post. His issue is that the numeric operations the phone executes to run the LLM is producing garbage. Arguably this could break all kinds of neural networks, such as voice transcription. He’s not complaining that the LLMs are themselves unable to properly perform math.

Sadly, at least in the North American market, Google’s Pixel phones are basically the last good phones you can reliably install your own ROM on.