I've been thinking about this problem and this is my thought process: 1. A tool like grammarly is super helpful for the majority of non-english speaking popula

I've been thinking about this problem and this is my thought process: 1. A tool like grammarly is super helpful for the majority of non-english speaking population. So, this is not something like Facebook that we can just ignore and boycott. This is a truly useful category of tooling. 2. But, sending everything that the user types over wire and predicting with a heavy model that can only run on massive servers is evil. 3. The only viable alternative is either come up with a tech that encrypts data over wire, such that sentences cannot be decontstructed back at the server, but knows what indices in the client string. should be replaced - I'm not even sure if something like this is possible. 4. Just ship the entire model to the edge, so that no data gets transferred. Google has managed to ship a light-weight model to its Picel devices, apparently but I'm not sure effective this could be: https://ai.googleblog.com/2021/10/grammar-correction-as-you-type-on-pixel.html https://ai.googleblog.com/2021/10/grammar-correction-as-you-...