[theamk]

Neat idea, shame about all this misleading information though.

The website very strongly implies (1) the only way to predict words is by uploading every keystroke to central server,(2) every keyboard other than theirs do that and (3) the only way to get privacy is to download their product.

This is all complete lies of course. The prediction database is small enough that you can bundle it with app and run all prediction locally. The "upload each keystroke" part is completely optional -- some keyboards have no such functionality, and some keyboards require explicit opt-it via the easily findable checkbox.

It's a pity that the authors of such original ideas have to resort to lies to promote their product.

[techwizrd]

It's quite misleading, actually. GBoard uses Federated Learning [0] so that models are trained locally on the device and keywords are not sent to the central server. Both Google and iOS use Federated Learning, Federated Evaluation & Tuning, Differential Privacy, and other privacy-preserving AI techniques and/or on-device ML to avoid sending user data to the server. Both have done so publicly since 2017 and have published many papers and talks on the subject.

0: https://arxiv.org/abs/1602.05629

[pshirshov]

A keyboard shouldn't have network permission at all.

[jstanley]

Or, better, the software should serve the users (and not the other way around) and then you don't have to care what permissions it has.

You don't need to tolerate software that you have an adversarial relationship with.

[kadoban]

Since there's very little way to verify that, permissions are a better metric. This was especially true before Google basically neutered permissions.

[michaelmior]

> The prediction database is small enough that you can bundle it with app and run all prediction locally.

I don't think this is necessarily true. It really depends on what you're trying to predict. That said, I would agree that it should be 100% optional for your keyboard to have network access.