@nprofile1q... We're opposed to it on privacy grounds rather than based on whether functionality is implemented using a neutral network. We don't have a fundamental issue with using a locally implemented neural network for specific tasks such as auto-brightness, text-to-speech, speech-to-text, etc. However, GrapheneOS is open source and that means any included features implemented that way need to have open source training data rather than being so called open models which are in fact closed source.