https://9to5mac.com/2024/01/24/next-generation-siri-ios-18/

https://www.macrumors.com/2023/12/07/apple-iphone-16-microphone-upgrade-siri-ai/

Writing in his latest Medium post, Kuo says that "strengthening Siri's hardware and software features and specifications is the key to promoting AI-generated content," adding that Apple's generative AI ambitions and integration of large-language models (LLMs) into Siri

This didn't happen today, but it seems that with iOS 17, on-device LLMs will be doing computation on-device for their autocomplete function.

iOS 17 makes iPhone more personal and intuitive

Apple today announced iOS 17, a major release that upgrades the communications experience across Phone, FaceTime, and Messages.

https://nickarner.com/notes/the-coming-of-local-llms-march-23-2023/

If the size of the model doesn't matter then calling it a large language model is pretty confusing. What is the actual standard here, because afaik Siri is already using language models so on the face of it this should resolve yes now.

@vluzko I'd say it's referring to the new architecture type of this models

@FranAbenza LLMs are not using a new architecture - Attention is All You Need was published in 2017. I cannot find confirmation that SIRI is using a transformer, because Apple publishes absolutely nothing about the technical details of the algorithm, but it would not be surprising if SIRI is already transformer based.

@vluzko afaik Siri is powered by Wolfram. But probably uses a transformer model to glue everything together. I am aware of the 2017 paper. But chatGPT for example adds some expansions to the architecture concept

There are not enough Indians in the world to replicate gpt to other services.