I would like to make a broader comment: perhaps as users we should ask what AI features even make sense to use? Starting last year I experimented heavily with Google Gemini interacting with Google WorkPlace apps. The technology was cool, and is even much better now, but I came to the conclusion I don't really need it.
For an iPhone local AI, I wrote an app for myself (although I think there are maybe 10 other people who use it) that chats with Apple's local model (that is fairly good) and switches to a Secure Enclave model on their servers and from the documentation it looks like using the cloud model is private and secure.
Even better now, I signed up for ProtonMail's optional Luma LLM Chat system with integrated private web search tools. It is surprisingly good, and I trust Proton that it is private.
Almost the only thing I frequently use commercial LLMs for now is a few times a week using gemini-cli for coding, and NotebookLM a few times a month, plus occasional Gemini use, but I pay for Luma (powered by Mistral models) so I routinely use it for AI search use cases.
Just because technology is incredibly cool, this doesn't mean that we have to use it if real productivity gains are slim or non-existent.
For an iPhone local AI, I wrote an app for myself (although I think there are maybe 10 other people who use it) that chats with Apple's local model (that is fairly good) and switches to a Secure Enclave model on their servers and from the documentation it looks like using the cloud model is private and secure.
Even better now, I signed up for ProtonMail's optional Luma LLM Chat system with integrated private web search tools. It is surprisingly good, and I trust Proton that it is private.
Almost the only thing I frequently use commercial LLMs for now is a few times a week using gemini-cli for coding, and NotebookLM a few times a month, plus occasional Gemini use, but I pay for Luma (powered by Mistral models) so I routinely use it for AI search use cases.
Just because technology is incredibly cool, this doesn't mean that we have to use it if real productivity gains are slim or non-existent.