Siri and Gemini

[Robert Brown (RLBrown)]

Hello SVMUG members and friends,

You may have read that Apple is negotiating a $1 billion per year contract with Google to have the Gemini AI model installed on private Apple Cloud servers. Here are some interesting points.

First, the integration of ChatGPT (OpenAI) with Siri has been awkward. From the user's perspective, the workflow involved submitting a query, which Siri would attempt to answer using an on-device AI. If it couldn't do so, it would send the query to an Apple AI on a cloud server, with strict user privacy protections. If that still couldn't resolve the query, the user would be asked for permission to forward the query to an OpenAI server. OpenAI assured that the user information would not be retained.

The problem is that almost everything seems to fall into the final stage, i.e., using OpenAI. There were trust issues. Users felt, despite assurances, that OpenAI may not be fulfilling its privacy pledge. Furthermore, having to give permission was annoying.

Second, it appears that OpenAI is moving forward without the necessary caution in AI development. Many experts are concerned that, because OpenAI is pursuing a trillion-dollar IPO, the safeguards for AI progress are being overlooked.

Third, some of us, at least I, find that Gemini provides superior responses as compared to ChatGPT.

This is how the partnership with Google Gemini is expected to function. Apple will be licensed to install a 1.5 trillion-parameter Gemini model on Apple's private cloud servers. The word "private" is crucial here. Just as they do for files, email, messages, and so on, Apple will encrypt queries and responses privately. A user's information never leaves the trusted boundaries of Apple services. The new Siri, set to launch in 2026, will use a billion-parameter model locally on devices and a 1.5 trillion-parameter model on Apple's private cloud servers. 

Users might choose from different AI vendors, but I believe Google Gemini is the genuine option here. 

Full disclosure: I maintain a Google Workspace, I have a Google Developer account where I run Google Compute Engine VM servers, and I subscribe to Gemini Ultra. So you might say I am entrenched in Google technology. 

I also have access to OpenAI, but I find I am using it less and less as compared to Gemini. I've noticed that several other services I use, particularly Adobe and Artlist, are utilizing Gemini to offer AI functions. 

Robert Lawson Brown mail