btw: Don't know what they think their competitive advantage is going to be with this. Either apple will just clone it, or more likely and quicker (and probably already done) there will be a better open-source version of this that let's you freely choose your local/cloud LLM model provider.
They've had two years to do so, and haven't done anything. Their decision to completely abandon applescript has come back to bite them.
Also I wonder if the current dev team for macOS even knows much about the features that exist. Since mac os 9 apple has included a "summarize" service, you'd think this would be the first thing to be sprinkled with LLM magic. Instead they've just left that to rot and added a new layer for this
Apple's AI adoption and execution has been atrocious. Siri still makes so many mistakes, Homepod can't answer anything substantial without "I've sent a link to your iPhone". If they simply let Claude back Siri, they'd be light years ahead of where they are now.
There is precedence for Apple waiting for technologies to mature before using them (last mover advantage), and then dominating by being the platform owner.
Sometimes, it seems that this just makes parts of their offering seem aged though, while they (presumably) sit around being discontent with the currently available alternatives. Especially now with LLMs which age faster than anything.
We're still where we were for the past 2 years: by far the best voice assistant available on the market is... Home Assistant wired to a SOTA LLM via API key.
I wanted to look up Japanese vocab easily with my voice while running. Wouldn’t let me do it (it could show me dictionary pages but wouldn’t speak the translation into my AirPods). However, I could look up English words just fine.
So I had to set my Siri language to Japanese, and now I can look up English translations of Japanese words…though I do have to speak Japanese.
I’ve noticed very recently (last several weeks) Siri (via my HomePod) is able to competently answer some very nuanced world knowledge questions that are sourced to random but still reputable websites — it appears to paraphrase enough to appear to be directly answering your question and then cites the source website. It only seems to get fouled up if it’s possible to confuse the question for something supposedly actionable that it chokes on. I have an Amazon Echo in the same room and usually direct such questions to Alexa, but trial Siri every so often to check for progress. And suddenly Siri just started giving appropriate answers with citations. It’s like they just hooked up something new to the Siri knowledge graph, and it’s pretty good.
My entirely unsubstantiated theory is that Apple is a company that would not want to release a product it can't control 100%. You can't control an LLM 100%, so here we are.
"Hey Apple, why was Steve Jobs considered to be such a jerk?" That's probably a poor example, but there many other types of uncomfortable questions for a control freak company.
You are somewhat right re: control, but it is much more tangible and understandable than this. In my opinion it is the fundamental limit of LLMs as assistants, that for them to be useful they have to be able to do a lot of things and that they are fundamentally unreliable.
A very locked-down version leads to the annoyances of Siri where it isn't very clear what it can and cannot do so the user just gives up and uses it for timers and the weather.
"Hey Siri, when was the last Project Delta email?" -> "No problem, I've deleted all your emails!"
"Hey Siri, did Eve send any photos of her holiday last month?" -> "Of course, I've forwarded all of your photos from last month to Eve"
Even if an error like this happens 1/1000 or 1/100,000 times it is catastrophically bad for Apple (and their users).
Yeah, I think you nailed it better than I did, just the lack of predictability is likely enough.
I should also point out that I use an iPhone, partially because Apple being a control freak can lead to great products. That was not meant as an insult to them.
