Tests of the New AI Siri | Hacker News Reader

Tests of the New AI Siri

pogueman.substack.com

19 pointsMBCook15 comments

I’ve been using it off and on since DB2.

David Pogue is right, it’s very impressive. He and basically everyone else make the same comment, and I fully agree: it’s hard to remember to even use it.

I used Siri a ton before, but always for basically the same four commands. Create reminder, call X, send a text to Y, set a timer for Z minutes.

And after years of that, it’s hard to remember to do more. But when I do it can often shock me.

It’s not fast. Which is quite annoying for those quick actions I used to do. But I’m OK with that on my phone. Not happy with it on my watch.

It’s not ChatGPT 5.6. And I’m sure they’ll get it faster over time. But in my mind they finally delivered.

I was just rambling about it to my wife - it's finally good and it finally can understand me correctly.

Apple always had access to app's internal API for AppleScript and accessibility, so now that they exposed it to an LLM (with upcoming claude and gemini support), they are a serious power. And iPhones are in the hands of a lot of folks around the globe. Exciting.

I seriously doubt they’re going to give anyone else access. They’re fighting the EU over it. But I guess we’ll see.

More than once has has told me something, and I had to ask how it knew. And it would tell me exactly how and what do you know it was there somewhere on my phone.

A lot of that list sounds like the "ai, please press the order me pizza button" meme.

Some of it is. He did the classic “book an entire vacation for me“ except he only asked it to plan.

No one is ever going to just tell a phone to book a vacation for them. It won’t know what you want well enough. Planning is much more reasonable because then you can just adjust whatever you don’t like.

But as a demonstration, it certainly shows how far Siri has come. It probably would’ve just offered to do a web search before.

The things I’ve found it really useful for so far is the “find this piece of information that must exist on my phone somewhere, but I don’t have the slightest clue where“ kind of query.

> Also, this is ethical AI: It deceives nobody, puts nobody out of work, and draws 100% of its power from renewable sources.

I like how this is cast as a selling point instead of tacit admission that Apple is not on the frontier.

I don’t see what those two have to do with each other. Could you explain?

If Apple was a frontier lab, this article would not begin with "This is unethical AI, it deceives people, puts people out of work, and draws 100% of its power from coal and LNG."

It's the same flavor of ethics-washing that we saw with Mac LLM inference. Many people rushed to call Mac inference "efficient" when it put up worse per-watt performance than cheap GPUs. It was a marketing platitude to make people accept subpar performance. Same for calling Siri "ethical AI" too.

Is this some sort of avant-garde sarcasm that is meant to be funny? I'm genuinely asking, I am having trouble modeling a live human being that reads any amount of llm-related news and says this without it being facetious.

It's an observation of how people respond to marketing. Apple objectively dragged their feet on LLM software and hardware research. After 3 (!!!) years we finally have a good LLM Siri, and the first note that this guy makes is "oh it's ethical" to cover for Apple's wasted time.

I find it pretty funny, but the author probably doesn't. Americans ought to know better than to utter "ethical" and "Apple" in the same breath.

I think this is good and not just whatever.

But hey perhaps i still don't understand your point properly

Here's a better example. Compare these two marketing tones:

Low confidence: "Our engineers work tirelessly to ensure that frontier research doesn't step on the toes of users. It's a deliberate touch to keep people in the drivers seat, emphasizing the importance of human coexistence with technology. Apple Intelligence can never replace our users or the personal touch they bring, and that's what makes Apple products so special to us all."

High confidence: https://variety.com/2024/digital/news/why-apple-ipad-ad-cont...

You realize this isn’t an Apple marketing piece, it’s an extremely well known tech reviewer who used to work for the NYT and was one of the two(?) who was allowed to test the iPhone before its release.

Actually think that’s a good thing for Apple. Compare how much money companies have absolutely shoved into the AI race, and how much circular investment is happening. Apple is one of the few massive players which hasn’t massively extended themselves and taken on a ton of financial risk. How do you make money on frontier models? By selling it to software companies, I guess.

Apple just needs a capable personal AI that respects privacy. They could eventually get that by self-hosting open model inference on their private cloud compute. They don’t need to light billions upon billions of dollars on fire to build data centers and do training to make a frontier model that’s nearly already a commodity. If they’re not selling tokens (and why would I want them to), that approach seems financially stupid and risky.

> If they’re not selling tokens (and why would I want them to)

That’s one of the two things that has stood out most to me. Siri AI answers my questions. And that’s it. She stops.

Google, OpenAI, and Anthropic all constantly try and get you to answer follow-up questions. Because if you do that, you’ll use up more tokens and get more “addicted“ and end up buying a plan.

Apple doesn’t have one. I guess iCloud, but let’s face it, they already pushed that a ton anyway. They’re not trying to get you to use it for the sake of use. It’s not trying to be your friend.

And just like David Pogue says, it doesn’t lie. I just asked it an incredibly stupid question. Can you get pregnant from a picture?

“While I'm not a doctor, no, you cannot get pregnant from looking at a picture of someone. Pregnancy requires physical contact and the transfer of sperm to an egg. For accurate information regarding reproduction, consult a healthcare professional.”

Look at that! It immediately says it’s not a doctor. It gives a good answer, and then it says you need to seek out real advice to be sure things are accurate. Of the summer, I also got responses that ended with something along the lines of “I am an AI, please check information for accuracy.”

I’ve gotten incorrect answers, but they come with disclaimers and it doesn’t act like some omnipotent god that is desperate to be your friend.

Reply on news.ycombinator.com