Apple finally admits that Siri is old

This post is part of our special coverage Syria Protests 2011

BEIJING TIME, JUNE 9, 2026, APPLE'S WWDC 2026 ARRIVED。

At the launch, it changed Siri to Siri AI, published in-depth collaboration with Google, trained its new generation of basic models with Gemini 's model capabilities, and extended Private Cloud Company for the first time to Google Cloud and Nvidia 's GPU。

It published five Apple Foundation Models, the smallest 3 billion parameters on the end side, and the largest cloud was dedicated to Nvidia GPU optimization. Almost every daily App was rewrited. Siri also has its own independent application, which preserves dialogue, synchronisation across equipment and memory。

It's the biggest event of all these years。

To tame a future

Apple's AI story dates back to autumn 2011, iPhone 4S launch, Siri first stood in front of the stage。

By then, Jobos was already very sick, and apples were standing at the crossroads of an era. Siri, like a little thing that comes out of a science fiction movie, you ask the weather, you ask the restaurant, you call it an alarm clock, and it answers you with a slight mechanical tone, and for the first time you think the phone is not just a glass without temperature。

Seri was originally a military-grade artificial intelligence assistant funded by the United States Defense Advanced Research and Planning Agency. Apple put it in the bag in 2010, and TechCrunch reported that the deal could be over $200 million. A year later, Siri followed the iPhone 4S and Apple said it understood the language of nature and could work for you like a personal assistant。

In that moment, apples got the best personal intelligence portal in the world. And then it took more than a decade。

Looking back today, Siri was the first to change the way people talk to machines. In 2011, iPhone was converting mobile phones from communication tools to personal computing devices, and App Store redefined software distribution and moved the mobile Internet from PC desktops to hands. Siri appears at the tip of a rising period. After entering the apple, it quickly turned from an ambitious personal assistant to a listening voice remote。

The apple bones believe in closure and control. But a real personal assistant must access more services, understand more context and tolerate more uncertainty. Uncertainty, on the other hand, means error, privacy risks, and apples are the least equipped to cope with disorder。

So Siri was only allowed to do certain tasks, like a domesticated future. It has a name, a voice, a personal packaging, but only a lack of the initiative and memory required for true personality. Users were initially shocked by it, and then joked about it, and then it simply didn't work。

Apple first put "Personal Assistant" in the phone, then shut it down。

Agent, what the whole industry is doing today, looks back, 2011 Siri is almost its prototype. The apple was the first company to make the Agent prototype, and then the last one to finish it。

UNLIKE AI, AI

Siri hasn't grown up all these years. Apple's AI is stuck

THE ANSWER IS THE OPPOSITE. APPLE DOES A LOT OF AI, JUST DOESN'T LOOK LIKE AI ANYMORE。

IF YOU PRESS THE PRESS, APPLES ARE LIKE 2024 WHEN THEY SUDDENLY START TALKING ABOUT AI. BUT IF YOU LOOK BACKWARDS ALONG THE TECHNOLOGY PATH, APPLES HAVE BEEN WORKING SINCE TEN YEARS AGO。

In 2015, it acquired two companies in succession, one for language supplementation and one for direct in-depth learning on mobile phones. The same year WWDC, speaking of Proactive Assistant, tried to get the system to give advice before the user spoke. The idea was advanced, but under the technical conditions, it was more like a slogan。

In the second year, SiriKit was introduced to open a seam between Siri and the developers, and to speak publicly about Differential Privacy, stating that it was necessary to learn from large-scale data while protecting individual privacy. In 2017, iPhone X brought Neal Engineering, Face ID and the camera to learn from equipment-end machines, while Apple launched Core ML, which allowed developers to run models on apple equipment, and bought Workflow, a subsequent fast-track command。

That's a lot of apple answers. It wants AI, and it doesn't want Google to bet on clouds and massive personal data. And developers, and they don't want Siri to become a pot of stew. So apples pick the hardest and slowest road, do the end, do the privacy, do the system integration。

Around 2020, Apple bought several companies to do low-cost edge AI and voice understanding. In the same year, M1 chip was released, 16 nuclear Engineering went up to Mac, and end-side AI calculates from the cell phone in the pocket to the computer. The following year, Live Text and Vital Look Up landed, the text in the photos can be copied directly, the camera can recognize grass and more voice requests can be processed without the machine。

Apple did not launch a single AI App for over a decade, but it did make the phone smart。

It makes sense to choose this path. AI on the phone isn't just a answering machine, it looks at photos, listens to voices, understands contacts, calls App, senses the amount, location and time. It would be better if it did something when it did not have a net, and it would be better not to upload the user ' s life to the cloud at every request. Apple's hardware controls make it eligible to go this way。

BUT THERE IS A DEEP GAP BETWEEN LOCAL AND OVERALL INTELLIGENCE. APPLES ARE GOOD AT BREAKING TECHNOLOGY INTO RELIABLE PARTS, GENERATING AI AND ASKING IT TO PUT THE PARTS BACK TOGETHER。

These parts are buried quietly in the system and await an opportunity。

The opportunity didn't come first. ChatGPT comes first。

At the end of 2022, when ChatGPT appeared, apples were not unprepared. Tim Cook has repeatedly emphasized on several occasions that AI and machine learning have been at the core of apple products for many years, and Bloomberg also revealed in 2023 that inside apples is the Ajax Large Model Framework and the inside Chatbot Project。

But the problem is not if the apple has a card, but the problem is that the rules on the table have changed。

ChatGPT has drawn user attention from "functional" to "capacity." Users start defaulting that there must be an AI on the phone and then go stronger than who. When ChatGPT had been able to organize a whole bunch of crazy ideas into an email, Siri said, "I found them online."。

2024 WWDC, Apple put Apple Industries on the stage. Writing tool, circular summary, photo search, Siri personalized understanding, ChatGPT access. Apple finally admits that it's based on self-study models, and in 2024 at least, it won't keep up with the expectations of its users. But the pie that it painted did not end up at the pace of propaganda。

Please, Google, be a tutor

The structure of the whole Siri team is not keeping up with this round of AI。

A number of media confirmed that Apple AI head John Giannandrea withdrew, Craig Federighi took over AI direction, that the Vision Pro head Mike Rockwell was transferred to lead the Siri team, and that a large number of Siri engineers were sent to study the AI programming tool. This is not a decent rotation, and the apples have realized that they will not be able to catch up with the original people and the original rhythm。

In January 2026, Apple and Google issued a joint statement that Apple will customize Apple Information for iPhone and other products using Gemini technology. Apple is reported to be paying Google approximately $1 billion a year, using a customized Gemini model of 1.2 trillion parameters to support Siri. Apple had tested the models of OpenAI and Anthropic before choosing Google。

This is completely different from the 2024 ChatGPT access. That time, ChatGPT was more like Siri's succour when he couldn't answer the user's request, the brand was OpenAI, the interface was window-based. This time Gemini went straight to the bottom and became part of the Apple Generation Foundation model。

The key action is distillation. Google gives apples full access to Gemini, and apples use large models in Google data centres to generate high-quality answers and reasoning processes, and then use these results to train smaller, cheaper models that can run on iPhone。

WWDC published a technical article the previous day on Apple, which packaged the collaboration into the third generation Apple Foundation Models, and Google customized five models. An AFM 3 Core with 3 billion parameters on the end, and an AFM 3 Core Advanced model with 20 billion parameters, but only part of which is activated as requested. The cloud contains AFM 3 Cloud and image models ADM 3 Cloud, and the strongest AFM 3 Cloud Pro。

More realistic changes are in the calculation. The end-side model is too smart to perform all the tasks, and it is difficult for Apple’s infrastructure to carry the complete Gemini level reasoning on its own, and some requests will run on Google Cloud’s Nvidia GPU. Apple then confirmed that the PCC had been extended beyond the Apple's own data centre for the first time, and that the technology vault covered the Nvidia Defence Company, Intel TDX and Google Titan chips. Apples emphasize that PCC software remains under their own control, that equipment only trusts the apple encryption approval process, and that the relevant binary documentation is open to security researchers。

Apples do not really give up control, but give up all self-researching dignity。

Bones are borrowed

UNDERSTANDING THE LOCATION OF APPLES IN THE AI ERA, YOU NEED TO SEE WHAT ITS CORE ASSETS ARE。

NOT CHIPS, NOT MODELS, EQUIPMENT. THE EQUIPMENT CONTAINS ALBUMS, MAIL, CALENDARS, MAPS AND PAYMENTS AND CONTAINS A LARGE NUMBER OF FRAGMENTS OF ORDINARY PEOPLE ' S LIVES. WHICH AI CAN MOBILIZE THESE FRAGMENTS, IT'S NOT JUST A CHAT ROBOT, IT CAN BECOME A REAL PERSONAL INTELLIGENCE HUB。

Apples began to pave the way for this hub long ago. Workflow, bought in 2017, became a fast-track command and Siri and system automation depth binding. The App Intents, launched in 2022, allows third parties to expose their capabilities to the system portal. In the Apple Inteligence era, these interfaces became the hands and feet of AI to call for real world action。

With these interfaces, OpenAI can come in, Gemini can come in, and the Chinese market can look for local partners. But the way they come in is not to take over the iPhone directly, but to be put into the Apple's terms of reference and privacy rules。

Apple is not the best model for anyone. It fears that users will start bypassing the system and hand over their lives directly to another entrance. If one day the user opens not App, but an AI assistant who can do everything for him, apples become a good shell。

So from then on, Apple Industries, in these words, is more representative of product control than of complete technological sovereignty. The skin is long, the clothes are tailored, but the bones are borrowed. Google provided skeletons, Nvidia provided joints, and apples were going to put on their own clothes and walk out。

Google got a huge endorsement from this deal, even apples admit Gemini's bottom capacity is more reliable. What Nvidia has received is another proof that even apples have the strongest consumer chip and self-study server ambition to get ahead of frontier reasoning and complex angent tasks, or not to bypass GPU clouds。

But the more bones are borrowed, the less the body is entirely its own. There is a business plan, regulation and technical rhythm for suppliers behind each of the borrowed bones. If one day someone had to pump the bone back and the apple could stand, it would not have to answer the question yet, but sooner or later。

New tenants in the system

Ordinary people do not care about model parameters. The ordinary people are concerned about whether the cell phone can be less annoying。

There are times when you expect more from Siri

It's almost an apology for apples。

And then it tried to make you see a different morning。

You wake up, there are 20 notices on the screen. You've got to cut it off in the past, and now the system has prioritized you, the boss's hair is ahead, and advertising and promotions are rounded up in a gray line. You opened the e-mail, a long working e-mail had been taken into three sentences, and you decided to reply, and Siri gave you a draft based on your tone of speaking to this man. You remember calling a businessman this afternoon to return the goods, but you haven't set it up yet, and the system has turned the order number over from your e-mails two days ago and posted it on the phone interface。

That's the story that apples are going to tell, the intelligence that lies under the system, saving you the cognitive work that you repeat every day. Read less nonsense, search less files and be interrupted once。

In order to tell the story, apples almost redo the entrance to Siri. The iPhone is put on the Vibrator Island and can talk. iPad and Mac combined on Spotlight. It has a stand-alone App, which preserves and continues the dialogue of the past across the device. Apple wanted Siri to become an AI assistant living in the system with memory and context, but tried not to make it look like ChatGPT。

Visuality is also an important direction. Siri Mode was added to the camera to give nutritional information by taking pictures of food, and to see things that you can't read and search. The system-level hearsay is not just a voice transliteration, but automatically adds a punting format that translates words into text that can be sent out directly。

The developers are paving the road. Apple opened the Core AI framework to allow third parties to load their own models on the device. App Intents upgrades to Siri make it easier to understand third-party applications. The Foundation Models Framewok no longer calls its own side model, but also supports access to Claude and Gemini as external suppliers. Apple is paving the way for the whole ecology, and after that, Siri, if he's going to cross the App, the developer has to give the content and the action to the system。

If these plans land, Apple AI will no longer be just a "talking Siri"。

It's just that this apple is a lot more cautious than it used to be. Siri AI was only open to users later this year in Beta, English first. And the same Apple Intelligence to China is probably not the same product anymore。

CHINESE USERS LOOK AT APPLE AI, BASICALLY TUICHI. THE PRESS CONFERENCE WAS LIVELY AND GOOD-LOOKING, BUT THE CHINESE REGION "RESPONSIBLE."。

The Chinese market has a set of rules for generating AI, for filing, content security and localization of data. Apples require local model partners and regulatory approval. Apple Intelligence is not just a problem in China a few months later, it may not be the same thing from the bottom。

U.S. users see a combination of self-study models plus Gemini, and Chinese users may see a common version of apple system privileges, local cloud services, indigenous models and regulatory requirements. They're all called Apple Intelligence, but the actual capabilities and the accessible borders may be completely different。

iCloud mainland Chinese services are operated by Yungase Guizhou. The disk saves the file, AI understands the document; AI looks at the cloud inventory photos, AI looks at the picture; AI syncs the disk memo, AI pulls out your plans, habits and relationships from the memo. These data have been used completely newly in the AI era and naturally subject to different weights of regulation。

More realistic threats come from competition. National mobile phone manufacturers are fast moving on the big end model, Chinese helper and Image AI. For Chinese users, $120,000 to buy a new iPhone, the core AI function is not working, so it's better to change the brand。

THE DAY-TO-DAY SCENE IN CHINA’S MARKET IS PARTICULARLY DIFFICULT FOR APPLES. WE BELIEVE, PAY FOR TREASURES, BEAUTY GROUPS, TREMORS, INTERNET TRAFFIC, GOVERNMENT SERVICES, HOSPITAL REGISTRATIONS, AND SO ON. AN AI ASSISTANT CAN HARDLY BE CALLED "SMART" IF HE CAN'T GET INTO THESE SCENES, AND CAN'T READ THE CONVERSATION, THE PAPERS, THE AUTHENTICATION CODES AND ALL KINDS OF EXPRESSIONS THAT ONLY LOCALS CAN UNDERSTAND。

Understand someone

Apple Intelligence has another problem, and it doesn't cover all iPhone。

iOS 27 can cover iPhone 11 and second generation iPhone SE, but Apple Intelligence requires at least iPhone 15 Pro and update models, M series iPad and Mac. The strongest end-side model also requires higher iPhone 17 Pro, iPhone Air, at least 12GB unified memory M4 iPad or M3 Mac。

OVER THE PAST FEW YEARS, THE CYCLE OF CHANGE HAS GROWN LONGER. THE SCREEN IS GOOD, THE PICTURES ARE GOOD, AND MANY PEOPLE DON'T CHANGE PHONES EVERY YEAR. AI MAY BE THE REASON WHY APPLES ARE RE-STIMULATING, AND THE END-SIDE AI DOES NEED STRONGER CHIPS AND LARGER MEMORY, AND THE HARDWARE THRESHOLD IS INEVITABLE. A PERSON'S ABILITY TO BE PACKAGED AS "BETTER UNDERSTAND YOU" TURNS INTO A PRICE THRESHOLD。

Apple has been asking "iPhone after" for more than a decade, trying watches, trying headphones, trying TV, trying the car project that was cut off after 10 years. In 2024, part of the car crew was transferred to the Generating AI team。

AI is just in time to give Apple a next-generation story that does not need to build new hardware from zero, but to modify the equipment that is already in the hands of billions of users. IPhone may be after iPhone, just it has to become something else。

Tim Cook's successor, Ternus, is responsible for planning the future of hardware products, suggesting the next step for apples. He is promoting an unpublished set of AI devices, glasses with cameras and wearable devices to understand the surroundings with computer vision. If these products come true, Apple Intelligence will continue to spread from mobile phones, which could become new senses, headphones, glasses and family hubs。

But no matter how far the senses extend, the core issue remains the same。

Relationships between people and mobile phones are, for the most part, not sitting down for a long time, but rather disturbing each other in a very trivial setting. You're on the subway, the kids are crying, the boss is rushing, there's 20 notices on the screen. Apple Intelligence's most specific meaning to ordinary people is not to be a helper, but to get a cell phone to start sharing some of your cognitive chores. Read less nonsense, search less files and be interrupted once。

APPLE HAS BEEN SHAPING ITSELF INTO A COMPANY ON THE SIDE OF THE USER. IT SAID THAT PRIVACY WAS A FUNDAMENTAL HUMAN RIGHT, THAT EQUIPMENT BELONGED TO USERS AND THAT TECHNOLOGY SHOULD SERVE PEOPLE. IN THE AI ERA, THIS IS A REAL TEST. BECAUSE ONCE A SYSTEM BEGINS TO UNDERSTAND YOU, IT'S NOT JUST PROTECTING YOUR DATA, IT'S SHAPING YOUR ACTIONS. IT GIVES YOU A SUMMARY, GIVES YOU ADVICE, FILTERS YOUR INFORMATION AND JUDGES WHAT IS IMPORTANT AND WHAT CAN BE IGNORED。

THE DIFFICULTY OF A PERSON'S INTELLIGENCE IS NEVER JUST THAT OF A PERSON. A PERSON ' S LIFE IS NOT A DATABASE, WHICH CONTAINS EMOTIONS, MISUNDERSTANDINGS, DISRESPECT AND CORNERS THAT DO NOT WANT TO BE SEEN BY ANY SYSTEM. AI CANNOT JUST USE EFFICIENCY PASSES TO ACCESS THESE PLACES。

Ishigami wrote in Clara and the Sun about an artificial intelligent companion, Clara. She spent her entire existence understanding a girl, learning to observe changes in light, learning to read expression and silence, and learning to be quiet when it was quiet。

But the most exciting part of the whole book is that Clara finally realized that the girl had something she could never touch. She is not ingenious, but she understands one thing and understands that one person and one person's data are completely different。

It took 15 years for Apple to admit Siri was not good enough. WWDC this night borrowed models from Google, calculations from Nvidia, another year of patience from users. It proved that it was willing to bow, but that it was only the beginning。

The next thing it's gonna learn is what Clara knew. Not how to get smarter, but to know where to stop after walking into a person's life。