Apple’s big model of Siri may not be what you think

Apple’s AI is late but arriving.

According to the Wall Street Journal, Apple is in discussions with Baidu about integrating generative AI into iPhones and other devices in the domestic market.

Although there is no official confirmation yet, two things are certain:

  • iPhone 16, iOS 18 and MacOS will be equipped with AI functions
  • Large models on Apple devices will be provided by different manufacturers at home and abroad

Compared with domestic brands that have already equipped AI assistants, Apple is unsurprisingly more than half a year late this time. Slowness seems to have always been Apple's label, but they can always bring some surprises while moving forward steadily.

However, the speed of progress in large AI models is measured in weeks or even days. Is Apple's late arrival once again a latecomer, or the beginning of a new era of falling behind?

A slightly compromised plan, the key is to get on the bus first

On the last day of last month, Apple announced in a 12-minute short meeting that it would give up building cars and move All in AI. Many members of the automotive team would be transferred to the AI ​​department.

The Titan Project, which has been dormant for ten years, failed in the last year of entering new energy vehicles. It will be a bit regretful for the future automobile market where a hundred schools of thought will contend. However, from the perspective of the long-term development of a technology company, this is nothing more than a long-term development. And the right choice.

AI is a basic application. At a time when all major companies are actively or passively embracing AI, Apple's "breakaway" is in line with the times, but how to embrace AI? What kind of artificial intelligence can occupy a place in the gradually divided market? This is the first problem they have to solve.

For foreign markets, Apple is actively negotiating with Google to add large AI models to iOS 18 to realize AI functions that other brands have already had.

Although "the two parties have not yet decided on the terms or brand of the artificial intelligence agreement, nor have they finalized how to implement it," among the many alternative partners (OpenAI and Anthropic), Google and Gemini should be the most suitable for Apple and iPhone. .

The Samsung Galaxy S24 series models released in February this year are out of the ordinary with AI functions. Functions such as call translation and creative writing have caught up with the domestic average. Instant search has shortened the search path and is very likely to become the main feature of AI mobile phones in the future. development route.

The overseas version of the S24 series is able to complete the above functions through the support of the large model Gemini.

Speaking from experience, Google has completed its initial attempts on the flagship model with the highest shipment volume in the world. Compared with manufacturers that are popular on the PC or Web side, they know the operating habits, usage scenarios, and adaptation of large mobile phone models better. What should the application do.

Furthermore, Google itself is more eager to get Apple's projects.

According to statistics from the international data company IDC, Samsung's global smartphone market share reached 19.4% in 2023, while Apple successfully reached the top of 20.1%.

If it wins Apple, Gemini's adoption rate in mobile phone terminals worldwide will reach 40%, which is extremely good for a large AI model company facing fierce competition.

In addition to Google, Apple also woke up from the dream.

Unlike other manufacturers that emphasize "self-research", Apple used cooperation to achieve AI integration from the beginning, and it also had its own considerations.

First of all, under the current situation of late start and slow progress, "use doctrine" is a good way to quickly compete for the market. Cooperation with Google can reduce R&D costs and charge high pit fees, while also easing the two companies' conflicts. current regulatory pressures.

Secondly, AIGC's technology is very good, but when it was implemented, it was criticized a lot for its shortcomings in ethics, privacy and other aspects. It was handed over to a mature third party, especially Google, which had successfully tested the water on Samsung models, saving effort and worry. And reduce public opinion and liability risks.

Another hurdle in this is technology localization. Each country and region has different requirements for the supervision and related laws of AI large models. The implementation of legal compliance is the prerequisite for competing for the market and developing technology. Therefore, the "domestic + international" two-pronged route has been born.

According to the initial fruitful cooperation between Samsung and Baidu, Apple will choose this route that has been "verified as true".

The AI ​​function on the National Bank version of the Samsung S24 series actually consists of technologies from multiple manufacturers: the instant search function is provided by Baidu and JD.com; the intelligent photo retouching is completed by Meitu Xiuxiu’s large model MiracleVision; the article summary, intelligent The writing adopts Baidu’s Wenxin Yiyan model.

Whether Apple will also cooperate with multiple manufacturers still needs to wait for follow-up news, but the cooperation with Baidu is already a certainty.

In the end, what Apple wants to do is not a smart voice assistant, but a complete set of AI terminals. However, according to Macrumor’s revelations, with the current self-research progress and technical achievements, Apple’s large models are still far from the level of companies such as Google and OpenAI.

Instead of rushing to launch an intelligent chatbot, it is better to use mature solutions as a transition first to gain more research time and room for improvement for self-developed large models.

The current market is important, but the core technology of the future is fundamental

Cooperation is the first step in Apple's AI globalization, and the ultimate goal is to have a large self-developed AI model.

This is a project that consumes money and energy. Not to mention regression, if you make a little slower progress, you may be eliminated next week. A large and competitive model often represents future dominance and bargaining power in the market.

Cook believes:

Breaking new ground in generative AI, we believe this technology can redefine the future.

Apple’s exploration of large models has actually always been on the agenda.

On the 15th of this month, Apple engineers quietly released a research paper, which detailed the development process of a new generative AI model called MM1.

MM1 is a multi-modal LLM series with up to 30B (30 billion) parameters, which is Apple’s latest research result in multi-modal large models.

In general, Apple's self-developed model still lags behind Gemini and GPT4V in terms of test results. It does not show as amazing results as Sora in generating results, nor does it explore a new technical route.

However, it can control various data variables and find out the most critical factors that affect the model generation effect in comparison. To put it simply, it is not inherently powerful, but it is good at observation, practice and summary. After repeated attempts, , can also achieve good results.

MM1 is composed of dense models and MoE (Mixed Expert) variants. When the instruction enters the MoE, whether you should go to the east market to buy a horse or the west market to buy a saddle will be clearly arranged by this command center.

While problems are refined and classified, computing efficiency is also improved and operating energy consumption is saved.

The release of this paper represents the staged results of Apple's exploration in the field of AI. Although MM1 did not subvert the industry or amaze the world, their progress can still be seen in obscure professional terms:

Our working model has always been to do the work first and talk about the work later, rather than being rude in front of ourselves. ——Tim Cook

Apple, which did not disclose too many technical details, is actually still planning another move: a large terminal-side model.

As early as the end of last year, Apple proposed a method for implementing large models into "memory-limited" devices such as the iPhone in a paper titled "Large-scale Language Models in Flash Memory: Efficient Large-Scale Language Model Inference under Limited Memory."

Researchers say they have successfully deployed LLM (Large Language Model) on iPhones and other memory-constrained devices using the latest flash memory technology.

This project is called Apple GPT. Its biggest function is to store LLM data directly in flash memory, such as integrating it inside Siri. Compared with the traditional running method, the new technology increases the inference speed of the CPU and GPU by up to 5 times. and 25 times.

"The efficiency methods we developed enable AI models to run within twice the current memory range of the iPhone," the researchers said.

In other words, it is feasible to carry large models on the side end. By reducing the amount of data transmitted by flash memory and improving the throughput of each transmission, LLM data can be stored directly in flash memory.

Technology aside, Siri is the bridge between us and AI

The progress is slow, the news is little, and the layout is large. This is an overview of Apple's exploration of AI.

Every time we see that a certain Apple technology lags behind the market and competitors, it gives people the illusion that it "started too late." In fact, when looking through relevant news and patent documents, you will find that it is often the first to be deployed. That batch, even that one.

As of 2023, Apple has acquired a total of 32 AI companies, ranking first among technology giants in acquisitions. The acquisition of Siri should be regarded as the beginning of Apple's entry into AI.

In 2010, Jobs made a phone call to Dag Kittlaus, the "father of Siri", which led to Siri joining Apple and starting the iPhone with a worth of more than 200 million US dollars.

Siri was originally positioned as an assistant to obtain information quickly and accurately, or to handle complex tasks.

In its most primitive version, Siri can connect to 42 network services – from restaurant review website Yelp and ticket sales website StubHub to movie review website Rotten Tomatoes and mathematical calculation website Wolfram Alpha.

Based on the prompts, Siri will integrate various information and reply to the user. Siri can help users buy tickets, reserve a restaurant or hail a cab without opening another app.

These "AI functions" that are now being vigorously promoted by AI Pin and other smart assistants seem to be just the "basic operations" of Siri more than ten years ago.

However, the actual experience of Siri has been greatly separated by the explosive development of large AI models.

Intelligent assistants are passive imitations of people, answering all questions and responding to requests.

The AI ​​terminal is an active approach to people. Based on the user's personal habits and preferences, after summarizing the past and reasoning, we will give you the most appropriate suggestions and answers at different times and places, and we can continuously learn and optimize to become "private and exclusive".

▲ Picture from: x.com

On the whole, Apple's lateness is only relatively late, because AI mobile phones are still in the early stages of development.

Indeed, most domestic brands have already made efforts in the stage of AI terminals, with roughly the same functions and different specialties. However, the usability of each large model can only be regarded as passing, except for AI elimination of OPPO photo albums and real-time call processing of Samsung. For segmented functions such as translation and Xiao Ai’s AI calling, most of the experience is still somewhat different from that of independent AI applications.

In addition to the technological breakthroughs of manufacturers, this is also related to the open interface of the App. For example, models that do not support WeChat voice call summary will lose a large area of ​​application space in daily life.

Therefore, the integration of large models, systems, and apps, as well as the exploration of new interaction methods, still have a long way to go. Prior to this, AI functions had not yet reached the level of influencing consumer purchasing decisions.

In the first year of AI’s launch, Siri’s goal is to close the gap of more than half a year with other AI assistants; and as an important part of Apple’s future layout, we are even more looking forward to what kind of “One more thing” Siri will bring in June. ”.

# Welcome to follow the official WeChat public account of aifaner: aifaner (WeChat ID: ifanr). More exciting content will be provided to you as soon as possible.

Ai Faner | Original link · View comments · Sina Weibo