Apple has given up on one of its most ambitious projects, building a car, and is considering moving 2,000 employees to its artificial intelligence department. Relevant people close to Apple revealed to Huxiu that people from the Vision Pro and automotive project teams have previously transferred to artificial intelligence projects.

As Steve Jobs said: "Deciding what not to do is just as important as deciding what to do."

Even though ten years of efforts to build cars were in vain, facing Think Different's OpenAI, Apple finally woke up from its dream and decided to fully join the AI battlefield.

The world says that Apple's full shift to AI is long overdue, but Apple just "gets up early and goes to work late." From Steve Jobs's early attempts at Atari's "Brick Breaker" game to Siri, which taught all voice assistants a lesson in 2016, Apple has written several significant strokes in the history of AI.

Now, Apple finally can no longer tolerate the "double-act" performance of OpenAI and Google, and has chosen to focus on investing in AI again. As the only technology company that may have full-stack capabilities such as computing power layer, middle layer, model layer, and application layer, will Apple's entry break new ground in the AI industry? In this AI war sweeping the world, what unique advantages does latecomer Apple have?

Looking at the global AI, Microsoft brings OpenAI to conquer the princes, but it is surrounded by wolves; Google is chasing OpenAI, but it is following suit; domestic manufacturers are trying to cross the river with OpenAI, but they have the advantage of nature. The pattern of this three-part world is yet to be determined. The vast world has a lot to offer. If given time, hegemony can be achieved and Apple can prosper.

01 The ancient AI evangelist

According to IIF data, total global debt will set a new record in 2023, reaching $313 trillion. Among the new debts, how early did Apple start in the field of AI? Tracing back to its origins, the starting time is far ahead of any current AI giant.

In 1975, Gang Leader Qiao’s great cause was not yet completed and Apple was not yet established. He was still upgrading and fighting monsters at Atari Game Company. Atari was at its peak in the gaming industry at that time, comparable to Nintendo in the mid-to-late 1980s and PlayStation in the 1990s. The 20-year-old Jobs and Steve Wozniak (Apple co-founder) spent four nights completing the hardware design of the game "Arkanoid".

"Arkanoid" was welcomed by players as soon as it was launched. It is famous in the history of game design and has also become the game system used by the DeepMind team to train deep learning.

Starting from the chess-playing program written by Turing and Shannon, the founders of artificial intelligence, early artificial intelligence was dedicated to creating a program that could defeat humans in the game.

So DeepMind combined reinforcement learning with deep neural networks to create a system that can learn Atari games. After thousands of trainings, the average score of the deep learning network in the game "Arkanoid" reached 10 times that of humans. DeepMind became famous and was acquired by Google a year later, and then AlphaGO, which started the last wave of AI, was born.

In the field of natural language processing and speech recognition, a major breakthrough in artificial intelligence, Apple was once ahead of any mobile phone company.

In 2010, Apple acquired Siri. In 2016, Apple launched the intelligent voice assistant Siri on the iPhone, which has significantly improved performance compared to other mobile phone voice assistants. "Sometimes, a performance improvement is so significant that you will test again to ensure that no decimal point is missed. The emergence of Siri is one of such cases." Apple engineers commented.

02 Missed the new wave of AI?

According to IIF data, total global debt will set a new record in 2023, reaching $313 trillion. Among the new debts, although Apple indirectly gave birth to the last AI wave, under the new wave of AI set off by OpenAI, Apple, which was caught off guard, although it made frequent moves, has always been cautious and low-key.

In the new era of AI, Apple only does three things, but none of them are perfect.

• First, acquire multiple AI startups. According to the latest report released by market research agency Stocklytics, Apple acquired a total of 32 AI companies in 2023 and applied the acquired AI technology to product improvements. For example, in 2020, Apple acquired Voysis, a Dublin-based voice AI technology company, to improve the Siri human-machine conversation experience.

• Second, develop large models and AI conversational robots. In July 2023, it was revealed that Apple was developing a large model Ajax and the internal chat robot Apple GPT. The Ajax system is built on top of Google's machine learning framework, Google Jax, and serves as the basis for the internal ChatGPT-style tool Apple GPT, but it is more of a driving force for internal products and is not open to consumers. But some Apple employees said it essentially copied Bard, ChatGPT and Bing AI and did not include any novel features or technology.

• Third, Siri is considered most likely to be connected to the large model. In 2018, Giannandrea, the former head of Google AI, joined Apple to lead the company's artificial intelligence and machine learning team, and took over the leadership of Apple's intelligent assistant Siri. Siri has been criticized for stagnating before, with Siri co-founder Dag Kittlaus saying that Siri may not have reached its full potential after being acquired by Apple. After ChatGPT subverted the experience of personal intelligent assistants, Siri was under greater threat.

It is not difficult to see that although Apple has spent a lot of time and energy on artificial intelligence, it has always been hesitant. In this regard, Cook has tactfully stated that Apple will add artificial intelligence to more of its products, but will do so "after careful consideration."

Previously, when asked about Apple's work on generative AI, Cook's answer was "stay tuned." "Apple really hasn't made a big splash in AI yet," said Brian Mulberry, an investment manager at Apple shareholders. In the era of the new wave of AI, we have not felt that Apple is as enthusiastic about AI as Microsoft, Google, and OpenAI.

03 Where Apple AI comes in

The day after it was revealed that he had given up building cars and switched to AI, Cook changed his previous attitude and stated at the shareholder meeting on February 29 that the company would "open up new horizons" in the field of generative artificial intelligence in 2024. He emphasized: "We believe This will bring transformative opportunities to users."

So, what are the advantages of latecomer Apple? Will it, as Cook said, break new ground in the AI world?

Qiu Chen, overseas partner of Huaying Capital, believes that Apple's advantage is that it may be the only technology company with full-stack capabilities such as computing power layer, middle layer, model layer, and application layer. In terms of its own artificial intelligence chips, cloud computing The superposition of multi-layered cost-reduction and efficiency-increasing optimization of algorithms and data will bring significant model system optimization effects.

Therefore, Apple, which has invested heavily in AI, can rely on its unique advantages to carve out an Apple-specific AI path and dominate the fierce AI competition.

Next, let's take a look at Apple's advantages at each of the above levels.

• Computing power layer: Currently, the AI training side is highly dependent on NVIDIA. Apple has its own data centers in China, Europe, and the United States, so it has accumulated a large number of NVIDIA GPUs, at least it will not face the problem of computing power shortage; on the inference side, Apple is in AI Continuous investment in chip design R&D and full-stack tool chains makes Apple fully capable of launching inference acceleration solutions based on self-developed chips, and can adapt to large self-developed base models. Google TPU and the startup Groq have already achieved results comparable to Nvidia in this field.

• Data layer: The iOS system with a huge user base can provide Apple with a large amount of corpus and user behavior. Since it was integrated into Siri of iPhone 4S in 2011, it has accumulated user conversation data spanning more than ten years.

• Application layer: The existing iOS system provides a platform and audience for content generation services such as text, pictures, audio and video for C-end users.

Qiu Chen believes that if Apple starts from the above-mentioned large base model and self-developed chips, combines software and hardware, and does the two things OpenAI and NVIDIA focus on at the same time, the cost will still be high, so it will temporarily give up building cars and concentrate resources. After the adjusted priorities of Apple's business, the order is likely to become: MR → large model → driverless car → embodied intelligence.

With the above advantages, in what areas will Apple, which has fully switched to AI, bring transformative opportunities to users?

Let’s start with the hot AI mobile phones nowadays. Although domestic rivals have already entered into large-scale end-side models a year ago, Apple’s continuous investment in chips and self-research capabilities allow it to use equipment more efficiently than other mobile phone manufacturers. Computing resources such as CPU, GPU, and NPU. In addition, running large models on the device has extremely high requirements on memory speed and memory resources. Apple can use its strong R&D and supply chain integration capabilities to solve hardware technical problems to the greatest extent possible. However, according to Apple's past promotional tone, it is likely that it will not mention AI phones and large terminal models, but will focus on applications. What solutions will Apple find that are different from those of its competitors? The next release of iOS18 and iPhone16 will reveal the answer.

For Vision Pro, the next-generation mobile terminal platform that Apple has high hopes for, AIGC's cost reduction and efficiency improvement for its 3D scene construction will be greatly reduced. Real-world and user data acquired through sensors and built-in cameras can also be combined with AIGC for big data analysis and modeling to obtain better real-time interaction and user experience feedback.

People close to Apple revealed to Huxiu that Apple’s abandonment of car building is likely to be a temporary delay in its car-building plans, waiting for breakthroughs in self-driving technology before launching more mature self-driving car products. Sora's learning and expansion of the physical laws of the real world and its ability to judge the trajectory of objects will bring new ideas to autonomous driving in difficult situations. It’s not that Apple has completely given up on building cars, but there is no need to build cars without autonomous driving. To achieve autonomous driving, AI is the only way to go. This is why Apple gave up building cars and turned to AI.

View more at EASELINK