The biggest mobile phone update in the past decade is coming soon

There is a state of chaos in the science fiction novel "The Three-Body Problem". When the planets fluctuate between the gravitational pull of the three suns and move irregularly, various unpredictable changes will occur.

The outbreak of AIGC has ushered in an era of chaos in the silicon-based world. We seem to be in the first year of the singularity, trying our best to understand this new world and live on the eve of a new civilization.

During the Spring Festival holiday that just passed, we witnessed the crazy evolution of the AI ​​industry.

OpenAI launched the AI ​​video generation model Sora, which can generate one-minute videos through simple prompt words. This may be an important foundation for AGI.

Google has successively launched Gemini Ultra1.0 and Gemini Pro 1.5, achieving 1 million token contextual capabilities and further improving multi-modal capabilities.

NVIDIA released the local AI chatbot Chat with RTX, and the largest “AI arms dealer” also joined the AI ​​application market

On New Year's Eve, OPPO also held the shortest press conference in history, announcing its entry into the era of AI mobile phones and launching the ColorOS AI New Year Edition.

At yesterday's OPPO AI strategy meeting, OPPO chief product officer Liu Zuohu also further elaborated on his understanding of AI phones. He said:

AI mobile phones will be the third major stage of change in the mobile phone industry after feature phones and smartphones.

This forms an interesting intertext with OpenAI CEO Sam Altman’s previous point of view: AI will change the world like mobile phones.

Before AI changes the world, people are scrambling to find a hardware carrier for AI. AI Pin and Rabbit R1 are both interesting attempts, but you will also find that the functional scenarios and interaction efficiency of these AI native devices do not seem to exceed existing hardware.

This does not mean that AI hardware is a false proposition. 2024 will indeed become the first year of AI hardware, but the most suitable AI terminal device in the next few years will be smartphones.

Smartphones are still the best AI carriers

OPPO regards the impact of AI mobile phones as the evolution from feature phones to smartphones. This watershed is actually the iPhone. It is said that Steve Jobs re-invented the mobile phone. What exactly did he invent? The reference answers for AI mobile phones are inside.

Back at the first-generation iPhone launch in 2007, in order to explain the difference between the iPhone and other mobile phones, Jobs drew a quadrant diagram for the coordinate system using ease of use and intelligence, giving a new definition to smartphones: Smartphones = easier. Use + high intelligence.

Compared with feature phones, the ease of use of smartphones is reflected in multi-touch and large screens. The two complement each other. The interaction of two-finger sliding operations is more efficient, the sliding unlock operation is intuitive, and the elimination of physical keyboards makes the screen more comfortable. Large space and a large screen mean better information output.

As for intelligence, in Steve Jobs’s original words, “This is a revolutionary breakthrough, bringing the real Internet to mobile phones.” It is precisely with the help of 4G and smartphones that there has been a mobile phone industry that has emerged in the past ten years. Internet era.

So when we are talking about AI hardware, does this measurement dimension fail? the answer is negative.

The digital world we live in today is actually an ocean of UI. From PCs to mobile phones, the fundamental changes brought about by each generation of general computing platforms actually improve the world's digitization level and transmission speed by changing the way human-computer interaction occurs.

Therefore, if we want to talk about what upgrades AI hardware should bring compared to existing products, it cannot do without the following dimensions:

  • Interaction efficiency (input and output)
  • Intelligence
  • -Portability

Combining these dimensions, it is not difficult to find that we do not necessarily need to re-create an AI-native device. Smartphones are the most suitable carrier at the moment.

After OPPO announced its AI strategy yesterday, Liu Zuohu also believed in the introduction interview that mobile phones will still be the most important part of AI hardware for at least 10 years .

Take the photo album of the mobile phone as an example. Recently, after taking pictures with Find X7, I often use the AI ​​elimination function. I just slide my finger in a circle to eliminate passers-by or debris, basically without the need for post-production.

There is no need to switch to other apps, finger sliding replaces mouse circle selection, which improves my interaction efficiency. The accuracy of AI-based elimination is a reflection of higher intelligence.

The emergence of ChatGPT allows us to see more intuitive interactions – natural language. Compared with the touch and click of GUI WYSIWYG, it turns out that we can interact in a simpler way.

This kind of interaction logic can be called Natural user interface (NUI). As early as 2008, August de los Reye, Microsoft's chief UI design director, publicly pointed out that the future evolution direction of GUI is NUI.

Large models with high computing power and natural language interaction are the levers that AI uses to significantly improve smartphones' ease of use and intelligence on two axes . This is difficult to achieve no matter how much improvements in imaging capabilities, screen materials, and processors are made.

Squeezing large models into mobile phones has become a reality. For example, on the Find X7 series, OPPO implemented a large model with 7 billion parameters on the device side for the first time, and put AndesGPT's Tiny model into the mobile phone. The memory required for the model to run on the mobile phone is rapidly reducing.

The normal model size of the 7 billion large model is 28GB. In order to truly deploy it on the device side, OPPO uses high-precision 4 bits to compress and lightweight the model. It can be compressed from 28GB to about 3.9GB, both in terms of storage and memory usage. It's this magnitude.

Jim Fan, senior AI scientist at NVIDIA, believes that we are entering a new era. Mobile chip performance is no longer measured by GHz, but by the speed at which tokens are generated.

Big language models are the new smartphone operating systems.

Perhaps one day in the future, the form of smartphones will be replaced by a new generation of computing platforms, but most people of our generation will start to experience the changes that AI brings to life from their smartphones.

What should an AI phone look like?

In 2024, we have already seen that mainstream mobile phone manufacturers such as OPPO are embracing AI, so what should an AI mobile phone look like?

Yesterday, OPPO and IDC jointly released the first AI mobile phone white paper, which mentioned four major characteristics of AI mobile phones:

  • First, we must be able to efficiently utilize computing resources to meet the computing needs of generative AI in the AI ​​era;
  • Must be able to keenly perceive the real world and understand complex information about users and the environment;
  • You also need to have strong self-learning ability;
  • In addition, AI phones will also have more creative capabilities and provide users with continuous inspiration and knowledge support.

The experience of AI mobile phones is also a combination of software and hardware, but the logic is different from that of current smartphones.

The OS relying on large models will be able to efficiently handle complex tasks through embedded intelligence, and will be able to actively create and provide more intuitive multi-modal interactions.

The hardware architecture has new measurement standards: high-performance AI computing power base, model library management optimization, and smart bionic sensing capabilities.

Lilian Weng, OpenAI application research director, once pointed out in an article that Agent = LLM (Large Language Model) + Memory + Planning skills (Planning) + Tool Use (Tool Use).

This is a four-component model that mimics the human process of performing tasks. LLM serves as the brain of the agent, and other key components perform tasks.

  • Disassembly and assignment of planning tasks
  • Simulate human memory types to acquire, store, retain and retrieve information
  • Call external API to extend model capabilities

In fact, OPPO's definition of AI mobile phones coincides with Lilian Weng's research on Agent. OPPO AI Center Product Director Zhang Jun also further introduced the core logic of OPPO AI super intelligence to Ai Faner.

The intelligent body will become the brain and core of the smart OS, but what we want to do is make full-stack AI changes. For example, we want to AI-ify the familiar devices such as screens and microphones, as well as the components in the OS.

This agent first understands user intentions based on knowledge graphs, document data and search engines. At the same time, based on the user's long-term memory and the current real-time interaction context, a variety of tools are called to provide services according to the needs of each user. This is the biggest difference from previous smartphones.

Many people think that AI mobile phones are equivalent to large models + mobile phones. In fact, a more accurate statement should be the combination of AI Agent and smartphones.

When AI can understand the user's natural language and gradually understand the user's intentions based on local data training, the services and functions provided by the past applications can be seamlessly integrated into the system and used in all scenarios of food, clothing, housing, transportation, entertainment, and work, the user only needs to speak Just need it.

Following this logic, the form of applications will also completely change. The entire interactive interface itself is a super app, and the way to use and develop applications will be completely different. This is the future that AI mobile phones are heading towards.

As for when we will see a real AI flagship, I am afraid we have to wait, but we can already see the prototype of AI phones on some flagships such as Find X7.

For example, when making a phone call, the AI ​​intelligent call summary function can help us sort out the key points of the call content, and accurately output the content summary with topics, key points, and to-do items in the form of a document. For someone like me who often interviews and conducts phone calls, Online meetings are very practical for professionals.

The new Xiaobu assistant launched during the Spring Festival, choose "Lianmai" to have a conversation experience that is closer to a real person, and can even simulate different roles, such as recommending recently released movies and collecting movie reviews, and voice recordings will automatically be converted into text for easy collection. Forward, get closer to a senior personal assistant.

As for the AI ​​elimination function mentioned earlier, OPPO said that the average number of daily uses per person has reached 15 times since it was launched, which also means that users’ experience with AI has begun to shift from early adopters to common users. This is also the result of the interaction of several factors such as interaction efficiency, intelligence and portability.

From these functions, we can also get a glimpse of OPPO's philosophy and route in making AI mobile phones: product-driven technology .

How to use multi-modal capabilities to solve users' high-frequency and high-perception complex task scenarios, adjust the training mode of large models based on scenarios and needs, and use different computing power will also be whether AI phones can really differentiate themselves from current smartphones. The key to a better experience.

2024, the first year of the singularity of AI mobile phones

2023 can be said to be the first year of large models. ChatGPT and a number of large models allow us to see a new possibility, that is, everyone can have a super AI assistant like Jarvis.

But even if ChatGPT has a mobile app, it still cannot truly become Jarvis, because many of the requirements cannot be processed in the execution chat box, so OpenAI launched GPTs to allow more people to develop their own agents.

But as we said before, the most suitable hardware carrier for intelligent agents is currently smartphones. Smartphones will adapt to the intelligent body from the system and derive a new application ecosystem, making the experience smoother.

At present, the conditions required for AI mobile phones are gradually maturing. On the one hand, general large models and industry large models, which are expensive in computing power, are beginning to find scenarios for productization. At the same time, the problem of running end-side models on mobile devices is gradually being solved, and memory with smaller capabilities can run large models with larger parameter scales.

This is why 2024 will be called the first year of AI mobile phones.

In fact, no matter which era, user needs should be discovered, not created out of thin air. Good products should be able to discern user needs under the trends of different eras.

As former Google CEO Eric Schmidt said in his book "Redefining the Company": "The future of technology lies in predicting users' needs and satisfying them before users are aware of these needs."

Ai Faner once pointed out in an article that AIGC is pushing society into an era of "intellectual surplus". Everyone has the opportunity to possess unlimited "external intelligence" and become a super individual.

Therefore, our portable devices should be the carrier of the most powerful external brain. Letting everyone have a dedicated super assistant in their pocket and realizing inclusive AI is also the biggest motivation for OPPO to make AI mobile phones. It has an insight into the biggest demand in the AIGC era.

In the past 20 years, OPPO is one of the few consumer electronics brands that has survived the cycle. From the era of feature phones to smartphones, OPPO's core competitiveness has never been a single point of technology, design or function, but the ability to see through the needs of the times, understand the nature of fashion, and not follow the trend.

This means that you cannot follow suit and must take the initiative to do things that have a certain risk of failure.

A103 was the first to define the music mobile phone category. It understood the entertainment and socialization methods of young people in small towns in the era of feature phones when portable media players were popular.

In the era of smartphones, from the first Ulike U701, which was truly focused on selfies and beauty, to the Find Use the "Beauty Filter" to enjoy a more real and natural state.

When the AIGC wave came, followers flocked to it, and there were more than 200 domestically produced large models. Being the first to announce an AI phone does not necessarily win the jackpot. Instead, it may be considered suspicious of following the trend. However, OPPO’s AI layout is actually much earlier than we imagined.

As early as 2020, OPPO launched its first large AI model and has continued to invest. Last year, OPPO released the self-trained Andes large model (AndesGPT), which is actually the accumulation of previous technology. It ranks in the Super CLUE knowledge and encyclopedia ability rankings. Second only to GPT 4.

At yesterday's OPPO AI strategy conference, although Liu Zuohu did not directly respond to the investment in AI, he told Ai Faner, "In terms of investment, we can say that there is no upper limit."

In January this year, OPPO established an AI center to bring together the backbone of product, R&D, operations, research institutes, digital engineering, and software engineering, allowing it to iterate as quickly as it does Internet products.

The Find X7 released at the beginning of the year is the first mobile phone to use 7 billion models on the device side. It is the first echelon of Android flagships in terms of processing power, processing latency, and visual generation response.

OPPO currently uses quantification and other technologies to significantly reduce the memory required to run models on mobile phones. At the same time, OPPO uses its self-built OPPO AI Marina Bay Data Center to deploy three-level models of Titan, Turbo, and Tiny to handle different scene requirements.

Device-cloud collaboration is currently the most ideal way to deploy large AI models on mobile phones while taking into account efficiency and performance. Device-side computing can better ensure real-time performance and privacy protection, and the cloud's stronger computing power can complete more complex tasks such as AI elimination.

IDC predicts that global shipments of new-generation AI mobile phones will reach 170 million units in 2024, accounting for 15% of overall smartphone shipments. The share of AI mobile phones in the Chinese market will also grow rapidly, accounting for more than 50% by 2027.

However, making AI experiences as influential as images in users’ consumption decisions will not happen overnight.

Big models have given birth to a new digital era, and our culture, economy and even identity are undergoing reshaping. As an important carrier of this change, AI mobile phones need more mainstream manufacturers like OPPO to serve as contributors and popularizers. More users will feel the changes brought by AI until it is integrated into people's daily lives.

For the past few decades, computing devices have been pursuing a future that automatically senses and responds to people's needs, providing an intelligent, natural and seamless user experience.

From PCs and smartphones to Vision Pro, which creates spatial computing, they are all taking us closer to this end point. Now, AI mobile phones are beginning to take over this historical task.

# 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