In early 2023, large AI models became popular around the world. At that time, some people said that this The best way for technology to reach the masses is through mobile phones.

No matter what consumers think of this sentence, mobile phone manufacturers, which have suffered from declining sales year after year, must have listened to it.

As a result, starting from the second half of 2023, all new mobile phone products will flaunt AI capabilities and claim to be AI phones. This feast continues to this day. Even the relatively conservative Apple has released "Apple Intelligence", and those second-tier mobile phone brands that were originally thought to have been squeezed into a lack of living space are also stubbornly relying on "All in AI" to stay. On the card table.

For a time, AI in the mobile phone market was buzzing with gongs and drums, red flags were waving, and there were crowds of people... No, the crowds of people might not necessarily be the case.

For users, facing the overwhelming number of AI mobile phones , the gorgeous press conference PPT may make you a little confused: What can AI bring to me? What are the differences between so many AI phones? The AI ​​I envision is the kind from science fiction movies. Can you give me a mobile phone?

In fact, since deep learning became popular in 2017, the combination of mobile phones and AI has continued. In the past few years, we have experienced changes from on-device AI computing power, AI photography, on-device large models, to system-level AI capabilities. If a mobile phone is an orchard, the fruit tree AI has been growing in it for many years.

So, by 2024, when AI mobile phones will have unlimited glory, will this fruit remain ripe?

Let us go through the complicated concepts and see what AI mobile phones bring.

When it comes to AI phones, we always have a very empty feeling I don’t even know what manufacturers and experts are excited about.

This sense of falsehood comes from the fact that there is no standardized identification of AI mobile phones in the industry. So each manufacturer, or even everyone, can interpret it independently. There are consensus and differences among these interpretations, and there are also a lot of paradoxes and word games. These sounds are intertwined and become a conceptual game.

Therefore, if you want to understand AI mobile phones, you must first avoid being disturbed by conceptual games. Just like picking fruit, you must first push aside the branches and leaves. How to achieve this kind of non-disturbance? A simple method is called "AI mobile phone, I have three things I don't look at".

First of all, you don’t need to look at the manufacturer’s “AI big hat”.

Not long ago, the AI ​​mobile phone field belatedly released the previousThe promise of Apple Intelligence. But this "Apple Smart" did not stir up much market splash. The reason is that Apple Intelligence is more like an upgraded package of many AI functions, rather than some new ability or new idea.

This kind of gives big names to many small AI functions The gameplay is relatively common in today’s mobile phone industry. Users thought they were getting something new, but in fact it was just old wine in new bottles and sold in bulk. Similar terms often include so-and-so smart, so-and-so GPT, and so-and-so native. They all mean the same thing. In many cases, this is not a project led by the R&D system at all, but the marketing department gathers together what AI capabilities each department has, and finally makes up a name and writes it into the PPT.

Secondly, there is a lack of standard technical terms that do not need to be read.

After any technology receives attention, it will go through a word-making stage. Some of these newly coined words will eventually stay and become conventional and relatively standardized expressions. But most of them will quietly pass away in historical memory. In the field of AI mobile phones, we often see a large number of similar new words. They all refer to something more or less, but as an ordinary consumer, there is no need to explore the context of each word, nor do they need to be confused by them.

There is no unified standard for high-level AI mobile phone vocabulary such as intelligent framework, intent recognition, active service, and global intelligence today. Enthusiasts can study it, but most consumers may as well stay away from it.

Again, there is no need to look at the large model parameters and other AI parameters. .

Since the emergence of large pre-trained models in 2018, there has been a phenomenon where leading manufacturers say the effect, and other manufacturers compete for parameters. Manufacturers that find it difficult to differentiate in terms of core capabilities tend to advertise that their model parameters are larger and their running scores are better. When it comes to the stage of AI mobile phones, it becomes a propaganda end-side large model with large parameters and a large number. But AI models are not hardware such as CPUs, and high parameters do not mean good performance. The model effect also depends on a series of conditions such as compression scheme and tuning effect, so it is not necessary to equate the model parameters promoted by the manufacturer with the product value.

What remains after these are distilled is the fruit of AI in mobile phones.

Many people are saying, what can AI do without applications? manage. But the problem is that ChatGPT is an application in itself. The prospect of creating stronger super applications on the application side is bright, but the application value of the large model itself has room to be combined with mobile phones.

The first step for AI mobile phones is to bring ChatGPT-like large model capabilities to the first-party functions of mobile phones. For example, in iOS 18.1, the key AI functions introduced include AI writing, email recommendations and replies, message summaries, etc. Thinking about it from another angle, we will find that these are all things that ChatGPT can do.

There is nothing wrong with this idea, and it must be led by the first-party mobile phone manufacturers. Because today’s smartphones provide many first-party functions, and there are many places where large model capabilities such as semantic understanding and content generation can be added. Therefore, the point-to-point combination of large language models and first-party functions of mobile phones is the most fruitful AI in mobile phones.

For similar functions, it is recommended to choose based on your own needs to select practical functions or function groups. For example, in November last year, OPPO released one-click generation of AI call summaries. This function can intelligently identify the call content and generate a summary of key information after the call ends. Another example is that Find X8 Pro can long press the power button to call up the screen question function, and AI will answer article content, picture information, etc. based on the screen display. Xiaomi has also launched a series of smart functions in ThePaper OS 2, such as AI writing, AI sound recognition, and AI wonderful painting.

But for mobile phone manufacturers, the functionality of large models Joining is obviously just a stepping stone. This single point upgrade cannot satisfy the manufacturer's ambitions for AI.

So, the intelligent agent came.

AI Agent is also called an intelligent agent and is widely Recognition is a new application form that truly embodies AI thinking. Agents can span different software workflows to achieve goal-led autonomous judgment. Putting this kind of operating logic in a mobile phone has two benefits.

For users, agents can be regarded as a new form of interaction independent of swipe interaction. Users can command the agent to complete instructions, and its understanding ability, generation ability and application range are far greater than previous voice assistants, closer to the user's basic imagination of a personal assistant.

For mobile phone manufacturers, smart phones are not only a new selling point, but also can bypass the barriers of APP, making first-party application permissions greater and the experience more important. By building an intelligent agent, there is a chance that user habits and third-party application services will be more closely tied to manufacturers.

Under such driving force, mobile phone manufacturers have been intensifying their efforts on the smartphone front for several months, vowing to have a posture of launching a decisive battle of smartphones in 2025. And from ALooking at the development of mobile phones online, mobile phones + AI Agent are an area that is divorced from the basic application framework of ChatGPT, and European and American terminal manufacturers represented by Apple have not followed up. The future achievements of China’s AI mobile phones may lie in the opportunities of smart bodies.

In terms of intelligent agents, Honor is currently the most powerful . At the Honor Magic7 launch conference not long ago, Honor CEO Zhao Ming used YOYO smart phone to automatically place an order for 2,000 cups of coffee, which suddenly brought the concept of smart phones out of the circle. Currently, Honor's main smartphone experiences include "turn off automatic renewal with one sentence", "order coffee with one sentence", "send files with one sentence", "turn off application permissions with one sentence", etc. It relatively completely and concretely demonstrates the value of intelligent agents in the mobile phone industry.

At the 2024 vivo Developer Conference, vivo also showed off its own smart solution Phone GPT, which not only has the function of identifying and operating the screen interface, but can also take over the audio and make restaurant reservations over the phone. and other functions. Extending the intelligence into the audio field has become a direction worthy of attention next.

On OPPO’s side, there are multiple consecutive rounds OPPO AI that allows for natural conversations and can operate across applications. In addition to providing AI capabilities such as object recognition, navigation, and question and answer, OPPO AI can also be combined with imaging capabilities and AIGC capabilities to help users complete functions such as image recognition, image processing, and typesetting and polishing.

Here at Huawei, the HarmonyOS NEXT version has the Hongmeng smart assistant Xiaoyi equipped with the Pangu large model. Its perception and reasoning capabilities have been greatly improved compared to previous versions, and it can realize the interaction of 23 major scenarios.

These attempts can be regarded as mobile + The progress in the direction of AI Agent is only different in the specific technical implementation methods, capability focus and promotional rhetoric of each manufacturer.

For consumers, if you are an in-depth mobile phone user, it may be time to try out smart phones today. However, if you are not troubled by traditional interaction methods and other problems, you may be able to use smart phones again. Be more mature.

At least the AI ​​Agent currently seems to have huge potential.

There are many solutions for combining large AI models with mobile phones. But I personally think that the best combination of the two is actually veryTo be clear, that's accessibility.

One of the core functions of the AI ​​large model is to convert between different media such as images, audio, and text, which happens to be what the visually impaired and hearing-impaired people need most. Therefore, when some people think that it is not necessary to integrate large AI models with mobile phones, I would say that looking at the great progress in accessibility, we can know that AI mobile phones must have a future.

Through communication with many hearing-impaired and visually impaired friends, as well as interviews with relevant organizations, we found that everyone wants to praise Huawei and vivo in real names. In the early years, it was generally believed that Apple's accessibility capabilities were very good, but now Huawei and vivo are leading the development of device-side accessibility capabilities, especially accessibility upgrades in the AI ​​era.

For example, vivo uses AI technology to help hearing-impaired, A range of help is available for the visually impaired and elderly. Last year, vivo’s “Sign Language Translator” software won first place in the CCL2024 sign language digital human translation quality evaluation. This year, vivo also upgraded the “Sign Language Learning” function to promote the popularization of sign language education.

Every step of the development of Hongmeng system is accompanied by Huge investment in end-to-end accessibility. For example, HarmonyOS 4 has launched a hearing aid direct connection function. During the Huawei Developer Conference 2023, Xiaoyi See the World was launched to help visually impaired people identify specific scenes and objects. This year, HarmonyOS NEXT is the first in the industry to launch Xiaoyi's voice repair function, which uses AI technology to analyze, repair and optimize the voices of speech-impaired people in real time. It can not only produce intelligible sounds, but also preserve the user's voice as much as possible. . This function is specially designed for people with level 4 and level 3 speech impairments. It can achieve voice restoration without text input, greatly expanding the boundaries of device-side accessibility capabilities.

For many users, the AI ​​capabilities in mobile phones may be the icing on the cake, but for people with disabilities, AI mobile phones may be the key to changing their lives and even their lives.

For this reason alone, AI must also exist, grow and flourish in mobile phones.

Looking back at the AI ​​capabilities promoted by mobile phone manufacturers, you will find that In fact, every step is still very green. If we are willing to admit the complexity of the combination of the two, then the fruit of AI mobile phones may not be too ripe.

If mobile phones only provide AI capabilities, manufacturers will face direct competition with application developers. While application developers include Internet giants and professional AI manufacturers, mobile phone manufacturers may not have the same technical capabilities and user appeal.Be persuasive.

As a result, mobile phone manufacturers hope to bet on breakthrough experiences such as smart phones, but at this stage, such experiences still need to be polished and upgraded. Many times, the instructions given by the user to the agent need to be confirmed repeatedly, which is not as efficient as directly swiping. The ability of AI to take over audio is even more uncertain. Faced with an agent that often makes mistakes and has a limited range of applications, users' patience and trust may quickly dry up.

More importantly, such cutting-edge capabilities also Not enough to influence consumers’ core decisions. There is a high probability that AI mobile phones will lead to a short-term improvement in the mobile phone market, but it is not yet to the extent that it can reverse the continued downward trend.

The next level up is more retreat. Many mobile phone manufacturers have been chanting slogans such as intent recognition, proactive service, and finding people for service for many years. But the reality is that even the most advanced large models cannot predict what people think. Consumers’ expectations for AI phones are all-powerful personal assistants in science fiction movies. In order to promote sales, manufacturers will also package and exaggerate AI mobile phones in their rhetoric, but in the end, within the visible range, the mobile phone industry can only violate consumer expectations again and again.

Despite these pressures, AI phones are still on their way. From the perspective of technical experience, it is indeed not enough to bring about disruption, but from the perspective of industrial development, the combination of large AI models and mobile phones is an excellent strategic opportunity. Large overseas models were unable to enter, so Google found another way and Apple developed rapidly. For Chinese mobile phones, AI is somewhat of a destiny.

This is an unprecedented situation, even if the fruit is not yet ripe. For most consumers, it might as well be like what Xie Yongqiang often said in "Country Love" - ​​"I have to go to the orchard."