"I have a special AI meeting every week." A mobile phone manufacturer said at a communication meeting after releasing a new phone that AI functions first need to conform to the user's intuition, which also reflects Mobile phone manufacturers are paying more and more attention to AI.
Technological development can never press the pause button. Every time it is thought that technological iterations have reached the bottleneck and cannot be rolled out, there will always be a force that breaks through the inherent framework, leads mankind out of the dead end, and brews new innovations. possible.
In the release of new phones at the end of the year, brand new phones including OPPO, Xiaomi and other brands mentioned the breakthrough of ultra-narrow bezels, which is behind the huge investment of time and money. The process requires glue filling and chip-level dust-proof requirements, so manufacturers have invested heavily. But at the user level, perception is limited.
The development of the hardware side of smartphones is approaching the physical limit, and there is not much room for improvements in screens, cameras, and processors. Just when the industry is saddened by this, it suddenly feels that the smartphone industry has entered the dusk. The burst of side AI opened another gap for it.
Although AI has been integrated into imaging, voice assistant and other scenarios by mobile phone manufacturers as early as 2017, 2024 may be regarded as the true "first year" of AI mobile phones. After all, this year, it has also stepped on the rhythm of the overall recovery of the mobile phone industry and the AI wave moving from the laboratory to the implementation.
In this context, AI mobile phones have also transformed from the initial "plus points" of a single category to a "must have" for all models. While this has promoted the industry as a whole to leap forward, the new business ecology affected by it is waiting to be activated.
AI mobile phones need self-certificationAI can be said to be 2024 In 2016, it was one of the driving forces behind the overall recovery of the mobile phone industry.
According to Canalys, the penetration rate of AI mobile phones will reach 17% in 2024. It is expected that the penetration rate of AI mobile phones will further accelerate in 2025, pushing the global penetration rate to 32% - it only took one year for AI mobile phones With more, it has completed the path that "Big Brother" folding screen mobile phones have not been able to achieve in six years.
Despite the rapid growth in data, the statistical caliber is a bit subtle. After all, against the background of mobile phone manufacturers beating the drums of war, no one distances themselves from AI in any newly launched models. AI has almost become the flagship nuclear weapon of every company—it can be used without it, but it must be available.
The reason is that when folding screens are half dead, the current smartphone industry needs a new gimmick to save lives.
This can be traced back to the past few years, when the smartphone industry went through "the gate of hell". At that time, Moore's Law, which had been pushing the consumer electronics industry forward, was facing a bottleneck, which directly caused mobile phone manufacturers and software manufacturers to "play a tacit understanding" and promote hardware replacement to become ineffective.
For this reason, panicked manufacturers even used the PC side to build anotherThreshold's ray tracing has been reproduced on mobile phones, but the mobile phone market is not as strongly tied to games and productivity as the PC market, so the final effect is not very good. However, national salvation routes such as imaging and folding screens have limited effects. On the whole, they can only serve vertical needs and are difficult to drive a large-scale replacement wave.
This is not the case with AI, because on-device AI taps into performance aspects significantly differently than smartphones in the past. An industry insider told Photon Planet that from a hardware perspective, AI models do not rely on small cores, but pay more attention to the capabilities of NPUs, and it is best to have large memory and high memory speed support.
Therefore, if a smartphone wants to become an AI phone or get a better experience on the AI application side, the hardware will inevitably be reconstructed. And this naturally means changing the phone. To tell this story, one step is extremely critical, which is to convince the broad consumer market to pay for AI phones.
Photon Planet’s survey in the middle of this year found that AI mobile phones of all major brands appear cold and indifferent in stores. Both customers and store employees are unspeakably enthusiastic about them. Recently, a SellCell survey showed that 73% of Apple Intelligence users and 87% of Galaxy AI users believe that AI lacks or has almost no value during their use.
This is not difficult to understand. After all, functions such as summary PPT, call summary, and picture generation are difficult for most users to use - no matter how iteratively upgraded and powerful, AI After all, it is difficult for mobile phones to get rid of the shackles of vertical demand and repeat the mistakes of folding screens.
Therefore, it has become a top priority for mobile phone manufacturers to present the value of on-device AI to users as early as possible.
Looking at the new phones released in the second half of this year, whether it is Apple, Samsung, or Rongmi OV, they have not skimped on ink at all. They have piled up materials in press conferences and offline stores, and have previously On the basis of AI photo retouching and AI dialogue, many system-level in-depth applications have been developed.
Among them, the most imaginative and attractive thing is the "autonomous driving" achieved by AI screen reading. AI obtains the mobile phone screen information and automatically interacts with the application to execute user instructions. ——Zhao Ming’s breakthrough of “ordering coffee with one sentence” at the Honor new product launch conference was precisely based on this ability. In the context of consumption, the honor of being the first to display this trump card also earns users' minds.
This means that at this stage, AI mobile phones have gone from integrating AI toys and tools to gradually deepening the mobile phone interaction and system levels, and showing their subversiveness to traditional mobile phones. This kind of deeply embedded AI interaction may not only promote a real reshuffle of device-side AI, but may also create new business models for mobile phone manufacturers.
AI is a life-saving straw“Excessive performance is a false proposition in the AI era. Games and photography scenes have very high performance requirements, and stacking materials It’s not completely useless work either.The advancement of chip manufacturing technology actually provides a broader space for AI. ”
A person from a mobile phone manufacturer said that device-side AI has very high system requirements, and even higher for consumer-grade products. For example, the simplest heat dissipation seems very simple, but in fact it is It’s difficult to understand. The mobile phone industry has just begun to solve the underlying memory usage problem. In the past year, many mobile phone manufacturers have also made many optimization actions for device-side deployment. However, due to poor consumer perception, Therefore, it is often ignored.
The biggest bottleneck of device-side deployment is the use of mobile phone computing resources, especially memory usage. Memory is reserved, so most manufacturers will dynamically allocate it according to the needs of the device-side AI function.
Take OPPO as an example. The previously launched device-side architecture "AI". LoRA", a dynamic deployment solution was adopted. The specific principle is that the architecture provides three end-side functions, one is occupied by the basic model, and the other two are dynamically adjusted according to the actual situation. It is reported that the peak memory usage of this solution is, It can save 75%.
The gradual maturity of the architecture means that a large-scale transformation of the ecosystem has begun. Some mobile phone manufacturers said that "the content of the conference next year is expected to be longer" because of the description of the hardware. Replaced by AI functions, which rely on experience and require more time to present.
< p>In addition to mobile phone manufacturers, large model companies including Zhipu, at the end of the year, launched the function of quickly executing complex tasks through natural language instructions based on the agent's task planning and screen information understanding capabilities. < p>Some people in the industry said that mobile phone manufacturers and The relationship between application developers is not completely antagonistic. At the current stage, it is mainly a cooperative relationship. “For developers, downloading an application does not mean using it. On-device AI can help increase the frequency of opening. Mobile phone manufacturers also need to cooperate with developers. Make the on-device AI assistant smoother to use. ”The most obvious advantage of application manufacturers is concentrated in consumption data in the vertical field. Take the photo trigger service as an example. Simply connecting to OTA requires tens of millions of vertical data for special training. Mobile phone manufacturers lack both There is no need to reinvent the wheel with relevant data, so that we can form stable cooperation with developers.
“In fact, one-step direct calls, such as navigation and shopping, are actually welcomed by application developers. This is because mobile phone manufacturers use direct methods to push services to users. "The person in charge of a mobile phone manufacturer said that the existing mobile services are already very saturated, and they are now paying more attention to how to allow users to quickly access services directly, especially when they need them. The shallower the entrance, the higher the frequency of calls. < /p>
Even so, in the AI function display of new phones at the end of the year, we can still see that the voice of mobile phone manufacturers is increasing with the acceleration of device-side deployment. When Zhao Ming used Honor to demonstrate Luckin ordering. , or Magic OS When 9.0 launched functions such as one-click shopping and price comparison, few people thought about it.Let’s ask a question: Should we open WeChat mini program, Meituan PP, or Ele.me? Is it based on user habits, or does it have a certain proportion of commercial weight?
There is no doubt that relying on direct access, smart assistants from various mobile phone manufacturers will likely change the existing form of mobile phone interaction - from mobile applications to smart assistants, from simple touch to touch and natural language , various interactive forms of intent recognition. This means that the "entry" status of mobile phone manufacturers has been further strengthened, and their control over application distribution will increase in the future - this will be another "traffic entrance" competition after pre-installed applications.
Can it change user habits?Whether users pay for AI phones is the only measurement indicator.
An earlier questionnaire survey sent by SellCell to 1,000 iPhone and Galaxy users showed an overall mismatch between purchasing decisions and usage habits. Most people are skeptical about the value of AI functions and have low willingness to pay, but nearly half of users believe that AI functions are an important basis for decision-making. In other words, users think AI functions are important, but the actual experience is not good.
The reason is that large models based on PC are somewhat "acclimated" to the mobile version. Most of the existing large model manufacturers rely on PC browser output services and focus on productivity tools, which is not fully suitable for mobile scenarios. For example, in public scenarios, voice interaction is obviously inferior to touch interaction; document summarization and article writing are not used as frequently as Moments copywriting generation.
A few days ago, Mark Gurman, a well-known technology reporter from Bloomberg, said that iOS19 will give up access to ChatGPT and instead adopt a self-developed model.
Another enemy of AI phones comes from habits. Nowadays, when the operating logic of traditional mobile phones is very mature, AI functions have indeed greatly simplified user usage habits. However, it still takes time for the user side to adapt to the new interaction. Just like the transition from digital machines to smartphones, it took several years for users to accept the mobile ecosystem and new touch interactions.
Currently, mobile phone manufacturers are constantly developing AI functions, such as AI screen reading, perhaps hoping to complete the exploration of scenarios or business models in the next era.
It is still uncertain whether technology can leverage changes in users’ usage habits. After all, as a tool, mobile phones give users a strong desire to control when using them, and AI may bring excessive “intelligence” and make users feel a sense of loss of control. How to gradually let users accept and rely on AI through step-by-step guidance is still a market problem that manufacturers need to overcome.