Smartphone manufacturers rush into the AI large model track, technological innovation or just following the trend?

In the current business environment, seemingly glamorous "opportunities" often become shackles that bind people.

Recently, global chip giant Qualcomm plans to lay off about 1,258 people in California. This personnel upheaval is expected to erupt on December 13. Qualcomm's last quarterly report has already hinted at this "wave of layoffs." From a financial structure perspective, mobile chips are Qualcomm's main source of income, accounting for more than half, but revenue in the third quarter fell by 21.6% year-on-year. The saturation of the smartphone market is quietly affecting upstream supply chain giants.

Since 2019, the wave of 5G-driven smartphone upgrades has lasted nearly four years. However, by 2022, the global smartphone replacement cycle reached an all-time high of 43 months. Over the past five years, the mobile phone industry has been seeking innovative breakthroughs. But when even market leaders struggle to launch shocking new features, it becomes even harder for other manufacturers to maintain their market position. More and more consumers are beginning to question the value of upgrading to a new phone.

Experts have pointed out that large models, especially in the AI field, may become a real breakthrough point. Although it is still unclear how to maximize the use of these potentials, domestic smartphone giants have set their sights on AI large models, attempting to carve out new battlefields.

Mobile Giants Compete in the Large Model Arena

Domestic smartphone manufacturers are scrambling to chase the big model craze.

At its annual conference, Xiaomi launched its self-developed AI large model, achieving impressive results on the C-Eval and CMMLU testing platforms. This 1.3 billion parameter large model has been deployed on mobile devices and can even rival the 6 billion parameter cloud model in certain scenarios.

Huawei announced that HarmonyOS 4 will integrate the "Pangu Large Model" to achieve a higher level of system integration.

OPPO recently publicly tested the "Xiao Bu Assistant" based on the AndesGPT large model technology. AndesGPT is an advanced generative large language model based on a hybrid cloud architecture, performing excellently in multiple authoritative evaluation rankings.

vivo plans to unveil its self-developed AI large model and new operating system at the developer conference on November 1. vivo has created an AI large model matrix covering three parameter levels: one billion, ten billion, and one hundred billion, aiming to meet the diverse needs of application scenarios.

Major brands are actively investing in the field of AI large models, hoping to create a more competitive image in the high-end market. Industry insiders believe that emphasizing AI features can not only stimulate user demand for high-end products but also drive up product prices, creating higher profits for brands. The next two years may witness a significant explosion of innovation in AI smartphones.

The competition for large models on mobile ends is gradually unifying the path

Although manufacturers promote that running large models on mobile phones is easy, there are many challenges in actual operation. Large models have high requirements for mobile hardware, especially the processor and memory. Too large a model may lead to decreased phone performance or even crashes. In addition, generation speed, power consumption, and heat dissipation are all key factors that need to be considered.

Therefore, the industry focus has shifted to edge-cloud collaboration. MediaTek has collaborated with manufacturers such as OPPO and vivo to jointly develop lightweight deployment solutions for large models on the edge. Large models on the edge can provide faster response speeds and better data security.

However, relying solely on mobile devices cannot solve all problems. Most manufacturers adopt a hybrid strategy that combines local and cloud solutions, determining whether to process data locally or transfer it to the cloud based on the complexity of the issues. This approach can effectively save costs while meeting users' needs in various aspects such as computing power, performance, energy consumption, and privacy protection.

Potential Challenges Under Manufacturer FOMO

Despite mobile phone manufacturers actively exploring the application of AI large models, they still face numerous challenges:

1. The definition of "large model" is ambiguous. The number of model parameters on mobile devices is far from that of a true large model, raising doubts about whether it should be referred to as a "large model."

2. Technical compression may affect performance. To adapt to mobile devices, manufacturers have to significantly compress the models, which may impact their deep learning capabilities.

3. The application scenarios are singular. Currently, mobile manufacturers seem to overly concentrate the application of AI large models on voice assistants, and whether this truly meets user needs is open to discussion.

4. The innovative value is limited. Although the new AI large model has debuted in the system, there is not much breakthrough in core functions compared to existing voice assistants.

Overall, the true popularization of AI large models in the mobile phone sector still requires time. The current various efforts are just the beginning of exploration. Mobile phone manufacturers need to find a balance between technological innovation and practical application, genuinely creating value for users rather than simply following technological trends.