Is it a blessing or a curse for workers to install large models on terminals?

Workers in various industries were worried about being replaced by AI at the beginning, but now "Zhenxiang" has begun to use this "sharp tool" to continuously improve efficiency and bold innovation at work. . According to a report jointly released by Sutu.com and Big Model House, the global large model market will reach US$21 billion by 2023, and its size will grow to US$109.5 billion by 2028.

However, from the perspective of user usage, the core of the cloud AI large model is universal and public knowledge, which is the accumulation of human knowledge base. However, it does not know the user's personal needs, because the user cannot let it learn too many personal living habits, which is very private. For example, large-scale models in the cloud can write science fiction, but it is difficult to write year-end reports of users over the past ten years and judge corresponding trends. Because this involves user privacy and security, large-scale cloud models cannot write reports without obtaining the user's work status over the past ten years. Therefore, truly helping users better manage their own affairs still requires the ability of on-device AI

Device-side AI can better provide users with personalized services and support, and help users better manage their own affairs. According to IDC estimates, by 2026, nearly 50% of terminal devices in the Chinese market will have AI engine technology in their processors. AI is expected to explode on the edge and end sides.

Terminal manufacturers have entered the "Battle of Hundreds of Models"

Mobile phone companies' exploration of AI large models has gradually advanced from the technology research and development stage to the application and industrial layout stage.

In August this year, Huawei announced at its 2023 Developer Conference that the HarmonyOS 4 system was fully integrated with the Pangu model, marking Huawei’s Harmony OS as the world’s first mobile terminal operation with embedded AI large model capabilities. The system is also the first mobile operating system with AI large model capabilities.

Huawei mobile phone smart assistant Xiaoyi has AI large model capabilities and has been enhanced in three directions: interaction, productivity improvement, and personalized services. It not only supports users to interact with it using natural language, but also helps users output short compositions. , pictures, videos, etc.

Shortly after Huawei announced the introduction of AI large models, on August 14, Xiaomi founder and chairman Lei Jun said in his annual speech that Xiaomi had organized a large model team in April this year and would work on it in business For application, the first application is "Xiao Ai Classmate", which will be upgraded to a large model version and will be invited to test starting from the 14th. Lei Jun said that the main direction of Xiaomi's large models is lightweight and local deployment. Xiaomi is a mobile phone company. It must use its own advantages to make large models available on mobile phones through lightweight and local deployment. Based on this, Xiaomi A bold idea was put forward to optimize tens of billions of large models.

Recently, Xiaomi released the XiaoAi large model version installation experience package, which all Xiaomi users can download and experience. According to the official introduction, after a new upgrade, Xiao Ai, which has joined the big model, has more functions, including text generation, personalized assistant, content summary, character customization, image generation, multi-language translation, Xiaomi product assistant, etc.

vivo is also one of the first technology companies to focus on AI application in mobile phone innovation.

In 2017, Shen Wei, the founder and president of vivo, proposed that artificial intelligence is the foundation of all technological innovations, and heavy investment in this field should be made for a long time to come. Since then, vivo has begun to form corresponding professional teams and prepare for the Global Research Institute of Artificial Intelligence to accelerate the accumulation of advantages in four aspects: data, talent, algorithms, and computing power. In 2018, vivo established Map The research institute builds a knowledge graph through a professional team and accumulates massive data. At present, vivo's overall map data has exceeded 2800T, and the cleaned data used for large model training is about 150T, which is equivalent to 160 million books, which is enough to meet the current demand for knowledge and information of large models.

At the same time, vivo has increased its investment and construction of artificial intelligence talent teams. Not only does its global AI research institute have more than 1,000 people, but by 2021, there will be more than 600 people focusing on the research and development of large models, laying the foundation for follow-up The explosion of technology reserves a solid talent pool.

While other mobile phone companies are still working hard to improve hardware configurations and compete for the market by lowering prices, vivo has achieved a leading position in the field of artificial intelligence research and development, and has invested more than six years to achieve a degree of saturation. This is also an important reason why vivo can once again lead other mobile phone companies and release a large self-developed AI model

AI intelligent artificial intelligence opens the second half

For mobile phone manufacturers, large models are a A technical mountain that must be conquered as soon as possible.

On the one hand, the universal capabilities of large models are regarded as the key for smartphones to get rid of the "parametric theory", and their broad application space has become a must-have for major mobile phone manufacturers; on the other hand, The innovative experience that large models can bring is accelerating the generational change of the mobile phone industry towards the "AI era". The differences in large model innovation and application capabilities may directly determine the market position of mobile phone manufacturers in the next stage.

In order to achieve this goal, mobile phone manufacturers need to strengthen the foundation of large-scale model technology, and at the same time find an effective way to combine large-scale models with smartphones, and transform the capabilities of large-scale model technology into mobile phone applications that can be perceived, understood and used by users. Service

Judging from official information, vivo’s self-developed AI large model will be fully integrated with OriginOS 4 and will be launched on vivo and iQOO’s flagship series models in the future. Specifically, at the Boao Forum for Asia in September this year, Zhou Zhou, vice president of vivo, vice president of OS products, and director of the vivo AI Global Research Institute, said that through in-depth thinking about the mobile phone ecology and user usage characteristics, vivo will next Released 5 self-developed large models and launched a series of solutions.

vivo plans to self-develop a large AI model, release its service capabilities with a smarter assistant image, empower the small v assistant with broader smart service capabilities, and bring disruption in learning, life, work and other scenarios. Innovation. In the future, vivo believes that the combination of smartphones and large AI models will provide high-level presentations in the form of “intelligent bodies”. Mobile phones must not only sense information, but also identify intentions, help with decision-making, and gradually move towards an active intelligent state of execution and management execution, ultimately promoting the entire "intelligent body" to achieve spontaneous intelligent evolution

At the same time, taking into account the large scale of AI Vivo chose to release five large models with three different parameter levels to resolve the cost, performance, power consumption and other issues caused by different model sizes and different deployment ends. Surround has said that vivo will contribute its 7 billion-level large model capabilities, training capabilities, tuning capabilities, and data hot-swapping capabilities in various industries, and take the initiative to assume the responsibility of the industry leader to build a win-win and joint construction , co-created multi-party value, promotes the prosperity and development of innovative applications of AI large models in the mobile phone industry.

It is not difficult to find that the fundamental reason why mobile phone manufacturers are rushing to grab large models lies in the advanced experience and innovative breakthroughs that large models bring to smartphones. Large models for smartphones are far from simple technology access. Behind the scenes, mobile phone manufacturers are tested on the integration of large models and underlying systems, algorithm allocation and optimization, cost and power consumption control, as well as understanding and judgment of application scenarios. Comprehensive ability.

At present, vivo’s self-developed large models have initially shown an industry-leading trend in terms of technical strength, landing scene experience, and number of large models. What is worth looking forward to is that at the upcoming 2023 vivo Developer Conference, the integration results of vivo’s self-developed AI large model and OriginOS 4 will be fully presented. At that time, vivo’s specific practice of human-computer interaction in the new era will also be oriented to Public release.

The above is the detailed content of Is it a blessing or a curse for workers to install large models on terminals?. For more information, please follow other related articles on the PHP Chinese website!