The new track of end-side AI big models has recently ushered in two more heavyweight players - Apple and Huawei - which provides new hope and momentum for smart terminals (mainly cell phones) to come out of the doldrums.

Apple GPT is an AI tool developed by Apple based on its self-developed Ajax framework, and is currently undergoing a small-scale internal test within Apple; Huawei, on the other hand, demonstrated its end-side AI big model at the application level through the intelligent assistant "Xiaoyi" of HarmonyOS 4, which was released on the afternoon of August 4 at the HDC 2023 developer conference. Huawei, on the other hand, demonstrated the end-side capabilities of AI big models at the application level by releasing the intelligent assistant "Xiao Yi" of HarmonyOS 4 at the HDC 2023 developer conference on August 4 afternoon.

Previously, Meta, OpenAI, Qualcomm, Google, Tencent and Baidu have launched or will soon launch applications or technical frameworks that support end-side AI models, and are rapidly outlining the technical and market prospects of this new direction.

HarmonyOS 4: What are the end-side AI big model capabilities

On August 4, Huawei officially released HarmonyOS 4. Huawei HarmonyOS is a full-scene distributed operating system for the Internet of Everything, supporting a wide range of terminal devices such as cell phones, tablets, smart wearables, and smart screens.

In the fourth version of this famous IoT OS from Huawei, the ability to run end-side AI big models (i.e., running big models on the phone side) becomes the key core.

"Today we have entered the era of big models, and Huawei Pangu big models will help the Hongmeng ecosystem." Yu Chengdong, Huawei's Managing Director, CEO of Terminal BG, and CEO of Smart Car Solutions BU, said, "With Pangu's underlying capabilities, Huawei will bring a new transformation of AI experience for users with smart terminal interactions, high-level productivity efficiency, and personalized services."

Text generation based on AI big models is part of HarmonyOS 4. With the growing development of the big model, Xiaoyi receives a new upgrade that enhances the capabilities in three areas: interaction, productivity, and personalized service.

Text generation and summarization capability, which is a standard capability for all the big model apps launched, HarmonyOS 4 is no exception. With Xiaoyi, content and text in images can be recognized, text content can be read aloud and more services can be accessed.

According to Ziad Asghar, senior vice president of product management and head of AI at Qualcomm, big models will rapidly reshape human-computer interaction.

The interaction changes in HarmonyOS 4 don't look significant. Based on voice interaction, it expands on multiple forms of input such as text, images and documents. For example, by communicating with the AI in a natural way with everyday speech, Little Art can automatically complete specified tasks. This has long been achievable in the iPhone through Siri.

However, as the first intelligent assistant with AI big model capability, Xiaoyi has a deeper understanding of natural semantics.

For example, Siri, the AI assistant of iPhone, requires clear and precise semantics of voice commands input by C-support users, while HarmonyOS 4 can understand voice commands with relatively less obvious meanings, and thus smartphones equipped with HarmonyOS 4 can complete tasks more accurately and quickly.

In addition, it can also access services faster than before due to the large-model AI generalization capability that Xiaoyi has and the fact that the phone's intelligent assistant connects to a wide range of services and special scenarios.

For example, if an event poster image is received and the C-end user gives instructions to Little Art, the AI can automatically recognize the address on the image and give a navigation button or save the poster phone number as a contact method.

If browsing an English news article with thousands of words, Xiaoyi can directly speed read the article, do the translation and complete the summary, and can also answer questions related to the news article.

HarmonyOS 4 also has generative capabilities. For example, it can automatically generate all kinds of business e-mail content or generate images; it can also use the personalized photos saved on the terminal, and Xiao Yi can call the AI drawing function to generate various styles of pictures.

As C-side users and Xiaoyi continue to interact, Xiaoyi will continue to improve its AI capabilities. These interaction data will be kept on the end-side to protect user privacy.

Personalization is a feature of smartphone applications. Xiaoyi has the memory ability to understand the "owner" more and more with the use of a long time, to give more intimate advice: for example, travel, before the trip to provide complete and rich travel information, and after the landing to provide local information, and according to the user's habits, to achieve personalized recommendations.

According to Huawei, small art of these new capabilities, will be opened in late August public test experience.

This is not Huawei's first attempt to access AI big models to mobile terminals. In March this year, Huawei released the P60 cell phone, which comes with a smart search function. This function is based on multimodal large model technology, which realizes the operation of large models on the cell phone side by doing miniaturization of the model on the cell phone side.

Apple moves slowly, Qualcomm is enthusiastic

Smart terminal vendors aiming at end-side AI big model capabilities are not only Huawei, but also Apple.

Apple is secretly developing GPT software called Apple GPT, based on Apple's self-developed Ajax framework development.

However, at present, Apple GPT lacks more details, and it is difficult for the outside world to know what the highlights of its technology or application capabilities are. In this regard, Apple made it clear that it has not yet decided on the future direction of Apple GPT development.

Previously, in the second fiscal quarter earnings call, Apple CEO Cook admitted that Al has great potential, but how to use AI technology still requires in-depth thinking. Cook said that Apple has integrated AI technology into its products and services, and will continue this process in the future.

The world's first intelligent assistant for the consumer-grade market, large-scale landing, the application of NLP technology, iPhone Siri, is 12 years old this year. With the new upgrade of Huawei Xiaoyi, millet Xiaoyi, Baidu Xiaodu a number of "juniors" compared to Siri is not only not intelligent, but also seems a bit "retarded".

The outside world that Apple landed GPT ability carrier is likely to be the presentation of the "twilight" state of Siri, but this has not been confirmed by Apple.

Nevertheless, the signs of Apple's entry into the end side of the AI model are very many. For example, in January this year, Apple launched a new program for Apple Books to add digital narration function, in order to automatically generate high-quality AI narration audio from written text; in the iOS17 update, Apple used the Transformer language model to improve the prediction and spelling correction function of the input method.

Also, the new AirPods Pro uses machine learning (ML) to implement an adaptive audio mode that automatically adjusts the volume by recognizing the external environment; iPadOS 17 uses a machine learning model to enable recognition of PDF fields; and Vision Pro uses the machine learning technology "encoder-decoder" Neural Networks to build digital bodies for users.

Apple how to layout AI track, this is a system of big problems, it is difficult to say in hundreds of words to understand. But the Siri development team of technical engineers to change the Siri "stupid" seems to lack of real motivation, behind the reflection of the reality of "organizational dysfunction and lack of ambition", is likely to affect the efficiency of Apple on the end side of the ground AI big model.

But Apple's inefficiency can not change the scale of deployment of end-side AI models in smart terminals, which is an indisputable fact. Because Qualcomm, Meta, OpenAI, Google, Amazon, Tencent and Baidu are all committed to realizing the lightweight deployment of AI models in mobile terminals.

The OpenAI team is launching a mobile ChatGPT app covering iOS/Android in the fourth week of July; Meta will work with Qualcomm to make its open source big model Llama 2 capable of running on flagship smartphones and PCs from 2024; Qualcomm says it is migrating more generative AI use cases to the edge side, and now AI models with more than one billion parameters are capable of running on run on cell phones with similar levels of performance and accuracy to the cloud.

Google, Tencent, Baidu and others have tightly integrated model compression technology with mobile model deployment frameworks/tools.

Among them, Qualcomm is enthusiastic about it. Qualcomm CEO Cristiano Amon emphasized in a statement that Qualcomm's ability to run AI models on cell phones rather than cloud servers provides Qualcomm with the opportunity to achieve an "inflection point" that will drive future growth.

"In short, we are uniquely positioned to help shape and capitalize on the upcoming Gen AI opportunity on devices." Amon said.

It remains uncertain when the overall smartphone market decline will stop; however, this wave of many B-side players ushered in by the end-side AI mega-modeling track is where new hope lies to reshape the application and market landscape of this dwindling consumer electronics category.

The new track of end-side AI big models has recently ushered in two more heavyweight players - Apple and Huawei - which provides new hope and momentum for smart terminals (mainly cell phones) to come out of the doldrums.

Apple GPT is an AI tool developed by Apple based on its self-developed Ajax framework, and is currently undergoing a small-scale internal test within Apple; Huawei, on the other hand, demonstrated its end-side AI big model at the application level through the intelligent assistant "Xiaoyi" of HarmonyOS 4, which was released on the afternoon of August 4 at the HDC 2023 developer conference. Huawei, on the other hand, demonstrated the end-side capabilities of AI big models at the application level by releasing the intelligent assistant "Xiao Yi" of HarmonyOS 4 at the HDC 2023 developer conference on August 4 afternoon.

Previously, Meta, OpenAI, Qualcomm, Google, Tencent and Baidu have launched or will soon launch applications or technical frameworks that support end-side AI models, and are rapidly outlining the technical and market prospects of this new direction.

HarmonyOS 4: What are the end-side AI big model capabilities

On August 4, Huawei officially released HarmonyOS 4. Huawei HarmonyOS is a full-scene distributed operating system for the Internet of Everything, supporting a wide range of terminal devices such as cell phones, tablets, smart wearables, and smart screens.

In the fourth version of this famous IoT OS from Huawei, the ability to run end-side AI big models (i.e., running big models on the phone side) becomes the key core.

"Today we have entered the era of big models, and Huawei Pangu big models will help the Hongmeng ecosystem." Yu Chengdong, Huawei's Managing Director, CEO of Terminal BG, and CEO of Smart Car Solutions BU, said, "With Pangu's underlying capabilities, Huawei will bring a new transformation of AI experience for users with smart terminal interactions, high-level productivity efficiency, and personalized services."

Text generation based on AI big models is part of HarmonyOS 4. With the growing development of the big model, Xiaoyi receives a new upgrade that enhances the capabilities in three areas: interaction, productivity, and personalized service.

Text generation and summarization capability, which is a standard capability for all the big model apps launched, HarmonyOS 4 is no exception. With Xiaoyi, content and text in images can be recognized, text content can be read aloud and more services can be accessed.

According to Ziad Asghar, senior vice president of product management and head of AI at Qualcomm, big models will rapidly reshape human-computer interaction.

The interaction changes in HarmonyOS 4 don't look significant. Based on voice interaction, it expands on multiple forms of input such as text, images and documents. For example, by communicating with the AI in a natural way with everyday speech, Little Art can automatically complete specified tasks. This has long been achievable in the iPhone through Siri.

However, as the first intelligent assistant with AI big model capability, Xiaoyi has a deeper understanding of natural semantics.

For example, Siri, the AI assistant of iPhone, requires clear and precise semantics of voice commands input by C-support users, while HarmonyOS 4 can understand voice commands with relatively less obvious meanings, and thus smartphones equipped with HarmonyOS 4 can complete tasks more accurately and quickly.

In addition, it can also access services faster than before due to the large-model AI generalization capability that Xiaoyi has and the fact that the phone's intelligent assistant connects to a wide range of services and special scenarios.

For example, if an event poster image is received and the C-end user gives instructions to Little Art, the AI can automatically recognize the address on the image and give a navigation button or save the poster phone number as a contact method.

If browsing an English news article with thousands of words, Xiaoyi can directly speed read the article, do the translation and complete the summary, and can also answer questions related to the news article.

HarmonyOS 4 also has generative capabilities. For example, it can automatically generate all kinds of business e-mail content or generate images; it can also use the personalized photos saved on the terminal, and Xiao Yi can call the AI drawing function to generate various styles of pictures.

As C-side users and Xiaoyi continue to interact, Xiaoyi will continue to improve its AI capabilities. These interaction data will be kept on the end-side to protect user privacy.

Personalization is a feature of smartphone applications. Xiaoyi has the memory ability to understand the "owner" more and more with the use of a long time, to give more intimate advice: for example, travel, before the trip to provide complete and rich travel information, and after the landing to provide local information, and according to the user's habits, to achieve personalized recommendations.

According to Huawei, small art of these new capabilities, will be opened in late August public test experience.

This is not Huawei's first attempt to access AI big models to mobile terminals. In March this year, Huawei released the P60 cell phone, which comes with a smart search function. This function is based on multimodal large model technology, which realizes the operation of large models on the cell phone side by doing miniaturization of the model on the cell phone side.

Apple moves slowly, Qualcomm is enthusiastic

Smart terminal vendors aiming at end-side AI big model capabilities are not only Huawei, but also Apple.

Apple is secretly developing GPT software called Apple GPT, based on Apple's self-developed Ajax framework development.

However, at present, Apple GPT lacks more details, and it is difficult for the outside world to know what the highlights of its technology or application capabilities are. In this regard, Apple made it clear that it has not yet decided on the future direction of Apple GPT development.

Previously, in the second fiscal quarter earnings call, Apple CEO Cook admitted that Al has great potential, but how to use AI technology still requires in-depth thinking. Cook said that Apple has integrated AI technology into its products and services, and will continue this process in the future.

The world's first intelligent assistant for the consumer-grade market, large-scale landing, the application of NLP technology, iPhone Siri, is 12 years old this year. With the new upgrade of Huawei Xiaoyi, millet Xiaoyi, Baidu Xiaodu a number of "juniors" compared to Siri is not only not intelligent, but also seems a bit "retarded".

The outside world that Apple landed GPT ability carrier is likely to be the presentation of the "twilight" state of Siri, but this has not been confirmed by Apple.

Nevertheless, the signs of Apple's entry into the end side of the AI model are very many. For example, in January this year, Apple launched a new program for Apple Books to add digital narration function, in order to automatically generate high-quality AI narration audio from written text; in the iOS17 update, Apple used the Transformer language model to improve the prediction and spelling correction function of the input method.

Also, the new AirPods Pro uses machine learning (ML) to implement an adaptive audio mode that automatically adjusts the volume by recognizing the external environment; iPadOS 17 uses a machine learning model to enable recognition of PDF fields; and Vision Pro uses the machine learning technology "encoder-decoder" Neural Networks to build digital bodies for users.

Apple how to layout AI track, this is a system of big problems, it is difficult to say in hundreds of words to understand. But the Siri development team of technical engineers to change the Siri "stupid" seems to lack of real motivation, behind the reflection of the reality of "organizational dysfunction and lack of ambition", is likely to affect the efficiency of Apple on the end side of the ground AI big model.

But Apple's inefficiency can not change the scale of deployment of end-side AI models in smart terminals, which is an indisputable fact. Because Qualcomm, Meta, OpenAI, Google, Amazon, Tencent and Baidu are all committed to realizing the lightweight deployment of AI models in mobile terminals.

The OpenAI team is launching a mobile ChatGPT app covering iOS/Android in the fourth week of July; Meta will work with Qualcomm to make its open source big model Llama 2 capable of running on flagship smartphones and PCs from 2024; Qualcomm says it is migrating more generative AI use cases to the edge side, and now AI models with more than one billion parameters are capable of running on run on cell phones with similar levels of performance and accuracy to the cloud.

Google, Tencent, Baidu and others have tightly integrated model compression technology with mobile model deployment frameworks/tools.

Among them, Qualcomm is enthusiastic about it. Qualcomm CEO Cristiano Amon emphasized in a statement that Qualcomm's ability to run AI models on cell phones rather than cloud servers provides Qualcomm with the opportunity to achieve an "inflection point" that will drive future growth.

"In short, we are uniquely positioned to help shape and capitalize on the upcoming Gen AI opportunity on devices." Amon said.

It remains uncertain when the overall smartphone market decline will stop; however, this wave of many B-side players ushered in by the end-side AI mega-modeling track is where new hope lies to reshape the application and market landscape of this dwindling consumer electronics category.