Vivo is not installing anymore! The sales king has launched PhoneGPT, which can refurbish all the phone calls, text messages and notes! It is worthy of being the model hiding leader in the domestic mobile phone industry!

Editor | Yifeng

Produced by | 51CTO Technology Stack (WeChat ID: blog)

How else can AI phones be popular?

The answer announced by vivo at the Developer Conference today is definitely eye-opening!

By integrating user memory, terminal models, and active execution capabilities, vivo has developed PhoneGPT, claiming to be the autonomous driving in the AI ​​assistant industry!

That is to say, PhoneGPT is no longer satisfied with simple task execution, but can make its own decisions, break down tasks, and finally take actions and complete tasks through technologies such as screen recognition.

This may still be a bit abstract, so it’s more intuitive to watch vivo’s Blue Heart Little V ordering video!

After the user sent a dining request to Xiao V, Xiao V automatically called up the relevant app, searched for the restaurant's phone number, and made a call to complete the reservation with the clerk! The entire execution process, including the conversation with the real person, was very smooth.

In addition to smart phones, the blue manufacturer's hard technology is not inferior at all, and the self-developed Blue Heart large model has online performance. On the terminal side, it put forward the conclusion that "3 billion parameters are the golden size of the terminal side", and released the new Blue Heart 3B, which beat the leaders of small models such as Gemma-2 and Phi-3-mini in the ranking.

One can't help but sigh that vivo is actually a hidden giant in the modeling industry.

A little knowledge, vivo has been sitting firmly on the throne of domestic mobile phone sales for three years, and the number of users of vivo AI has exceeded 500 million worldwide.

picture

If you want to understand the future of AI phones, the AI ​​strategies of vivo and Blue Heart Intelligence are definitely worth a look.

1. PhoneGPT: AI must not only “execute” but also “understand you”

The core highlight of vivo Blue Heart Intelligence is that they have achieved a deep integration of big models and operating systems.

Vivo Vice President Zhou Wei said that the effect of this implementation is that the big model can understand the user's personal context and preferences, allowing the system to proactively provide personalized services.

picture

Specifically, the mobile phone system reconstructed by AI has brought about a revolution in three experiences, namely, interactive experience, service experience and connection experience.

1. Interactive experience

Many people have complained about the rigidity and mechanicalness of Apple's Siri. Before AI 2.0, our mobile assistants could only perform some simple execution tasks.

With the big model, AI assistants can truly understand intentions from all angles at the text, voice, image and even behavioral levels, achieving the so-called "execution and understanding you".

In terms of interactive experience, the example given by others is voice interaction! As the most commonly used and most intuitive form of interaction in our daily life, what tricks can vivo come up with?

picture

Xiao V’s dialect dialogue means that users no longer need to switch between “Trump” and “Cantonese-Mandarin”, and Xiao V will also answer in dialect, which shortens the distance between AI and users.

2. Service Experience

In terms of service improvement, I think vivo has really lived up to the often-quoted saying, "all functions are worth redoing with AI."

vivo has refurbished basic functions such as phone calls and text messages. For example, the content of phone calls can be transcribed into text or translated, the note-taking function can be directly polished using AI, and photos can be directly scanned and recognized, etc.

The editor feels that these designs are not false needs, but actually use AI to help users solve some problems.

picture

Since it is a developer conference, vivo generously announced that it will open up its platform capabilities so that third-party applications can also access these AI black technologies.

picture

3. Link experience

The connection experience mentioned here mainly refers to the changes in the relationship between users and services in the AI ​​era.

AI has indeed injected vitality into services. Now people and services can go both ways! In terms of developers, vivo has made efforts to better distribute the services provided by the APP and find users who need it.

picture

Especially for lightweight and atomic services, vivo has built system-level perception and intent recognition to enable the system to actively distribute services.

picture

On the user side, vivo proposed an interesting concept: "shared memory between people and devices."

It sounds like AI is more like a friend who will always be with us. As we use our mobile phones, AI will continue to expand our knowledge graph, which is equivalent to the mobile phone system becoming "better and better with use."

What specific function can this achieve?

picture

Using this on-site example, we immediately understood it.

For example, a user calls up Xiao V and says, "Please help me find the photo of Xiao Bei and I at Universal Studios last year, send it to him and wish him a happy birthday." This example is very challenging for AI, as the smart assistant needs to remember the user's social relationships and the other person's appearance, find photos in the album that meet the conditions of person, time, and location, find people with the note "Xiao Bei" across applications to send messages, and finally provide exclusive services such as recommending birthday restaurants.

picture

But I feel a little uneasy about leaving all these things to AI, as if someone has gone through all my photo albums.

However, vivo also attaches great importance to security issues. It even overcame huge technical challenges and not only moved the call summary, document summary, local knowledge graph and other privacy information that you don't want others to see to the client, but also thoughtfully implemented the audit function on the client. Now you can breathe a sigh of relief and use AI with confidence.

picture

Then, since there are multiple intelligent agents, it is best to let them cooperate to achieve the effect of 1+1 greater than 2. If it is equipped with intention understanding, the center responsible for scheduling can also be completed by AI to complete part of the work, and it can make autonomous decisions based on understanding people.

This combination realizes the phoneGPT at the beginning of our article that can provide an intelligent driving experience!

The restaurant ordering in the video is based on the multimodal function of the Blue Heart model. It can actively complete tasks based on user intent. Through screen recognition, AI can operate the controls and execute each step by itself.

picture

According to the information provided by the user, the accuracy of each step of phoneGPT is 94%, and each step takes only 2 seconds. For the sake of safety, this function will also be developed on the terminal side, and the execution effect will be further improved.

Wow, the era of AI true personal assistant has arrived!

2. New Blue Heart large model matrix released, Blue Heart 3B surpasses Phi-3-mini

Since the release of GPT-4o, the multimodal capabilities of the model have become a focus.

In the new model matrix, Blue Heart’s speech, image and multimodal large models are the first to appear.

picture

Since it is a mobile phone manufacturer, the terminal model is definitely the top priority of development.

Based on the previously developed 1B, 7B, and 13B edge models, vivo has been continuously testing. After comprehensively considering the model effect, performance, memory, and power consumption, it tried to find the optimal solution and proposed that the 3 billion parameter level is the golden size for the edge.

picture

Under the guidance of the research, vivo's 3 billion blue heart end-side large model - Blue Heart 3B was officially launched. Compared with the 7B large model launched last year, Blue Heart 3B has achieved a 300% performance improvement, and also reduced power consumption by 46% and memory usage by 63%. The performance of Blue Heart 3B is terrifying, achieving an extreme word output speed of 80 words per second, and achieving an extremely low energy consumption of 450mA, and only occupies 1.4GB of memory.

picture

Zhou Wei said that Blue Heart 3B has broken through the impossible triangle of "strong capability - low consumption - small model" of the edge model, and has achieved industry-leading level on many lists.

picture

vivo strives to create "the big model that understands Chinese people best". Specifically in terms of multimodal big models, they have improved the ink style of the raw image function and the performance of Chinese character-related creations.

picture

Looking at this list, we can see that vivo has a solid accumulation of model capabilities. As for the voice list, voice has always been Doubao's strong core technology. Being able to surpass Doubao in voice interaction means that daily conversation tasks are basically no problem.

picture

3. Conclusion: The best AI technology should be implemented in applications

After showing off its AI strength, vivo began to introduce its self-developed system OriginOS 5.

Just as they themselves said: Not only do we need good technology, but it also needs to be implemented at the product level.

So, we saw vivo's work based on deep thinking about interface and user operation - Atomic Island.

Based on intent recognition, users only need to copy or drag text and other content to the Atomic Island, and it will automatically recommend the APP you want to open, truly freeing your hands and making technology one step ahead of others.

When I first saw the words "Atomic Island", it might be easy to think that this is an imitation of Apple's Smart Island. But to be honest, the former brought me unexpected surprises, which shows how deeply vivo thinks about how AI can change the form of interaction.

vivo's developer conference has given people a new understanding of this domestic mobile phone sales champion. When it comes to the strategic concept of AI, we not only have the skills, but also the principles.