Just! Musk, big announcement!

Musk is constantly moving when it comes to AI.

Today, Musk announced that the early beta version of Grok Voice Mode is now live on the Grok app. Grok 3's voice mode allows users to converse with AI through natural language, enabling more intuitive and efficient information acquisition and interaction.

According to Musk's previous photos, the Grok app topped the list of free apps in the App Store in the United States, surpassing OpenAI's ChatGPT app. The xAI team revealed that the next phase of AI cluster construction has been launched, which will be about 5 times stronger than the current cluster.

At present, the global "AI war" is intensifying. In China, with the release and explosion of the DeepSeek-R1 model, the AI application model is ushering in important changes. Following mainstream domestic mobile phone manufacturers such as Huawei and Honor, Xiaomi has also officially accessed DeepSeek-R1. In addition, domestic cloud computing giants such as Huawei Cloud, e Cloud, Tencent Cloud, Alibaba Cloud, Volcano Engine, China Unicom Cloud, Mobile Cloud, and Baidu Intelligent Cloud have also been connected to DeepSeek.

Musk announced

On February 23, Beijing time, Musk announced on the social platform X that the early beta version of Grok voice mode is now available on the Grok app. "Even though this is an early beta version and there may be some issues (but we'll fix them quickly), it's still great."

Although Musk admits that the Grok voice mode is still in the early testing stage and may have some issues, he still praises its performance. The launch of Grok voice mode means that users can interact with Grok through voice, which will greatly improve the user experience and expand the application scenarios of Grok.

Grok 3's voice mode allows users to converse with AI through natural language, enabling more intuitive and efficient information acquisition and interaction. The mode offers two different voices (Ara and Grok) as well as different personalities, the ability to add custom captions, a share button, and the ability to record and share conversations.

On the 20th local time, Musk's artificial intelligence startup xAI announced that Grok 3 is open for free use (until the server is overwhelmed). Previously, Grok 3 required an X Premium+ or SuperGrok subscription to use it.

This move quickly detonated the global AI circle. Subsequently, Musk posted a photo of the Apple App Store's free app download rankings, showing that the Grok app surpassed the ChatGPT app to take the first place.

Musk even announced that Grok 3 will be used for SpaceX Mars mission calculations in the future, and predicted that "a Nobel Prize-level breakthrough will be achieved within three years", which undoubtedly adds more imagination to the future development of Grok 3.

On February 18, local time, Musk and the xAI team officially released the latest version of Grok, Grok 3, in a live broadcast.

Judging from the data displayed at the scene, Grok 3 scored higher than Google's Gemini model, Anthropic's Claude model, and OpenAI's GPT-4o model in the relevant tests of mathematics, science, and programming.

Musk said the chatbot was trained on synthetic data and was able to reflect on its own mistakes by reanalyzing the information to achieve better logical consistency. It is reported that the biggest highlight of Grok 3 is its introduction of the "Chain Of Thought" reasoning mechanism. This mechanism allows Grok 3 to solve complex problems in a step-by-step manner, just like a human.

"Grok 3 has more than 10 times the computing power of Grok 2." "We're constantly improving the model every day," Musk said. ”

The xAI team also revealed for the first time that the next phase of AI cluster building has now begun, and that "a model better than Grok 3 must excel in all aspects of deep learning science and engineering, but this is no easy task." "We've already started working on the next AI cluster, which will be about five times stronger than the current one." ”

The tide of AI applications is coming

As the "AI war" intensifies, industry insiders believe that 2025 will be the first year of the outbreak of AI terminal applications.

Recently, with the release and explosion of the DeepSeek-R1 open-source model, the AI construction and application model is ushering in important changes, which has attracted enthusiastic attention at home and abroad.

After mainstream domestic mobile phone manufacturers such as Huawei and Honor have announced their access to DeepSeek-R1, Xiaomi has also officially accessed DeepSeek-R1. On February 23rd, Xiaomi Super Xiaoai has launched DeepSeek-R1, which uses online search by default, and can be turned on by typing "Open Deep Thinking".

In addition, since the DeepSeek open source model can be deployed on public and private clouds, domestic cloud computing giants such as Huawei Cloud, e Cloud, Tencent Cloud, Alibaba Cloud, Volcano Engine, China Unicom Cloud, Mobile Cloud, and Baidu Intelligent Cloud have all connected to DeepSeek.

An Yun, deputy director of the Institute of Artificial Intelligence of Saizhi Industry Research Institute, said that domestic large models have achieved breakthrough technological progress through innovations such as open-source strategies, low-cost and efficient reasoning, and reinforcement learning combined with hybrid expert architecture (MoE). Open source has broken the technological monopoly of large enterprises and promoted the universalization of AI technology. Its low-cost algorithm optimization model has changed the long-term dependence on computing power stacking and promoted the efficiency-oriented competitive landscape.

At the same time, domestic large models with low-cost and efficient inference models have also affected the upstream and downstream of the AI industry and affected the capital market. In the field of manufacturing, it helps factories to achieve production process optimization and quality control; In the financial sector, the efficiency of AI model-driven risk assessment has been effectively improved. In the medical industry, AI-assisted diagnosis will shorten the diagnosis and treatment cycle. In addition, data-driven decision-making support in the fields of government affairs, education, and transportation through large models will promote both efficiency and service quality.

Guojin Securities believes that if 2023 is the first year of AI training, and 2024 is the first year of AI inference, then 2025 will be the first year of the outbreak of AI terminal applications. Domestic large models have opened a new stage in the development and application of global large models, and we continue to be optimistic about AI infrastructure, focusing on AI applications.

Minsheng Securities pointed out that the most beneficial direction of DeepSeek is the cloud industry chain. Cloud resources have become "hard currency", and cloud vendors hold computing resources and open the upper limit of the threshold. In the context of enterprises scrambling to deploy the DeepSeek model, cloud vendors are expected to quickly feed back on the premise of having sufficient computing power "ammunition" and extensive user coverage.

Analysts believe that cloud computing, as the underlying computing power support of large models, is expected to continue to benefit, and the parity of AI computing power brought about by technological progress will not only benefit traditional AI computing power suppliers, but will also extend to non-traditional large cloud manufacturers. On the one hand, leading cloud computing companies have successively connected to DeepSeek, which is expected to promote the rapid growth of computing power leasing and AI service revenue of cloud service providers. On the other hand, DeepSeek's cost reduction is expected to accelerate the prosperity of the application side and further boost the demand for cloud computing power. For domestic cloud vendors, Deepseek will accelerate the digital transformation of enterprises to the cloud, and further improve the profit margin of cloud business under the scale effect.