After months of hard work, the Tongyi Qianwen Qwen series open-source model has been upgraded to Qwen2, including five models of different scales: 0.5B/1.5B/7B/57B/72B, supporting a maximum context length of up to 128K tokens.

360 AI search prediction capabilities are showcased. After the national college entrance examination Chinese test ended this morning, some netizens discovered that among the 10 composition topics predicted by Zhou Hongyi, three of them matched the composition topics of the Beijing, National A, and New Curriculum I papers in this year's college entrance examination.

What are the other hot topics in the domestic AI industry that deserve attention in the past day? Let Crow Tab take you to take a look.

/ 01 / Domestic News

1) Alibaba Cloud Launches Tongyi Qianwen Qwen2 Large Model and Opens Source

Alibaba Cloud officially launched the Tongyi Qianwen Qwen2 series of large models and opened source them on the Hugging Face and ModelScope platforms. The Qwen2 series includes five models of different scales: 0.5B/1.5B/7B/57B/72B.

These models support extended context lengths, up to 128K tokens. The performance of the Qwen2-72B model surpasses that of the US Llama3-70B and closed-source models in China such as Wenxin 4.0, Doubao pro, and Hunyuan pro. In addition to training in Chinese and English, Qwen2 has incorporated high-quality data from 27 languages.

2) Tencent Opens Source the Portrait Photo-to-Video Model V-Express

Tencent has open-sourced an innovative model called V-Express that can generate videos using portrait photos. The model achieves comprehensive control over posture, input images, and audio by balancing different control signals. The step-by-step discarding operation of the V-Express method solves the problem of weak signals under strong signal interference, successfully achieving audio-controlled video generation.

3) Vidu Video Large Model Upgrade: Duration Extended to 32 Seconds, Supporting Audio-Video Synthesis

The first long-duration, high-consistency, and high-dynamism video large model in China, Vidu, jointly released by Shengshu Technology and Tsinghua University, has recently completed three significant technological iterations: It can generate 32-second videos with one click, support audio-video synthesis (Text-2-Audio), and support 4D generation.

4) WeChat Releases New Rules, Requiring AI-Generated Content to Be Labeled

According to the official WeChat Coral Security news, WeChat will further regulate the "information source labeling" of self-media messages. WeChat stated that when publishing information generated using AI and other technologies, the content must be labeled as AI-produced; when publishing fictional or interpreted content, the content must be labeled as fictional.

For labeled information, related prompts will be displayed at the bottom of the message to better help users accurately understand the information. For content that does not comply with the information source labeling regulations, the platform will remind users and restrict its spread.

5) Kuaishou's Sora Kailing Open Test, Capable of Generating 2-Minute Videos

Kuaishou AI team's self-developed "Kailing" video generation large model has been launched. The model supports generating high-definition videos with a maximum length of two minutes, 30 frames per second, and 1080P resolution. It is claimed that in terms of the generation quality of 5-second videos, the Kailing model has surpassed Google's video model Voe.

6) ByteDance Launches AI Music Tool Sponge Music, Now in Beta Testing

According to ByteDance's official news, the AI music generation tool Sponge Music is now in beta testing, and the official website is now online. Sponge Music performs excellently in vocal processing, especially in Chinese songs, reducing the use of electronic music and improving articulation clarity and singing fluency.

7) Mianbi Intelligence Announces Free Commercial Use of MiniCPM

Mianbi's MiniCPM is now available for free commercial use. The MiniCPM and MiniCPM-V weights will be fully open to academic research, and enterprises and individuals are also allowed to use them commercially after filling out a questionnaire registration. Community use of the MiniCPM series of models needs to follow the Apache 2.0 and MiniCPM Model Community License Agreement.

8) Zhou Hongyi's Bold Prediction of College Entrance Examination Composition Topics Hits 3 out of 10: Demonstrating 360 AI Search Prediction Capabilities

At the 360AI new product launch conference on June 6, Zhou Hongyi boldly predicted the topics for the 2024 college entrance examination compositions using 360AI search. After the national college entrance examination Chinese test ended this morning, some netizens discovered that among the 10 composition topics predicted by Zhou Hongyi, three of them matched the composition topics of the Beijing, National A, and New Curriculum I papers in this year's college entrance examination.

The composition topic of the National A paper deeply explores the psychological issues of teenagers in interpersonal communication, which coincides with the focus of 360AI search on adolescent mental health in modern life.

The topic "Enduring and Refreshing" in the Beijing paper aligns well with the analysis in 360AI search's topic prediction suggestions regarding the relationship between historical inheritance and contemporary innovation, as well as how to promote social innovation and development while respecting history.

/ 02 / Overseas Developments

1) Local AI Operation, Support for Simplified Chinese, Samsung Announces "Live Translate" Compatibility with Third-party Apps

The Live Translate feature in Samsung's latest Galaxy AI suite will be compatible with third-party apps, meaning users can use AI-based translation functionality in their favorite apps.

2) X's Wallpaper Traffic Boost, Grok 1.5V to Be Released Soon, May Integrate with Midjourney

Insiders have learned that Grok 1.5V is即将发布, possibly integrating with Midjourney to support image generation, allowing users to generate images in Grok and directly publish them to X. This move explains the previous reason why X pushed traffic to wallpaper bloggers.

3) 37 "Unicorn" Enterprises in the Global AI Field: 17 New Additions in the Past Year, 5 in China

According to Nikkei News and research firm CB Insights, as of the end of April this year, there are 37 "unicorn" enterprises in the generative AI field. Compared to 20 companies at the end of April last year, the number of AI "unicorns" has almost doubled in the past year, and the trend of targeting new generation potential technologies is pushing up enterprise valuations. There have been 17 new additions in the past year, with 5 in China.

4) The United States is Expected to Launch an Antitrust Investigation into Microsoft, OpenAI, and NVIDIA

According to insiders, US federal regulators have reached an agreement to allow an antitrust investigation into the dominant positions of Microsoft, OpenAI, and NVIDIA in the AI industry. It is understood that the US Department of Justice and the US Federal Trade Commission reached this agreement in the past week. The US Federal Trade Commission will play a leading role in reviewing the conduct of OpenAI and Microsoft.

5) Siri Co-founder: Apple's Partnership with OpenAI Will Not Last Long

Previously, it was reported that Apple had reached an agreement with OpenAI, a leading generative AI company, to provide generative AI chatbot functionality for Apple's iOS18 operating system. However, Siri co-founder Dag Kittlaus said that such a partnership may only be a stopgap measure and will not last long.