1. Qwen2.5 updates with million-token long context and a 4.3 times increase in inference speed; netizens say RAG is about to become obsolete.

The recent update of the Qwen2.5-Turbo model has made significant waves in the AI community, showcasing its ability to handle a staggering 1 million tokens of long context while achieving an impressive inference speed increase of up to 4.3 times. This new model not only allows for rapid summaries of lengthy texts but also outperforms competitors in various benchmarks, highlighting its enhanced efficiency and affordability at just 0.3 yuan per million tokens. As users express excitement over its capabilities, Qwen is positioning itself as a formidable player in the field of large models.

Original Chinese article: https://www.qbitai.com/2024/11/220926.html

English translation via free online service: https://translate.google.com/translate?hl=en&sl=zh-CN&tl=en&u=https%3A%2F%2Fwww.qbitai.com%2F2024%2F11%2F220926.html

2. Peking University team released the multi-modal model LLaVA-o1, whose reasoning capabilities are comparable to GPT-o1!

Peking University researchers have released LLaVA-o1, the first visual language model capable of spontaneous systematic reasoning, achieving performance comparable to GPT-4v. The 11B-parameter model introduces a novel “slow thinking” mechanism with a four-stage reasoning process, outperforming competitors like Gemini-1.5-pro and GPT-4v-mini across six multimodal benchmarks. Built on Llama-3.2-Vision, the open-source model demonstrates an 8.9% improvement over its base model through its structured reasoning approach.

Original Chinese article: https://www.aibase.com/zh/news/13322

English translation via free online service: https://translate.google.com/translate?hl=en&sl=zh-CN&tl=en&u=https://www.aibase.com/zh/news/13322

3. ByteDance’s PixelDance and Seaweed video generation models launched on Jimeng AI, offering free public access.

ByteDance has launched its video generation models PixelDance and Seaweed on the Jimeng AI platform, making them available for free public use. These tools allow users to create videos using AI technology, further expanding ByteDance’s footprint in the AI video generation space.

Original Chinese article: https://www.jiqizhixin.com/articles/2024-11-18-2

English translation via free online service: https://translate.google.com/translate?hl=en&sl=zh-CN&tl=en&u=https://www.jiqizhixin.com/articles/2024-11-18-2

4. Xianyu: Over 600 million registered users, AI technology has been applied to idle transaction scenarios.

At the 29th UN Climate Conference, Alibaba Xianyu’s CTO Chen Jufeng announced that the platform has surpassed 600 million registered users while introducing its innovative AI model aimed at enhancing the user experience in secondhand transactions. The Xianyu AI will optimize the entire trading process, focusing on trust through advanced features like credit scoring, smart listing, and an AI model for virtual fitting. Looking ahead, Xianyu plans to deepen its integration with AI technology to provide users with personalized AI advisors for a seamless trading experience.

Original Chinese article: https://news.qq.com/rain/a/20241118A05M6V00

English translation via free online service: https://translate.google.com/translate?hl=en&sl=zh-CN&tl=en&u=https%3A%2F%2Fnews.qq.com%2Frain%2Fa%2F20241118A05M6V00

5. Precision and generality cannot be achieved simultaneously; Peking University and Huawei’s theoretical proof shows that scaling laws are difficult to achieve under low precision.

A joint study by Peking University and Huawei has proven that achieving scaling laws at low precision in large language models is infeasible. Quantization, frequently used to reduce the computational costs of model inference, limits performance gains when using lower precision levels like int8 or int4. This research highlights the trade-off between model precision and benefits from scaling laws, challenging assumptions that reducing precision consistently increases efficiency.

Original Chinese article: https://www.jiqizhixin.com/articles/2024-11-18-10

English translation via free online service: https://translate.google.com/translate?hl=en&sl=zh-CN&tl=en&u=https://www.jiqizhixin.com/articles/2024-11-18-10

6. UNITX secured $46 million in Series B funding to power smart manufacturing with AI industrial vision large models.

UNITX has raised $46 million in Series B funding to further develop its AI-powered industrial vision solutions. These models focus on visual inspection in smart manufacturing and are already integrated into supply chains for sectors like automotive parts and lithium batteries. This funding will expedite their technology development.

Original Chinese article: https://36kr.com/p/3040205464482051

English translation via free online service: https://translate.google.com/translate?hl=en&sl=zh-CN&tl=en&u=https://36kr.com/p/3040205464482051

7. SEELE AI completed a pre-A round of financing in the tens of millions of US dollars and will release the world’s first AI game generation platform.

SEELE AI has secured tens of millions of dollars in Pre-A funding to support the launch of the world’s first AI game generation platform. The company aims to revolutionize the game development industry with this platform, although specifics about its technology and features have not yet been disclosed. This significant investment underscores investor confidence in SEELE AI’s innovative approach.

Original Chinese article: https://www.163.com/dy/article/JHA7BOKQ0511B8LM.html?spss=dy_author

English translation via free online service: https://translate.google.com/translate?hl=en&sl=zh-CN&tl=en&u=https://www.163.com/dy/article/JHA7BOKQ0511B8LM.html?spss=dy_author

8. Daimon Robotics Secures Two Consecutive Angel+ Rounds Totaling ¥100M, Focusing on Optical Tactile Sensors and Embodied AI Systems.

Daimon Robotics, focused on developing optical tactile sensors and embodied AI systems, has successfully completed two rounds of Angel+ financing, each totaling over ¥ 100 million. These sensors have applications in humanoid and surgical robots, as well as automotive sectors like smart cockpits.

Original Chinese article: https://36kr.com/p/2990834468563720

English translation via free online service: https://translate.google.com/translate?hl=en&sl=zh-CN&tl=en&u=https://36kr.com/p/2990834468563720

9. The Smart Interaction Innovation Competition has concluded, and the Harbin Institute of Technology’s AI Smart Recitation Assistant has won the Grand Prize | OPPO Intelligent Body Platform.

The Intelligent Interaction Innovation Competition, co-hosted by OPPO and Zhejiang University, concluded with Harbin Institute of Technology’s AI Recitation Assistant winning the grand prize. The competition focused on advancing human-machine interaction technologies and showcased innovative applications of AI, demonstrating significant potential in enhancing educational efficiency and cultural preservation.

Original Chinese article: https://www.qbitai.com/2024/11/220690.html

English translation via free online service: https://translate.google.com/translate?hl=en&sl=zh-CN&tl=en&u=https://www.qbitai.com/2024/11/220690.html

—————————————————————————————————————————————————
That’s all for today’s China AI Native Industry Insights. Join us at AI Native Foundation Membership Dashboard for the latest insights on AI Native, or follow our linkedin account at AI Native Foundation and our twitter account at AINativeF.

The post China AI Native Industry Insights – 20241119 – Alibaba | Peking University | ByteDance | Huawei | more appeared first on AI Native Foundation.