Mini Program: Daily summary of artificial intelligence dynamics.
1. OpenAI and Meta will train large models using African languages to benefit more populations.
OpenAI and Meta will begin training large AI models using African languages to address the shortage of models for thousands of dialects on the African continent. This project also involves the French telecom operator Orange SA, which will start in the first half of next year. Initially, the company will focus on two West African languages: Wolof and Pulaar, which are used by 22 million people in the region. Orange SA ultimately hopes to expand the project to more AI technology companies, building large language models to help them communicate with customers in 18 markets in the Middle East and Africa. Africa has about one-third of the world's languages, and this scale, combined with limited funding and incomplete infrastructure, means that few of these dialects are included in AI training models. Orange will use local language models to interact with customers and plans to offer them for free to public health and education services, as well as local companies and merchants. Orange's Chief AI Officer Steve Jarrett stated that next year, the project will target other sub-Saharan African languages such as Lingala, Swahili, and Bambara.
2. Trump reportedly considers establishing a new position in the White House, specifically to coordinate AI affairs.
According to media reports citing insiders, the elected President Trump is considering establishing a new position in the White House to coordinate artificial intelligence (AI) affairs. Insiders say that the AI commissioner may be established, but it has not yet been finalized, and the world's richest man Elon Musk will not take this position. Nevertheless, Musk is still expected to play a key role in future discussions and practical applications of AI.
3. Shusheng·Puyuy large model is fully opened, InternThinker strong reasoning model breaks through complex tasks.
On the evening of November 25, the Shanghai Artificial Intelligence Laboratory opened the Shusheng·Puyuy large model to social users and released the strong reasoning model InternThinker on the model interface. It is reported that the InternThinker model has long thinking capabilities and can reflect and correct during the reasoning process, achieving better results in complex reasoning tasks such as mathematics, code, and reasoning puzzles.
4. Xiaopeng Motors announces VR remote operation robot patent.
Tianyancha's intellectual property information shows that recently, Guangzhou Xiaopeng Automobile Technology Co., Ltd. has applied for a patent for 'a robot remote operation system, method, and device based on VR headsets'. The abstract shows that this invention relates to the field of robot remote operation technology, and the system includes motion capture equipment, VR headsets, upper computers, and robots. The motion capture equipment collects the pose of one or more body parts of the wearer and sends it to the VR headset; the VR headset filters the pose in the motion capture signal and sends the filtered poses and VR headset poses to the upper computer; the upper computer processes the poses and VR headset poses according to the robot control algorithm to generate at least one control signal and sends it to the robot, which receives at least one control signal to perform corresponding movements, increasing the redundant degrees of freedom used to control the robotic arm, effectively solving the problem of the robotic arm losing control in singular positions and improving the control accuracy over the robot.
5. Philips and Amazon expand strategic cooperation to promote HealthSuite cloud service development.
Amazon stated that Philips and Amazon Web Services (AWS) are expanding their strategic cooperation to promote the development of HealthSuite cloud services, aiding generative artificial intelligence workflows.
6. China Nonferrous Metals will acquire a Brazilian tin mining company, as artificial intelligence boosts demand prospects.
China Nonferrous Metals Group's subsidiary agrees to acquire equity in a Brazilian tin producer. For this metal used in electronic products, the development of artificial intelligence has boosted its demand prospects. Minsur SA, headquartered in Lima, Peru, announced on Tuesday that the trading company under China Nonferrous will acquire equity in Mineração Taboca indirectly held by Minsur for $340 million, provided that certain conditions are met. Taboca, located in the Pitinga mining area of the Brazilian Amazon, is one of the richest tin ore reserves in the world. This deal will also allow China Nonferrous to acquire a processing plant near São Paulo, as well as supply niobium and tantalum.
7. AI company DeepGlint's general manager Zhao Yong resigns: third-quarter losses doubled year-on-year.
AI company DeepGlint announced on November 26 that it recently received a written resignation letter from Chairman and General Manager Zhao Yong, who applied to resign from the position of General Manager due to the company's business development needs. His resignation will take effect from the date it is delivered to the board of directors. The announcement stated that after resigning as General Manager, Zhao Yong will still serve as Chairman and other positions, and will be a core technical staff and Chief Scientist of the company. As of the date of this announcement, Zhao Yong indirectly holds 45,174,904 shares of the company, accounting for 17.44% of the total share capital of the company. On November 25, 2024, DeepGlint held the eighth meeting of the second board of directors, and it was agreed to appoint Ms. Wu Yizhou as the general manager of the company, with a term from the date of board approval until the end of the second board term; it was also agreed to appoint Mr. Zhang Qiang as the deputy general manager of the company, with a term from the date of board approval until the end of the second board term.
8. IBM Granite 3.0 enterprise AI model announces compatibility with Amazon's latest Trainium 2 chip, now available on the AWS platform.
As a prelude to the Amazon AWS re:Invent technology conference to be held next week, Amazon AWS has announced that it will further deepen its cooperation with IBM in the AI field. IBM's Granite 3.0 enterprise AI model will land on the Amazon cloud platform, and it will also further adapt to Amazon's latest Trainium 2 chip. According to reports, the Amazon Trainium 2 chip is built for training AI models, with a computing power of 65 exaflops, claiming to be able to train foundational models and large language models in a very short time. In terms of integration and collaboration on the cloud platform, IBM's AI model lifecycle management tool watsonx.governance will further integrate with Amazon SageMaker starting in December, allowing users to directly operate SageMaker through the watsonx.governance management interface, while the information from the SageMaker model registry can also be shared with watsonx, thereby achieving transparency in AI operation management. In addition, IBM's Guardium AI Security tool is now available in the Amazon AWS Marketplace. This tool focuses on managing security risks of sensitive AI data and AI models, capable of detecting 'shadow AI' (AI models created by employees privately), and can integrate with watsonx.governance to ensure AI security.
9. Anthropic announces the Model Context Protocol open-source agreement, utilizing a unified interface to 'unify AI databases'.
On November 25, Claude AI developer Anthropic announced an open-source agreement called Model Context Protocol, which claims to provide a standardized and unified interface for AI tools and model databases in the industry, allowing major AI tools to access various data sources through a single protocol, thus accelerating model response speed and generation quality. Anthropic stated that the industry has invested a lot of resources in model functionality development and has made rapid progress in improving reasoning capabilities and quality. However, even the most advanced models are still limited by data isolation issues. Each new data source requires a customized integration plan, making truly interconnected systems difficult to scale. This Model Context Protocol is said to be aimed at 'unifying AI model interfaces'.
10. Endpoint intelligence speech is about to officially launch: optimized based on Intel Core Ultra chip, providing local processing 'privacy mode'.
The Intel New Quality Productivity Technology Ecological Conference opened today in Chengdu. Zhizhi stated that based on optimization of the Intel Core Ultra chip, the previously network-dependent Zhizhi Speech now has offline operation capabilities. The official stated that the endpoint Zhizhi Speech jointly created with Intel is about to officially launch, with the following capabilities: in offline mode, endpoint Zhizhi Speech can handle AI dialogues, analyze long documents, answer knowledge base questions, and other complex tasks; it provides a local processing 'privacy mode' option to ensure the safety of user data and personal document information; in combination with cloud capabilities, endpoint Zhizhi Speech supports building intelligent agents, AI image recognition and drawing, and efficient AI search capabilities.
11. Kimi Math version launched: based on the dark side of the moon k0-math model, claimed to match the capabilities of OpenAI o1.
The dark side of the moon released the new generation mathematical reasoning model k0-math on November 16. The official announcement today states that the Kimi Math version based on this model is now online. In the Kimi web version, select the 'glasses' icon in the sidebar to use the Kimi Math version based on the k0-math model, and the official also stated that a mobile version will be launched later. The official recommends using LaTeX format for mathematical formulas, which can be screenshot or photographed and sent to the regular version of Kimi, allowing Kimi to convert the image to LaTeX format, then copy the question to the Kimi Math version.
12. Xuan Ya International: OrangeGPT AI video remix tool has been developed to enhance video creation capabilities.
Xuan Ya International stated on the interaction platform that the company has created OrangeGPT around application scenarios related to its main business to meet different scenario creation needs for marketing. The company's R&D team has developed AI intelligent creation tools such as 'OrangeGPT AI video remix tool' and 'OrangeGPT AI poster tool' based on business application scenario needs, significantly improving the video creation capabilities and image generation effects of OrangeGPT.
13. Kunlun Wanwei: The Aijiekexin AI high-performance chip project is progressing smoothly and has not yet achieved mass production.
An investor asked Kunlun Wanwei about news stating that Kunlun Wanwei's holdings in Aijiekexin have started small-scale chip production. Is this news true? Kunlun Wanwei stated on the interaction platform that the AI high-performance chip project under Aijiekexin is progressing smoothly. As of the current reply, Aijiekexin has not yet achieved mass production.
Article forwarded from: Jinshi Data.