Big Language Information:
Google AI Bard is fully upgraded: image analysis, support for more languages, landing in more regions
On Thursday, Google updated its artificial intelligence (AI) chatbot Bard. Google announced that it will open the Bard service to more users in the European Union and Brazil, regions home to hundreds of millions of people, in the biggest expansion of the product since its launch in February.
Bard has been added with more language capabilities and has image analysis capabilities, where users can upload images and ask Bard to provide information about the image or ask it to make a caption based on the image.
Huawei Releases New AI Large Model Storage Products to Support models with 10 billion parameter scales
Huawei today released two new AI large model storage products, OceanStor A310 and FusionCube A3000. Among them, OceanStor A310 is a deep learning data lake storage for basic and industry large models, which supports data weaving and near-memory computing. FusionCube A3000 is a training/push-hyperfusion machine for industry large model training and inference, which can support models with tens of billions of parameter scales and support models with higher parameter scales through horizontal scaling.
Users complain that GPT-4 has become "lazy" and "stupid" in the industry: probably related to OpenAI's rearchitecture
According to the Science and Technology Innovation Board Daily, some doubts have begun to appear on the Internet recently, many users complain that OpenAI's GPT-4 has become less powerful, and some people say that the model has become "lazier" and "dumber" compared to its previous reasoning ability and other outputs.
Some in the industry believe that this may be related to OpenAI's major redesign of the system, speculating that OpenAI uses hybrid expert (MoE) models to make generative model output higher-quality, less costly, and more responsive.
Chongqing's first self-developed large language model "Zhaoyan" was released
On July 13, the "Zhaoyan" big model release conference of Chongqing Institute of Artificial Intelligence of Shanghai Jiao Tong University was held in the Western (Chongqing) Science City. At the press conference, the latest research and development results of the institute - "Zhaoyan" large model products were released.
This is the first self-developed large-language model in Chongqing, and it is also the first release of results since the institute landed in Science City. "Zhaoyan" can present accurate and professional text output for users such as academic research, business analysis, and life consulting.
Leyard's virtual moving point has reached AI cooperation with JD Technology, focusing on "spatial computing" and "AI large model" and other technologies
According to the news of the virtual moving point official Weibo on July 14, on July 13, Leyard's virtual moving point reached a strategic cooperation with Jingdong Technology.
With JD.com, the two parties will give full play to their respective advantages on the basis of existing cooperation, continue to develop and upgrade their business breadth, focus on core technologies such as "spatial computing", "digital human" and "AI big model", continue to make efforts in film and television, education, industry, games, sports and other fields, empower thousands of industries with science and technology, and promote virtual moving points to move forward in the AI era. According to the data, virtual moving point is empowered by AI to build an "integrated three core" development strategy with spatial computing as the strategic main body and AI large model, hardware equipment and spatial data as the three cores.
Giant Network and Alibaba Cloud reached a comprehensive cooperation of "game + AI"
Giant Network and Alibaba Cloud announced the "Game + AI" comprehensive cooperation, and the two sides officially signed a memorandum of cooperation in Shanghai.
According to the memorandum of cooperation, Alibaba Cloud and Giant Network will jointly build a "game + AI" intelligent computing platform. Based on the game R&D platform of Giant Network, Alibaba Cloud Tongyi Qianwen model and AIGC product capabilities, the two parties have carried out comprehensive cooperation to create AI scene landing projects in the game field.
The co-founder of Stability AI sued the company's current CEO
The co-founder of artificial intelligence company Stability AI sued the company and its boss on July 13, claiming he was tricked into selling a stake in the $1 billion startup for just $100.
Cyrus Holtz co-founded Stability in 2020 with current CEO Emad Mostark. Host's lawsuit alleges that Mostark convinced him that "the company he helped build was essentially worthless" and then bought all 15 percent of his shares.
Shinye Technology launched the first generation of LLM (Large Language Model) application development framework E-LADF
Recently, Shinye Technology launched the first generation of LLM (Large Language Model) application development framework E-LADF.
Based on basic components such as ChatGPT-like large model engine and vector database, E-LADF aims to provide a series of functional interfaces to help developers build and deploy LLM-based applications. At present, E-LADF includes four core interface functions: local knowledge base management, general scenario streaming dialogue, local knowledge base based question answering, and long text entity extraction and summary.
Cube Shuke launched the AI training platform Cube PAI
According to the Securities Times, recently, the Cubic Cloud PAI (hereinafter referred to as "Cubic PAI") independently developed by CubeShuke has been reviewed by the National Copyright Administration and successfully obtained the "Computer Software Copyright Registration Certificate".
Cubic PAI provides a complete AI model training and resource management capability platform, which is easy to scale, supports on-premise, on-cloud, and hybrid environments of all sizes, and provides many AI user-friendly features to make it easier for end users and administrators to complete daily AI tasks.
New Oriental's subsidiary added AI software development business
Tianyancha App shows that recently, Beijing Layout Future Education Technology Co., Ltd. has undergone industrial and commercial changes, and its business scope has added artificial intelligence application software development, computer software and hardware and auxiliary equipment wholesale, residential interior decoration, construction engineering design, etc.
The company was established in July 2016 with a registered capital of 550 million yuan, the legal representative and manager is Chai Mingyi, the chairman is Yu Minhong, and it is wholly owned by New Oriental Education Technology Group Co., Ltd.
The demand for AI talent in U.S. businesses has surged
According to the latest data from Adzuna, a global job search platform, the demand for AI talent in the US technology industry is rising with the rapid development of the artificial intelligence (AI) industry, and many of these jobs can easily reach six figures in annual salaries.
Adzuna data shows that there were 7.6 million open positions in the United States in June, and more of them require skills in the field of artificial intelligence, and the number of AI-related job openings has now risen to 169045, of which 3,575 positions specifically indicate the need for generative AI skills.
Microsoft executive: Instead of dividing traffic, Bing chat boosts publishers' traffic
Jordi Ribias, Microsoft's vice president of project management at Bing, recently spoke to Fortune about the impact of Bing Chat bots and website traffic, and talked about the future of Bing Search and more.
Ribas emphasized that instead of diverting traffic to content publishers, Ribas emphasized that Bing Chat drove traffic.
The U.S. Federal Trade Commission launched a full investigation into OpenAI
According to market news, the US Federal Trade Commission (FTC) recently launched a comprehensive investigation of OpenAI. It is reported that the investigation is because its product ChatGPT may have violated consumer protection laws.
The Federal Trade Commission asked OpenAI to explain how it addresses the risks posed by its AI models, and the investigation poses a significant regulatory threat to OpenAI's business in the United States. The FTC specifically requires OpenAI to provide a detailed description of any complaints about its products making false, misleading, defamatory, or harmful representations to individuals.
Key Papers:
Alibaba DAMO Academy launched PolyLM, an open-source big language model
Alibaba DAMO Academy recently released a paper announcing an open source multilingual big language model PolyLM.
PolyLM demonstrates remarkable ability to understand, reason, and generate natural language instructions, and the model is available in 1.7B and 13B sizes. The researchers also propose a multilingual self-directed method that can automatically generate 132.7K diverse multilingual instructions for model fine-tuning.
Paper address: Click here
Google announced HyperDreamBooth, which can generate multiple styles of face images
Google Research recently released a paper announcing a hypernetwork product called HyperDreamBooth.
HyperDreamBooth is able to generate face images with a variety of backgrounds and styles from a single image, while maintaining the details of the picture. According to the research team, HyperDreamBooth can model faces in about 20 seconds, 25 times faster than DreamBooth and 125 times faster than Stable Diffusion's Textual Inversion.
Paper address: Click here