Large Model

11 月 20 日消息，据晚点 LatePost 昨日晚间报道，“大模型六小龙”公司之一 —— 零一万物近期正计划分拆出一个 AI 应用公司，将以零一万物大模型为基础搭建游戏等方向的应用。该公司正以数千万美元的估值对外融资。 注：零一万物、百川智能、智谱 AI、月之暗面、Minimax 和阶跃星辰被行业称为“大模型六小龙”。 报道称，分拆后的新公司名为“绿洲”，由零一万物联合创始人、副总裁马杰担任…

据雷峰网今日消息，小米上周进行一系列人事变动，其中基础技术平台部成立 AI 平台部，张铎（duó）担任 AI 平台部负责人，向基础技术平台部总经理冯宏华汇报，组织上基础技术平台部属于小米技术委员会。 张铎在 2016 年~2021 年间任职小米，曾被雷军在微博公开表示：张铎是小米的大神。今年 9 月 10 日，张铎在社交平台上表示重新回归小米。 报道还提到，多位小米员工确认，小米集团副总裁马骥已不…

小米官方今日发文宣布，大模型小爱音乐能力升级，多端设备支持「音乐问答」功能。 据介绍，该功能基于 AI 能力升级，支持歌曲信息追问、搜索专业音乐知识等内容。用户将小爱同学大模型升级至最新版本后，可在手机、车载端体验新功能。 小米小爱同学在今年 7 月末迎来大版本更新，全量升级「大模型小爱」，支持自然问答、图片编辑、车外唤醒防御等功能，覆盖手机、平板、电视、音箱、汽车等核心品类设备。 在今年 10 …

Tencent announced the launch of the Hunyuan-Large model, which is the largest Transformer-based MoE model that has been open sourced in the industry, with 389 billion total parameters (389B) and 52 billion activation parameters (52B). Tencent has open sourced Hunyuan-A52B-Pretrain, Hunyuan-A52B-Instruct, and Hunyuan-A52B-Instruct-FP8 at Hugging Face, and released...

Quark announced a comprehensive upgrade of its "AI Search" product, which allows users to search for all kinds of new questions and problems through cell phones, tablets, computers and other devices, and the AI will explain the topics step by step and support users to ask questions at any time. According to reports, Quark AI can also provide in-depth explanations around the knowledge points and test points in the questions, accurately locate the key steps, and enable users to not only learn the difficult questions, but also learn a class of questions by example with the experience of "heuristic" learning. At the same time, Quark released the "Spiritual Knowledge" learning model. The official said that in the latest evaluation, Quark's "Spiritual Knowledge" Learning Model has a positive performance on the math problems of the exam...

Baichuan big model public number today announced that Baichuan Intelligence launched a one-stop big model commercialization solution, including the whole link universal training data, Baichuan4-Turbo, Baichuan4-Air two models and the whole link domain enhancement tool chain, which can help enterprises to achieve low-cost private deployment, and achieve the availability of multiple scenarios of up to 96%. It is reported that the program has the characteristics of "rich tools, fast response, significant effect, low cost", based on Baichuan4-Turbo, Baichuan4-Air, and combined with ...

Recently, there are rumors in the AI circle, in the overall financing difficulties, valuation and constantly push up the background, some big model startups become difficult, in and out of the valley, which is highly regarded by everyone's attention "six tigers" in a two tigers, and even have given up the basic big model research and development, turned to the AI application. Because, if not transformed, I'm afraid that the "six little tigers" are going to be cool. If the tigers really do this, to some extent means that they began to take the initiative to go down the AGI table, from the grand narrative of the stars and the sea, to the reality of the grounded tea, rice, oil and salt. No way, or that old saying, the situation is stronger than people, the current environment, make money...

On October 24th, at the 2024 Tech Data Xunfei Global 1024 Developer Festival, Tech Data Xunfei officially released "Xunfei Starfire 4.0 Turbo". Xunfei Starfire 4.0 Turbo is said to have ushered in a brand-new upgrade, with seven major capabilities surpassing GPT-4 Turbo, and its mathematical and code capabilities exceeding GPT-4o. In terms of mathematical capabilities, Xunfei Starfire 4.0 Turbo has completed the verification of algorithms for ultra-long thought chain, tree search and self-reflective evaluation, and it is expected that it can realize the difficult mathematical capabilities of class o1 by the end of this year. From ...

From the "Henan release" was informed that on October 17, the first domestic application of clinical ophthalmology big model - VisionFM ("Fuxi Hui-eye") released in Zhengzhou. Wang Ningli, president of the Henan Provincial Academy of Medical Sciences, "Fuxi Hui-eye" in layman's terms is both a cutting-edge platform, but also an artificial intelligence application. It has a strong "brain", aggregated from 26 countries and regions of 560,000 individuals of 8 modalities, a total of 3.4 million eye images, is currently the world's largest ophthalmic image database. The large model was developed by Prof. Wang Ningli's team in collaboration with more than 30...

Byte Jump released a factual clarification announcement about interns destroying big model training". Byte Jumping big model training was attacked by interns, and the person involved has been dismissed The announcement said that recently the media said that Byte Jumping big model training was attacked by interns", and after the company's internal verification, there was indeed a serious disciplinary violation of the interns of the commercialization technology team, and the interns have been dismissed. Byte Jumping said that the relevant reports also exist in part exaggerated and inaccurate information, the special explanation is as follows: First, the intern involved maliciously interfered with the model training task of the commercialization technology team's research project, but it did not affect the commercialization of the official project and the online business, and also...

According to Ali International AI team AIDC-AI, Marco, the latest translation big model developed by Ali International, was officially launched today, supporting 15 global mainstream languages such as Chinese, English, Japanese, Korean, Spanish and French. Officially, based on Flores, a widely used open source evaluation dataset, the Marco translation big model is ahead of the benchmark translation products on the market, such as Google Translate, DeepL, GPT-4, etc. in terms of BLEU automatic evaluation indexes. "Your baby is on the way" will no longer be translated as "Your baby ...

On October 16th, Yi-Lightning, the AI company founded by Kai-Fu Lee, released its new flagship model. On the blind test list LMSYS, Yi-Lightning surpassed OpenAI, GPT-4o-2024-05-13 released by Anthropic, and Claude 3.5 Sonnet, and was ranked sixth in the world and first in China. No. 1 in China. According to Zero One Everything, this is the first time that a Chinese big model has realized "surpassing the O...

From the official website of China Mobile, we learned that at the main forum of the 12th China Mobile Global Partner Conference, China Mobile released the JiuTian ShanZhi Multi-Modal Base Model and launched 30 self-developed industry models. China Mobile's introduction to the JiuTian ShanZhi multimodal base model is as follows: The JiuTian ShanZhi multimodal base model released is the latest R&D achievement of the JiuTian series of general-purpose models, and the model's four major functions have been significantly improved: first, intelligent parsing of long text, supporting text comprehension of over 100,000 words, automatically analyzing and extracting key information from multiple documents, accurately parsing the core viewpoints behind the information, and ultimately realizing the graphics and text. ...

As AI technology continues to evolve, a number of domain-related concepts and acronyms have been appearing in various articles, such as Prompt Engineering, Agent Intelligence, Knowledge Bases, Vector Databases, RAGs, and Knowledge Graphs, etc., but it is also true that these technologies and concepts play a crucial role in the development of the big AI models. These technological elements collaborate with each other in diverse forms to drive AI technology forward. In this article, we will delve deeper into the high level models that are always paired with big models including, but not limited to, the above...

October 6 news, Guoxing Aerospace today announced the completion of the satellite in-orbit operation of AI big model technology verification, the space adaptability of Guoxing Aerospace AI big model in-orbit operation, the reliability of the arithmetic satellite platform and the high-performance arithmetic load in-orbit effective arithmetic verification. Time Node: At 10:31 a.m. on September 24, "One Arrow, Three Stars" of Guoxing Aerospace was successfully launched in Haiyang, Shandong Province. At 20:46 on September 25, the test team uploaded the AI task to the satellite in the form of remote control instruction through the ground station; at 21:00 on September 25, the test team uploaded the AI task to the satellite in the form of remote control instruction through the ground station; at 21:00 on September 25, the test team uploaded the AI task to the satellite in the form of remote control instruction.

JetBrains officially released JetBrains Al Assistant based on AliCloud's Tongyi Big Model at the 2024 Cloud Conference this week. JetBrains Al Assistant is deeply integrated with several JetBrains products, and is a customized "productivity tool" for Chinese developers, according to JetBrains' official introduction. "JetBrains Al Assistant is deeply integrated with a number of JetBrains products and is customized for Chinese developers. JetBrains Al Assistant supports full-link optimization development, and can be used in specific IDE tool windows as an internal...

At the 2024 Cloud Root Conference, AliCloud CTO Zhou Jingren released Qwen2.5, a new generation of open source models, of which the flagship model, Qwen2.5-72B, is claimed to outperform the Llama 405B. Qwen2.5 covers multiple sizes of large language models, multimodal models, mathematical models, and code models, and each size has a base version, a command-following version, quantized versions, totaling more than 100 models on the shelf. Qwen2.5 language models: 0.5B, 1.5B, 3B, 7B, 14B, 32B, and ...

Shanghai Internet Information Office released the "Shanghai Generative Artificial Intelligence Service Filing Information Announcement" on September 9, adding 7 new generative artificial intelligence services that have completed the filing, and a total of 41 generative artificial intelligence services have been completed for the record. From the list of the latest filed generative AI services, we can see that the services of Mihayou's subsidiary, Shanghai Mihayou Mystery Technology Company Limited, have passed the filing, and the model name is "Glossa". Cai Hao Yu, the founder of Miha Tour, once said that "AIGC (Artificial Intelligence Generated Content) will revolutionize the game development industry", and he believes that the future...

The World Digital Technology Academy (WDTA) published a blog post on its official account, announcing that it officially released the international standard "Large Model Supply Chain Security Requirements" at the Bund Conference. The standard was jointly compiled by experts from the Cloud Security Alliance (CSA) Greater China and dozens of companies including Ant Group, Microsoft, Google, Baidu, NIST, Meta, PrivateAI, etc. This is the industry's first international standard for large model supply chain security, marking another solid step forward in international cooperation in global AI governance. Li Yuhang, Chairman of the Cloud Security Alliance (CSA) Greater China, said that the operation of large model systems depends on...

On September 5, Tencent announced the launch of a new generation of large models "Hunyuan Turbo". Compared with the previous generation model, Tencent Hunyuan Turbo has significantly improved performance, with training efficiency increased by 108%, inference efficiency increased by 100%, inference cost reduced by 50%, and decoding speed increased by 20%. The effect is comparable to GPT-4o in multiple benchmarks, and third-party evaluation ranks first in China. At the Tencent Global Digital Ecosystem Conference, Qiu Yuepeng, Vice President of Tencent, COO of the Cloud and Smart Industry Business Group, and President of Tencent Cloud, announced that Tencent Hunyuan Turbo was launched on Tencent Cloud, and the input and output prices are only half of the previous model...

Recently, OpenAI executives are planning a jaw-dropping subscription plan - the upcoming new large models "Strawberry" and "Orion" will cost up to $2,000 per month, 100 times higher than the current $20 for ChatGPT Plus! This sky-high price is undoubtedly eye-popping. There are two main reasons why OpenAI dares to set such a price: one is the shortage of funds, which requires a lot of funds to support the huge computing power demand; the other is the extraordinary strength of the new product, coupled with the existing hundreds of millions of user base, Ope...

According to CCTV News and Heilongjiang Daily, on August 30, the largest single-cluster intelligent computing center of domestic operators, China Mobile Intelligent Computing Center (Harbin), was completed and officially put into operation. The intelligent computing center has the characteristics of the largest single-cluster computing power and the largest domestic network equipment networking. The single-cluster intelligent computing center builds all AI accelerator cards into one cluster to support the training of large models with trillion-level parameters. According to reports, China Mobile Intelligent Computing Center (Harbin) has achieved a localization rate of 100% for AI chips, and for the first time passed domestic network equipment...

Zhipu AI recently released its latest large-scale model GLM-4-Plus, which demonstrates powerful visual capabilities comparable to OpenAI GPT-4, and announced that it will be open for use on August 30. Major update highlights: Language base model GLM-4-Plus: A qualitative leap in language parsing, instruction execution, and long text processing capabilities, continuing to maintain a leading position in international competition. CogView-3-Plus: The performance of the text model is comparable to the industry's top MJ-V6 and FLUX models. Image/video understanding model GLM-4V-Plus...

According to Alibaba Cloud today, at the 2024 Big Data Expo on August 29, the Institute of Geochemistry of the Chinese Academy of Sciences and Alibaba Cloud jointly released the world's first "Lunar Science Multimodal Professional Big Model". The big model is built based on Alibaba Cloud's Tongyi series models. At present, the accuracy rate of lunar impact crater age and morphology has reached more than 80%. According to Alibaba Cloud, the lunar professional big model is based on the Tongyi series models such as vision, multimodality and natural language, combined with RAG retrieval enhancement and other technologies, and fine-tuned and trained in Alibaba Cloud Bailian exclusive version. At present, the lunar professional big model...