Large Model

March 17 news, @ millet technology official microblogging said today, millet big model team in the field of audio reasoning breakthrough progress. Inspired by DeepSeek-R1, the team took the lead in applying reinforcement learning algorithms to multimodal audio comprehension tasks, and in just one week it topped the internationally authoritative MMAU audio comprehension evaluation list with a SOTA accuracy of 64.5%, which is now open sourced at the same time. 1AI's official article reads: Reinforcement Learning Shows "Counterintuitive" Advantage -- Xiaomi's Big Model Team Tops MMAU Audio Reasoning Chart Faced with a video of a car driving...

March 11, OpenAI today announced its latest research, trying to use CoT (chain of thought) monitoring way to monitor the reasoning model's "ideas", so as to stop the AI big model nonsense, hide the real intention and other behaviors, at the same time, this is also one of the effective tools to supervise the super model. Note: AI will occasionally talk nonsense, which we generally call AI hallucinations. In addition to this, AI can be lazy or deliberately bad, such as tricking the user or simply giving up when a problem is too difficult. OpenAI uses the newly released cutting-edge model o3-mini as the monitored...

February 24 news, the current global technology giants power AI, e-commerce platform companies have also joined. According to Lei Feng network news today, Jinduoduo internal has formed a large model team of e-commerce recommendation, the person in charge of the core members of Baidu Phoenix Nest before. According to sources, this person was previously in Baidu when it is already ten million annual salary, and this time Pinduoduo give the sincerity is also very sufficient, its annual package is several times in Baidu. Another source said that Pinduoduo recruited a group of people from Baidu last year to do big model related projects. At present, Pinduoduo big model products are divided into several applications: price comparison system, recommendation, advertising, search...

Feb. 24 - Clement Delangue, co-founder and CEO of Hugging Face, a community for AI big models, announced in a post early this morning that DeepSeek-R1 has received more than 10,000 likes on Hugging Face, making it the most popular big model out of nearly 1.5 million models on the platform. "It's astounding!" Late last month, Hugging Face also announced the Open-R1 project, which plans to fill the DeepSeek R1 un...

Since the Spring Festival, DeepSeek's heat continues to climb, accompanied by a lot of misunderstandings and controversies, some people say that it is "the light of the national product to beat OpenAI", and some people say that it "is just a copy of the foreign model of the work of the small smart". These misunderstandings and controversies are mainly focused on five aspects: 1, excessive myths and brainless disparagement, DeepSeek in the end is not the underlying innovation? Is there any basis for the so-called distillation of ChatGPT? 2, the cost of DeepSeek, really only 5.5 million dollars? ...

It's been about a week since DeepSeek released R1, an open-source inference model that competes with OpenAI o1, generating global attention. Now, Hugging Face, a community of big AI models, has announced the Open-R1 project, which plans to fill in some of the components of DeepSeek's R1 that are not open-source, adding "more fuel to the fire" of this global craze. According to Hugging Face, while the weights used by DeepSeek are known, the datasets and code used to train the models are...

January 27 news, January 23, by the Port of Qingdao, Shandong Port and Shandong Port Technology Group to build the world's first port service class big model -- Ark TaaS (Trade as a Service) big model officially opened for public testing, covering the six core application scenarios. 1AI collated major The main contents are as follows: Scenario 1: intelligent planning of transportation routes and estimation of costs using the big model to help cargo owners and forwarders choose the best transportation solutions and improve logistics efficiency. Scene 2 Smart Way Pilot: Provide personalized business consulting and port area...

On January 25, Liu Qingfeng, chairman of KU Xunfei, made clear at the company's annual meeting the strategic ecological position of Xunfei's Starfire big model: adhering to autonomous and controllable, facing the people's livelihood and immediate needs, facing the needs of national strategy, and empowering scientific research and production. He revealed that in the new stage of the development of general artificial intelligence, Xunfei will be a full-stack autonomous and controllable big model national team, "focusing on autonomous and controllable base capabilities and key application tracks, and maintaining strategic stamina in the spirit of long-termism." 1AI noted that Liu Qingfeng also revealed the 2025 business development strategy: "Strengthen the C-end, deepen the B-end, choose the G-end, product...

January 21, 2011 - Step-2 Star launched two new models of Step-2 series yesterday -- Step-2 mini, which has a smaller number of parameters and is more cost-effective, and Step Literature Master, which is a model specially designed for the content creation field. 1AI learned from the official introduction that Step-2 mini, compared with the trillion-parameter model Step-2, retains its model performance of more than 80% with a parameter count of about 3%. At the same time, Step-2 mini has a faster generation speed and a high price...

According to CCTV News on January 19, the world's first large model of seismic waves with billions of parameters, the "Shimeji" large model, was released in the Chengdu Center of the National Supercomputer in the third stage of testing this month (January 17), and this version has been used for the data processing of the 6.8 magnitude earthquake in Tingri, Tibet, to verify its effectiveness. This version has been used in the data processing of the Tibet Tingri 6.8 magnitude earthquake, which has verified its effectiveness. According to the report, the "Shimeji" seismic wave model will be open for fine-tuning, inference framework and supporting data processing process in 2025, and users can use "Shimeji" directly on the supercomputing platform for business analysis and scientific research. 1AI inquired about the Institute of Geophysics of the China Earthquake Administration (CEA) and learned that ...

January 13 news, according to the official Fangda Special Steel, January 10, Fangda Special Steel Technology Co. (Huawei). 1AI learned from the official press release that the two companies will work closely in the areas of enterprise data governance system and application, application of big models in the steel industry, information security, and digital talent training. In terms of data governance system and application, the two companies will continue to enrich the sharing and application scenarios of enterprise data assets and services, and actively participate in the formulation and promotion of enterprise and industry standards. For the big model in ...

According to a report by Caixin News Agency, recently, a document on the minutes of an expert meeting on DeepSeek's release history and optimization direction has been circulating in the industry. In response, DeepSeek responded that the company has not authorized any personnel to participate in brokerage investor exchanges, and that the so-called "DeepSeek experts" are not company personnel, and the information exchanged is not true. DeepSeek said that the company has strict internal rules and regulations, which explicitly prohibit employees from accepting external interviews, participating in investor exchanges and other types of investor-oriented organizations in the market information exchange meetings. All related matters are subject to public disclosure. 1AI...

At the first edition of the 2024 Ideal AI Talk held on December 25th, Ideal Automobile CEO Li Xiang started a conversation on AI and other topics, and he said that a real big model product must be able to use all the devices autonomously, and will have all the services, which is the real artificial intelligence. 1AI attached Ideal AI Talk interview transcript part of the highlights are as follows: Q: The first time you said internally that you want to be a global leading AI enterprise this sentence, it was in January last year, it was ChatGPT was born just two months...

According to 36 krypton "intelligent emergence" reported today, big model unicorn step leap star has completed the B round of financing, the total financing amount reaches hundreds of millions of dollars. The financing has the participation of state-owned capital, strategic and financial investors, the core investors include Shanghai State-owned Capital Investment Company Limited and its fund, and strategic and financial investors include Tencent Investment, Wuyuan Capital, Qiming Venture Capital and so on. The financing is used to continue to invest in basic model research and development, strengthen multimodal and complex reasoning capabilities, and increase coverage of C-end application scenarios through products and ecology. Jiang Daxin, CEO of StepStar, once said, "Unimodal - multimodal...

At the 2024 China Game Industry Annual Conference, Giant Network announced the "QianYing" audio game generation model, which includes YingGame, a video game generation model, and YingSound, a video dubbing model. YingGame is a video generation model for open-world games, jointly developed by Giant Network's AI Lab and Tsinghua University's SATLab. YingGame is a video generation model for open-world games, jointly developed by Giant's AI Lab and Tsinghua University's SATLab, which is said to be the first time to realize the interactive control of diverse character movements, and can be customized to the game character, and at the same time, has better physical simulation characteristics of the game. Combined with the self-developed video dubbing...

Blogger @i Ice Universe broke the news on December 6 that Samsung Galaxy S25 series will debut the official version of One UI 7, and the difference between the current Beta version of the S24 test is that the S25 series will debut a new wave of AI features, as well as back to the peak of the new Bixby, which is officially connected to the big model. 1AI has learned that Samsung officially announced the One UI 7 Beta on December 5, starting with the Galaxy S24 in Germany, India, South Korea, Poland, the United Kingdom, and the United States....

During China Telecom's "2024 Digital Technology Ecological Conference", the first "TeleAI Developer Conference" was held in Guangzhou. At the conference, TeleAI announced a series of innovative technologies, products and scientific research results, including video generation model, visual model production and application integration platform, embodied intelligence, and smart transmission network, as well as the developer industry alliance program. During the conference, TeleAI announced the release of the China Telecom AI Developer Industry Alliance Program with 19 partners including Huawei, ZTE, and AsiaInfo, aiming to aggregate the strengths of all parties to accelerate the development of the AI industry, and...

According to Huawei's official news, on November 29, the National Pipe Network Group's "Pipe Network" big model was formally put on line for application. Based on Huawei's arithmetic base and technology stack, the "pipe network" model creates a system of "pipe network system and process + data + IT + big model", and builds a big model program with pipe network characteristics. According to reports, through the deployment of the first 200P centralized computing power, the "Pipe Network" big model has been put online for more than 20 application scenarios, with a cumulative number of uses of more than 1.23 million, and has become one of the first in the oil and gas industry to have passed the "dual filing" of industry big models and algorithms by the relevant state departments.... It has also become one of the first oil and gas industries to pass the "dual filing" of industry large models and algorithms by the relevant state departments...

CCTV released a blog post yesterday (November 29), reporting that the first pumped storage large model independently developed by Southern Power Grid was officially put into use, marking the realization of the depth of intelligent management of nearly one-fifth of the nation's installed capacity of pumped storage equipment, which will significantly enhance the grid's peak-peaking and energy storage capacity, and safeguard the country's energy security. 1AI quoted the press release as saying that the pumped storage grand model, which is embedded in China's third-generation pumped storage AI data analysis platform, is like an experienced expert with knowledge of all pumped storage unit equipment maintenance, and that it understands the technicians' commands and automatically collects the set...

Dark Side of the Moon Kimi and MADSys Lab at Tsinghua University 2024 jointly released a design for the Mooncake inference system underlying Kimi. The system is based on the KVCache-centered PD separation and store-for-store architecture, which improves the inference throughput. Recently, in order to further accelerate the application and promotion of this technology framework, Kimi, the Dark Side of the Moon, and MADSys Lab at Tsinghua University, together with 9#AISoft, AliCloud, Huawei Storage, Noodle Intelligence, and Convergence Technology, have released an open source project ...

OpenAI and Meta will begin training AI macromodels in African languages to address the shortage of models for the continent's thousands of dialects. The project, which also involves French telecom operator Orange SA, will launch in the first half of next year, 1AI understands. Initially, the company will focus on two West African languages, Wolof and Pulaar, which are spoken by 22 million people in the region. Orange SA eventually hopes to expand the program to more AI technology companies, building large-scale language models to help it engage with 18 markets in the Middle East and Africa...

November 20, 2011 - LatePost reported yesterday evening that one of the "Big Model Six" companies, Zero One Everything, is planning to spin off an AI application company, which will use the Zero One Everything Big Model as the basis for building games and other applications. The company is in the midst of a multi-million-dollar round of fundraising. The company is raising funds at a valuation of tens of millions of dollars. Note: ZeroOne, Baichuan Intelligence, Smart Spectrum AI, Dark Side of the Moon, Minimax, and Step Star are known as the "Six Little Dragons of Big Models" in the industry. According to the report, the new company after the spin-off is called "Oasis", and is headed by Ma Jie, co-founder and vice president of ZeroOne...

According to today's news from Lei Feng network, millet last week to carry out a series of personnel changes, in which the basic technology platform department set up AI platform department, Zhang Duo (duó) as the head of the AI platform department, to the general manager of the basic technology platform department Feng Honghua report, organizationally the basic technology platform department belongs to the millet technology committee. Zhang Duo served in Xiaomi between 2016 and 2021, and was publicly stated by Lei Jun on microblogging: Zhang Duo is a big god of Xiaomi. On September 10 this year, Zhang Duo said on social media platforms to return to Xiaomi. The report also mentioned that a number of Xiaomi employees confirmed that Xiaomi Group Vice President Ma Ji has not...

Xiaomi official today announced that the big model small love music ability upgrade, multi-end device support "music Q&A" function. According to the introduction, the function is based on the upgrade of AI ability, supporting the song information to ask questions, search for professional music knowledge and other content. Users can experience the new function on the cell phone and car end after upgrading the Xiaoxi classmate big model to the latest version. Xiaomi Xiao Ai classmates ushered in a big version update at the end of July this year, upgrading the "big model Xiao Ai" in full, supporting natural question and answer, picture editing, and waking up defense outside the car, covering cell phones, tablets, televisions, speakers, cars, and other core categories of devices. In this year's 10 ...