1. The representative of the "two sessions" party members and the big model of the AI model landing
2. Anthropic the strongest model release performance is fully surpassed GPT-4
3. AWS announced that the Claude 3 model access
4. "The Strongest Wensheng Tu Model" research papers are publishedIndore Investment
5. Openai Google Microsoft META co -signed AI open letter
6Kanpur Wealth Management. ChatGPT adds new text read aloud function
7Kanpur Stock. 3D generating model triposr open source 1 second to turn a single picture to 3D model
8, 53 pdf exposes Openai’s Q*model promotion timetable
9. Mac’s exclusive big model framework is online
10. Nvidia’s market value Super Saudi Ame is the third largest listed company in the world
11. Puchen Technology Open Source SORA Reapping Architecture Plan 46%
12. It is rumored that Ali led the new round of financing minimax
13. Lianhui Technology completed a new round of hundreds of millions of yuan in strategic financing
14. AI Creative State Baseston received $ 40 million in Series B financing
15. Xindan Intelligent Complete Ten Thousand Level Angel Ring financing
16. Deng Jiang, vice president of former Zhongguancun Department, joined Baichuan Smart
17. Tencent Cloud AI series of product promotion AI painting 19.9 yuan and 1,000 times
18. Ali push AI e -commerce tool "painting frog"
19. Huawei Yunpan Gu Da model passes standard compliance verification
20. Tom Cat: Promoting the application to access the Sora model
21. Shi Zhi AI Wisemodel Open Source Community launched a half -year transcript
22New Delhi Investment. Cooperative AI industry highland in Haidian District and Tsinghua University
23. Tsinghua will build 100 AI empowerment teaching pilot courses
24. JPMorgan Chase AI Auxiliary Cash Flow model can reduce artificial work by 90%
25. Google founder appeared as a Gemini image generating problem
26. OPENAI developer Relations Leader left
27. AI tax reporting robot flip: the test questions are wrong, half
28. AMD is hindered by selling AI chips to China
29. India restricted generation formula AI tool release
1. The representative of the "two sessions" party members and the big model of the AI model landing
According to the Beijing Business Daily, at the National Two National Sessions this year, the representative members positively made suggestions on topics such as the trend, application, and supervision of large AI models.The technological industry representatives that have proposed related suggestions are Yang Jie, a member of the National Committee of the Chinese People’s Political Consultative Conference, the secretary of the Indian mobile party group, and the chairman of the chairman of the National People’s Congress, Lei Jun, a representative of the National People’s Congress, Chairman of Xiaomi Group, Zhou Hongyi, member of the National Committee of the Chinese People’s Political Consultative Conference, the founder of 360 Group, deputies of the National People’s Congress, and a director of HKUST XunfeiHe Liu Qingfeng, member of the National Committee of the Chinese People’s Political Consultative Conference, deputy general manager of Tianyu Digital Section, He Yi, member of the National Committee of the Chinese People’s Political Consultative Conference, founder and CEO of Zhihu and CEO.
2. Anthropic the strongest model release performance is fully surpassed GPT-4
Anthropic, a great enemy of OpenAI, announced last night that the Claude 3 model family, including Claude 3 Haiku, Sonnet, Opus, all supported more than 1 million tokens inputs, added more modular functions, and supported processing images and charts.Among them, OPUS has the strongest performance, and is better than GPT-4 in the common evaluation benchmark of most AI systems; Sonnet reaches a balance between intelligence and speed, which is especially suitable for enterprise workloads. The Haiku model is the fastest and most cost-effective.Read the arxiv (~ 10K tokens) with charts and graphic information and data dense research papers in less than 3 seconds.Both Opus and Sonnet are available, and Haiku will soon be released.
3. AWS announced that the Claude 3 model access
Amazon Cloud Technology announced today that the Claude 3 series model dedicated to AI security and research company Anthropic will provide services based on Amazon Bedrolk.Customers will access the Claude 3 Opus, Sonnet and Haiku through Amazon Bedrolk’s three -supported visual features.Among them, the Sonnet model is now available in East (North Virginia) and West (Oregon) regions. The OPUS models and Haiku models will be launched in the next few weeks.
4. "The Strongest Wensheng Tu Model" research papers are published
Stability AI released a research papers of the "Strongest Wensheng Model" Stable Diffusion 3 today to describe technical details.This paper will soon be open on Arxiv.After assessment, Stable Diffusion 3 is equal to or better than the current most advanced Wenshengtu system.The multi -mode diffusion transformer (MMDIT) architecture is an independent weight set for images and languages. Compared with the previous version of the Stable Diffusion 3, it improves text understanding and spelling ability.
5. Openai Google Microsoft META co -signed AI open letter
SV Angel issued an AI public letter last night, calling on "everyone to establish, deploy and use AI to improve people’s lives."OPENAI, META, Google, Microsoft, Salesforce, Hugging Face, Mistral AI, DataBricks, Scale AI, Character.ai, Elevenlabs and other institutions have signed."AI", "There is a common responsibility to make a choice to maximize the interests of AI and reduce risks."
6. ChatGPT adds new text read aloud function
OpenAI announced today that ChatGPT’s "reading" function can be read, which can read the content of the conversation text.The operation method on iOS or Android systems is to click and hold the message, and then click the "Read Aloud" button.Openai is also scrolling at the web. Users can click the "Read Aloud" button under the message.
7. 3D generating model triposr open source 1 second to turn a single picture to 3D model
STABILITY AI announced today that it cooperates with the domestic 3D generation model, the VAST, which is open source for a fast 3D object reconstruction model Triposr, which can generate high -quality 3D models from a single image within 1 second.TRIPOSR runs under low reasoning budget, which is completely available for users without GPUs.When testing at Nvidia A100, it generates 3D output (texture grid) with draft quality in about 0.5 seconds, which is better than other open images to 3D models, such as OpenLRM.Under the MIT license, the Triposr model weight and source code can be downloaded, allowing business, individuals and research and use.
8, 53 pdf exposes Openai’s Q*model promotion timetable
In the early morning of yesterday, a netizen named Jackson on the social platform X released a 53 -page PDF file, called "revealing Openai’s plan to create AGI by 2027".According to this file, OpenAI began training a multi -mode model of 125 trillion parameters in August 2022.The first stage is Arrakis, also called Q*.The model was trained in December 2023, but it was canceled due to the high cost of reasoning. This is the GPT-5 originally planned to be released in 2025.GOBI (GPT-4.5) was renamed GPT-5 because the original GPT-5 was canceled.The next stage of Q*was originally GPT-6, but it was later renamed GPT-7. It was originally planned to be released in 2026, but it was put on hold for recent Musk’s lawsuit.Q* 2025 (GPT-8) is planned to be released in 2027 to achieve a comprehensive AGI.I do n’t know how high the reliability is, and Openai has not responded to this.
9. Mac’s exclusive big model framework is online
According to social platform X and GitHub, the Apple Computer Deficiency Model Framework Chat with MLX created by an employee of an OPENAI employee has recently launched.This framework supports 11 languages such as Chinese, English, French, etc., which can easily integrate any open source models compatible with HuggingFace and MLX. As long as the two lines of code can complete the deployment, support Apple Computer to run a local model.
Github address:
10. Nvidia’s market value Super Saudi Ame is the third largest listed company in the worldPune Investment
According to Bloomberg, Nvidia has a market value of Saudi Arabia on Monday, becoming the third largest listed company in the world after Apple and Microsoft.As of the closing of the Eastern Time on Monday, Nvidia’s total market value was US $ 2.13 trillion.
11. Puchen Technology Open Source SORA Reapping Architecture Plan 46%
宣 潞 潞 潞 announced yesterday that its already open source Sora reproduction architecture OPEN-SORA.It is said that this architecture can also reduce the cost of reappearing by 46%and expand the length of the model training input sequence to 819k Patches.At present, Open-Sora has covered: provides a complete Sora reproduction architecture, supports dynamic resolution, supports a variety of model structures, supports multiple video compression methods, and supports multiple parallel training optimizations.
Github address:
12. It is rumored that Ali led the new round of financing minimax
According to the Venture Capital Daily report yesterday, MINIMAX, the starting project of the GM model, is conducting a new round of large -scale financing, and Ali is the core investment party.Not long before, another large model of the company created the dark side of the moon also reported the news of completing $ 1 billion in financing. Most of the funds were said to come from Ali.
13. Lianhui Technology completed a new round of hundreds of millions of yuan in strategic financing
Lianhui Technology announced on February 28 that it has completed a new round of hundreds of millions of yuan in strategic financing. Investors have moved and invested by the Indian Mobile Industry Chain Development Fund, the Qianhai Fund of Qianhai Fangzhou and Qilu Qianhai FundComposition of head state -owned assets and market -oriented institutions.Lianhui Technology was established in 2003. Focusing on multi -modal data analysis, visual semantics understanding, pre -training large models, AIGC and other AI technologies, that is, industrial landing. At presentSmart and other products.This round of financing will be used for the technology research and development, product innovation and market expansion of multi -mode and large models and independent intelligence, and further expand the leading advantages of major B industries such as operators, energy power, and media.
14. AI Creative State Baseston received $ 40 million in Series B financing
The American AI founder Baseten announced today that it has received $ 40 million in Series B financing, led by IVP and Spark Capital. Greylock, Conviction Partners, and angel investor Lachy Groom participated in the investment, with a valuation of more than 200 million US dollars.Founded in 2019, the main business is to help customers adopt training open source or custom AI models and put them into public applications.
Customers can deploy their models on the Baseten cloud without having to set back -end infrastructure.Baseten’s software automatically adjusts the number of computing resources paid by the company’s payment according to the traffic of the AI model within a certain time.If customers need GPUs, it can also help get scarce GPUs.
15. Xindan Intelligent Complete Ten Thousand Level Angel Ring financing
According to the investment community, recently, the Shenzhen AI startup Xindan Intelligent completed the angel round financing of millions of levels, which was jointly invested by the global artificial intelligence enterprise APUS and the senior investor of the AI industry.Xindan Intelligence was founded in 2023. It consists of Tsinghua, Berkeley, Tencent, Meta and other academic and engineering elites. Since its establishment, Xindan Intelligent has used 7B and 30B models to reach TOP1 on the global recognized list MT-Bench, which has achieved it, which has achieved it.SCORE: 8.58, second only to GPT4.
16. Deng Jiang, vice president of former Zhongguancun Department, joined Baichuan Smart
According to the "Science and Technology Board Daily" reported on March 4, Deng Jiang, vice president of former Zhongguancun Science and Technology, joined the domestic head of large models, Baichuan Smart, as a business partner as the president of the financial business group.And the establishment of cooperative ecology.Deng Jiang has been engaged in the research and development of core business systems, product design and team management at the Headquarters of the Indian Agricultural Bank of China and the CITIC Bank head office.He joined Zhongguancun Kejin in 2022. He served as the establishment of the financial industry line, and successfully delivered nearly a hundred acceptance projects.
17. Tencent Cloud AI series of product promotion AI painting 19.9 yuan and 1,000 times
Tencent Cloud announced yesterday that it launched a series of AI product promotional activities, including voice recognition, voice synthesis, AI painting, digital wise man, face core and other fields.In the first single area, the new user purchased a minimum of 0.4 % off, including AI painting services 19.9 yuan and 1,000 times, a face fusion of 14.9 yuan and 1,000 times, and a recording file recognition of 14.9 yuan and 60 hours.The face of the face was 50 % off, the voice recognition was 40 % off, AI painting was 50 % off, and 50 % off the audience.
18. Ali push AI e -commerce tool "painting frog"
According to the TECH Planet yesterday, Ali established an AI e -commerce team, which is currently creating a "painting frog" for merchants and experts.It is reported that the main body of the painting frog development is Hangzhou Lianfan Information Technology Co., Ltd., which is 100%held by Alibaba (India) Co., Ltd."Painting Frog" will improve creative efficiency for Taobao, Tmall merchants and experts in terms of generating marketing documents, training exclusive AI models, etc., and will be quickly "planting grass".
19. Huawei Yunpan Gu Da model passes standard compliance verification
According to Huawei Indian officials, in yesterday, in the trusted AI model standards organized by the Indian CITIC Institute, the Huawei Yunpan Gu Gu model successfully completed the financial model standard compliance verification and obtained excellent levels (4+ level)Rating, this is also the highest level among the manufacturers of this large financial model evaluation.Huawei Yunpan Gu Da Model has become the first batch of products that have passed the standard compliance verification of the financial model standard. It is also the first product to pass the industry’s large model standard compliance verification.
20. Tom Cat: Promoting the application to access the Sora model
Tom Cat answered investors on the interactive platform on March 3 and asked: "The company is promoting the Sora model of applying for an OpenAI and plans to conduct the test of related material production. At present, Sora is not yet officially accessible."
21. Shi Zhi AI Wisemodel Open Source Community launched a half -year transcript
Shi Zhi AI Wisemodel.cn Open Source Community Platform announced last night that the number of half -year transcripts since the official announcement of official announcement on September 4, 2023: The number of open source models of Wisemodel reached 390, and the number of open source data sets reached 58.At the same time, the Wisemodel community officially launched code and computing power modules. The open source project code in the AI field can also be released simultaneously in the Wisemodel community.
The Wisemodel community also launched an open source reappearing Sora plan on February 19 this year. At present, more than 40 research members from domestic and foreign universities such as Tsinghua, Peking University, Shanghai University of Science and Technology, Hong Kong University of Science and Technology, and Tokyo UniversityResearch members of the entrepreneurial team joined.The duplication plan team is divided into model groups, data groups, evaluation groups, and INFRA groups, which have carried out the corresponding parts of research and verification.
22. Cooperative AI industry highland in Haidian District and Tsinghua University
On the morning of March 2nd, the signing ceremony of the strategic cooperation between Haidian District and Tsinghua University to jointly build the artificial intelligence industry highland.According to the agreement, the two parties will cooperate around the five major aspects of building an artificial intelligence innovation platform, the development of key core technologies, the transformation of major scientific and technological achievements, the construction of an artificial intelligence talent zone, and the planting artificial intelligence industry ecology to form new productivity.
The cooperation between the two parties to build an artificial intelligence industry highland is of great significance for schools to improve artificial intelligence scientific and technological innovation and achievement transformation, serve Haidian to promote the upgrading of the intelligent industry and economic development, and help Beijing to accelerate the construction of artificial intelligence innovation with global influence.
23. Tsinghua will build 100 AI empowerment teaching pilot courses
According to the public account of Tsinghua University, on the afternoon of February 29, Tsinghua University held a 2024 school faculty and staff conference and central school party member centralized training. The principal Li Luming made a school work report, focusing on the school’s exploration and layout of the school’s deep integration of education and teaching.It means that the school will build 100 AI empowerment teaching pilot courses and equip each 2024 freshman with "AI growth assistant".
24. JPMorgan Chase AI Auxiliary Cash Flow model can reduce artificial work by 90%
According to Bloomberg report yesterday, Morgan Chase Group, one of the largest financial service agencies in the United States, has helped some corporate customers reduce labor in classification and visualization payment process through its AI -based cash flow management tools.Decisions still need to be made by humans.Since its launch about a year ago, the company now has about 2,500 customers using the product.Morgan Chase CEO Jamie Dimon said that this technology can eventually shorten the weekly working hours to 3.5 days.
25. Google founder appeared as a Gemini image generating problem
According to social platform X, Google co -founder Sergey Brin appeared at the AI hacker marathon venue on the weekend.test.Sheri gave a nearly 40 -minute speech on topics such as Gemini development inside story and AGI. When asked about the view of the issue of Gemini painting in Google’s big model, he said that "there is indeed we have not fully understood the model in the model."Part" and apologized.
26. OPENAI developer Relations Leader left
OPENAI developer relationship head, Logan Kilpatrick, posted on March 1st local time on social platform X, announced that he had left Openai and he had just passed the last day of OpenAI.Logan had previously led the relationship between the developer of OpenAI and supported developers to use Dall-E, API and ChatGPT for development.In addition to OpenAI, he is a chief developer community in Julia’s programming language and a teacher of the CSCI E-33A of Harvard University Expanded School Curriculum.Logan was previously Apple’s application machine learning engineer and software engineer, as well as community managers in Julia’s programming language.
27. AI tax reporting robot flip: the test questions are wrong, half
According to the Washington Post, after testing, the AI Tax Assist of Turbotax and H & R Block is unreliable. He answered half of the wrong or error information in 16 test questions.These performances release a dangerous signal. For products with vital accuracy, you must not trust the answer given by the AI tool.
28. AMD is hindered by selling AI chips to China
According to Bloomberg report today, according to people familiar with the matter, AMDs were trying to sell a custom AI chip to the Indian market, and they encountered obstacles from the US government.AMD hopes to obtain the approval of the US Department of Commerce to sell this AI chip to Indian customers, and says that the performance of this chip is lower than the performance of chips sold outside AMDs outside India, and the design meets US export restrictions.However, US government officials informed AMD that the performance of the chip is still too strong. AMD must be permitted by the Industrial and Security Agency of the US Department of Commerce in order to sell the chip.
29. India restricted generation formula AI tool release
According to Reuters today, the Indian Information Technology Department stated in a consultation report issued to various platforms last Friday that the use of tools, including the generation AI, and its "availability on the Internet on the Internet, must be obtained by the availability of users.The Indian government’s clear permission ".India requires technology companies to seek permission before public release of "unreliable" or the AI tools that are being tried, and pointed out that they should be labeled because these tools may return error answers to user inquiries.
Varanasi Stock