Summary: IBM announced the establishment of the AI Alliance with more than 50 global companies/institutions, including IBM, Meta, AMD, Intel, Oracle, Hugging Face, Sony Group, Stability AI, Cornell University, University of California, Berkeley, University of Tokyo, Yale University, and others. Google released the AI large model Gemini 1.0, which it considers to be the largest and most powerful AI model, outperforming GPT4 in 30 academic benchmark tests. Gemini Ultra achieved a high score of 90.0%, becoming the first model to surpass human experts on MMLU.
Image Source: Generated by Unbounded AI
Financing
French AI startup Mistral AI is raising approximately 4.5 billion euros in a new round of financing, with participation from Nvidia and Salesforce
According to Bloomberg on December 5, sources revealed that the French artificial intelligence startup Mistral AI is in the final stages of raising approximately 4.5 billion euros (about 4.87 billion US dollars) from investors including Nvidia and Salesforce, valuing the company at about 20 billion US dollars. Sources stated that the deal includes over 325 million euros in equity provided by investors led by Andreessen Horowitz, and Nvidia and Salesforce have also agreed to provide an additional 1.2 billion euros in convertible debt. Details are still to be finalized and may be subject to change.
AI software and storage startup Vast Data completes $118 million Series E financing, led by Fidelity
According to TechCrunch on December 7, artificial intelligence software and storage startup Vast Data announced the completion of a $118 million Series E financing, led by Fidelity Ventures, with participation from New Enterprise Associates, BOND Capital, Drive Capital, Nvidia, Dell, Goldman Sachs, Tiger Global, Commonfund, Norwest, 83 North, Greenfield, and Next47.
AI startup AssemblyAI completes $50 million financing, led by Accel
According to TechCrunch on December 5, artificial intelligence startup AssemblyAI completed a $50 million financing round, led by Accel, with participation from former Salesforce Co-CEO Keith Block, former GitHub CEO Nat Friedman, Daniel Gross, Insight Partners, and Y Combinator. AssemblyAI's total financing now amounts to $115 million.
Indian AI startup Sarvam raises $41 million in financing, with participation from Lightspeed and others
According to TechCrunch on December 7, Indian AI startup Sarvam AI, founded 5 months ago, announced that it has raised $41 million in funding through seed and Series A rounds. The seed round was led by Lightspeed and Peak XV Partners, while the Series A round was led by Lightspeed, with participation from Peak XV and Khosla Ventures.
MIT startup Liquid AI raises nearly $40 million in financing to build a new type of liquid neural network AI
According to Zhanzhang Zhi Jia on December 7, the MIT-incubated startup Liquid AI recently successfully completed nearly $40 million in seed round financing, with participation from multiple venture capital firms and well-known angel investors, valuing the company at $303 million. The company aims to build a new type of AI called liquid neural network, which is designed to provide a more streamlined, interpretable, and dynamically adaptive AI system. Liquid AI plans to commercialize this technology, providing private AI infrastructure and platforms for customers, covering areas such as autonomous driving, weather forecasting, and medical data analysis.
Replicate raises $40 million in financing for its open-source AI model repository, led by a16z
According to Forbes on December 6, AI startup Replicate announced that it has raised $40 million in Series B financing, led by Andreessen Horowitz, with participation from Nvidia's venture capital arm NVentures, Heavybit, Sequoia, and Y Combinator. According to sources familiar with the matter, this round of financing values the San Francisco-based startup at $350 million, bringing Replicate's total financing to approximately $58 million.
UK AI startup AutogenAI secures $39.5 million in financing, with participation from Salesforce
According to Bloomberg on December 6, UK AI startup AutogenAI has secured $39.5 million in financing, with participation from Salesforce and other investors. AutogenAI aims to provide AI services for areas such as public contract bidding. Building on ChatGPT and similar services, AutogenAI has developed an automatic writing tool designed to expedite the preparation of bid documents for public and private projects.
Generative AI platform Leonardo.Ai raises $31 million in financing, with participation from Samsung Next
According to TechCrunch on December 6, the Sydney-based generative AI startup Leonardo.Ai announced that it has raised $31 million in financing, with investors including Blackbird, Side Stage Ventures, Smash Capital, TIRTA Ventures, Gaorong Capital, and Samsung Next. The new funds will be used to hire more sales and marketing team members. Additionally, Leonardo.Ai plans to establish its own engineering team.
AI marketing service provider Connectly.ai raises $18 million in Series A financing
According to IT Juzi, marketing service provider Connectly.ai has raised $18 million in Series A financing, led by Volpe Capital, with participation from DST Global and RX Ventures. Connectly.ai focuses on simplifying enterprise users' conversion of customer information into data flows, allowing companies to create and send interactive and personalized marketing campaigns on a large scale through AI-driven mini-bots.
Large Models
Google releases AI large model Gemini 1.0, outperforming GPT4 in 30 academic benchmark tests
AI New Intelligence News, on December 6th, Google officially announced the release of Gemini 1.0, which it considers to be the largest and most powerful AI model capable of processing information in different content forms such as video, audio, and text. It was introduced that Google has released three different versions for different scenarios: Gemini Ultra, the largest and most powerful model suitable for highly complex tasks; Gemini Pro, the best model for various tasks; and Gemini Nano, the most efficient model for edge devices.
In addition, Google evaluated the performance of the two models in various tasks, from natural image, audio, and video understanding to mathematical reasoning. Gemini Ultra has surpassed GPT-4 in 30 out of 32 common academic benchmarks and achieved a high score of 90.0% in the Massive Multitask Language Understanding (MMLU) test, becoming the first model to surpass human experts on MMLU. It is reported that MMLU comprehensively uses 57 subjects including mathematics, physics, history, law, medicine, and ethics to test the ability to understand world knowledge and solve problems.
IBM and Meta join forces with over 50 global companies and institutions to establish the AI Alliance
AI New Intelligence News, on December 5th, IBM announced the establishment of the AI Alliance with over 50 global companies/institutions, including IBM, Meta, AMD, Intel, Oracle, Hugging Face, Sony Group, Stability AI, Cornell University, University of California, Berkeley, University of Tokyo, Yale University, and others.
The alliance will establish member-driven working groups, as well as a management committee and a technical oversight committee dedicated to advancing work and developing overall project standards and guidelines. In addition to bringing together leading developers, scientists, scholars, students, and business leaders in the field of artificial intelligence, the AI Alliance also plans to collaborate with existing important initiatives of governments, non-profit organizations, and civil society organizations.
HuggingFace announces the latest open-source large model leaderboard: Alibaba Cloud TongYi QianWen tops the list
According to Jinshi on December 8th, the world's largest open-source large model community, HuggingFace, recently announced the latest open-source large model leaderboard, with Alibaba Cloud TongYi QianWen topping the list. It is understood that HuggingFace's open-source large model leaderboard (Open LLM Leaderboard) includes hundreds of open-source large models globally, covering six major evaluation dimensions including reading comprehension, logical reasoning, mathematical computation, and fact question answering.
Insiders: OpenAI may release GPT-4.5 by the end of this month
AI New Intelligence News, user @apples_jimmy on the social platform X recently posted that OpenAI may release GPT-4.5 by the end of December, and also suggested that OpenAI's competitor Anthropic may also release multimodal models within this month.
Stability.ai open sources the 3 billion parameter large language model StableLM Zephyr3B
According to AIGC Open Community on December 8th, the well-known open-source generative AI platform stability.ai has open-sourced the 3 billion parameter large language model StableLM Zephyr3B on its official website. Zephyr3B is designed for mobile devices such as smartphones and laptops, featuring small parameters, strong performance, and low computational power consumption, capable of generating text and summarizing content, comparable to models with 7 billion and 13 billion parameters.
Applications
Baidu AI Cloud launches "Super Assistant" AI native application based on Wenxin Yiyuan
According to Cailian News on December 4th, Baidu AI Cloud has launched the AI native application "Super Assistant" based on Wenxin Yiyuan, and has officially started public testing. In the future, Super Assistant will also launch independent PC and mobile applications. In addition, Baidu AI Cloud will continue to release industry-specific versions of Super Assistant.
Meitu launches AI visual large model Miraclevision 4.0 and opens commercial API
AI New Intelligence News, on December 5th, Meitu launched the AI visual large model Miraclevision 4.0. This version features AI design and AI video functions, and will be gradually applied to Meitu's products including Meitu Xiuxiu, Meitu Beauty Camera, Wink, Meitu Design Studio, and WHEE starting from January 2024. In addition, MiracleVision has officially opened its commercial API and offers three tiered packages.
Meta introduces an independent AI image generator, currently free but only supports English prompts
According to IT Home on December 7th, Meta recently launched a new, independent AI image generator called Imagine with Meta, allowing users to create images based on natural language descriptions.
It is reported that the new AI image generator is supported by Meta's existing Emu image generation model and can create high-resolution images based on text prompts. It is currently free for English-speaking users in the United States (whether it will be charged in the future is unknown), and each prompt will generate four images.
Regulation
In addition to the "AI Act," the EU will add additional rules and constraints for higher-capacity models such as GPT-4
According to IT Home on December 10th, the EU has reached a preliminary agreement on the world's first AI regulatory legislation "AI Act," which includes general artificial intelligence systems such as ChatGPT, and is considered a key step in comprehensive AI regulation.
According to a document obtained by Bloomberg, developers of all general artificial intelligence systems (or powerful models likely to be widely used) must meet basic transparency requirements, unless these systems are freely open-source. The requirements include: establishing acceptable usage policies; providing ongoing updates on model training information; summarizing the data used for model training and reporting detailed summaries; and establishing policies that respect copyright laws.
According to the document, models considered to pose "systemic risks" will be subject to additional rules. It is reported that the EU will determine the existence of risks based on the computational power of the training model, with the threshold set at over 100 trillion operations per second. Experts have stated that the only model currently able to meet this threshold is OpenAI's GPT-4.
Large Companies
Meta and Microsoft to purchase AMD's latest AI chip Instinct MI300 X
According to Cailian News, Meta, Microsoft, and OpenAI stated at an AMD investor event that they will use AMD's latest artificial intelligence (AI) chip Instinct MI300 X in the future.
AMD unveiled its first product, the Instinct MI300 X accelerator, which consists of 8 MI300 X GPUs and can provide up to 1.5 TB of HBM3 memory capacity. Compared to Nvidia's H100 HGX, the Instinct MI300 X accelerator has significantly higher throughput and lower latency when running large language model inferences, and also performs better in various AI and HPC projects.
McDonald's and Google to collaborate to introduce generative AI for fresher food by 2024
According to Zhanzhang Zhi Jia on December 7th, McDonald's announced a collaboration with Google to introduce generative artificial intelligence (AI) to optimize the operations of its global stores by 2024. In this collaboration, McDonald's will undergo large-scale hardware and software upgrades, covering multiple systems including ordering kiosks and mobile applications, aiming to provide faster and fresher food.
McDonald's stated that customers can expect to experience more AI-based automation services in the future at more drive-in restaurants through generative AI. However, McDonald's did not provide detailed information on the specific use of AI, only stating that hardware and software upgrades will be carried out, including local upgrades within the stores and services provided through Google Cloud. This updated system will help the restaurant management "quickly identify and take action to reduce business interruptions."
AMD launches AI GPU accelerator MI300 X, training AI models 60% faster than Nvidia's H100
According to IT Home, AMD officially announced its flagship AI GPU accelerator MI300 X at the "Advancing AI" event held at 2 am today, with performance 60% higher than Nvidia's H100.
AMD mentioned that in terms of training performance, the MI300 X is on par with its competitor (H100) and offers competitive price/performance, while excelling in inference workloads. In addition, the MI300X AI accelerator software stack has been upgraded to ROCm 6.0, improving support for generative AI and large language models.
Opinions
Foreign media: Most OpenAI employees "are not really interested in working for Microsoft"
According to Business Insider on December 7th, several current and former OpenAI employees revealed that most employees "are not really interested in working for Microsoft," and the initial agreement to join Microsoft was only to allow Altman to return to the CEO position.
A current OpenAI employee candidly stated that although almost all employees signed an open letter at the time to leave the company with Altman, "no one wanted to go to Microsoft." He also stated that Microsoft is the "largest and slowest" company among all major tech companies. In addition to these factors, compensation also affected the motivation of OpenAI employees. It was reported that the base salary at OpenAI is usually starting at $300,000, and employees can receive stock options worth $2 million within four years, with some employees receiving even higher compensation. Although Microsoft agreed to hire all OpenAI employees at the same compensation level, it was only a verbal agreement at the time and had no legal binding. One employee also believed that Microsoft is unlikely to compensate departing employees for the stock options they would lose when joining the tech giant. In addition, the original plan to offer existing vested stock options to external investors will be canceled.
Zhou Hongyi: Many people have an unrealistic worship of large models, and some even asked him to create a large model for pig farming
According to IT Home on December 10th, at the 2023 China Enterprise Leaders Annual Meeting, Zhou Hongyi, the founder of 360, expressed his views on the recent AI large model craze. He stated that many people have an infinite longing or unrealistic worship of large models, and someone even asked him to create a large model for pig farming. He believes that the technology breakthrough of large models has only been a few years, and there are still many shortcomings.
Research Reports
Hugging Face and research team: Energy consumed by image generation can charge a phone 950 times
According to IT Home on December 5th, a research team composed of Carnegie Mellon University and Hugging Face found that the rise of generative AI not only brings significant changes to various industries but also has a significant impact on the global environment. Researchers studied the carbon footprint of various AI models, including chatbots, paper editors, image classifiers, and image generators, and measured the energy consumption and carbon emissions of these models after performing a given task 1000 times.
The study found that text-based models, such as chatbots and paper editors, are relatively energy-efficient, with the energy required for a thousand chats equivalent to 16% of the energy used to charge a mainstream smartphone. However, image-based models, such as image generators, have the lowest efficiency, with the least efficient image generation model consuming 11.49 kWh per thousand times, equivalent to the energy used to charge a mainstream phone 950 times.
Survey: Only 10% of enterprises adopted generative AI solutions in the past year
According to IT Home on December 6th, a 2023 ML Insider survey conducted by cnvrg.io, a platform focused on AI and large language models (LLM) under Intel, showed that only 10% of enterprise organizations integrated generative AI (GenAI) solutions into production in the past year. The survey also showed that among the enterprises adopting generative AI, about half of them reported improvements in customer experience (58%), efficiency (53%), product capabilities (52%), and significant cost reduction (47%).
Despite the clear benefits, the road to adopting generative AI is not without challenges. The survey showed that 46% of respondents believed that infrastructure is the biggest obstacle to developing LLM into products, with other obstacles including complexity, lack of AI talent, high implementation costs, and compliance issues. 84% of respondents stated that due to the growing interest in LLM technology, they need to improve their skills, with only 19% feeling confident in their understanding of this technology mechanism.
IDC: AI terminals to account for 55% of the Chinese market next year
According to Jinshi News, on December 7th, at the first AI PC Industry Innovation Forum, Lenovo Group and IDC jointly released the first "AI PC Industry (China) White Paper." The white paper pointed out that in the future, everyone will be able to have a personal AI PC running their own "personal large model," and AI PCs will become indispensable personal AI assistants for individuals, families, and businesses.
In addition, IDC released the 2024 China Intelligent Terminal Top Ten Insights Report, predicting that over 70% of terminal devices in the Chinese market in 2024 will be equipped with AI functions, and AI terminals will account for 55% of the market. AI terminals refer to terminal devices with integrated AI engines.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
