The first official "large model standard compliance evaluation" results in China have been released, and the large model products of four enterprises including Alibaba Cloud, Baidu, 360, and Tencent have passed the evaluation. The epic upgrade of Midjourney V6 is stunning!
Image Source: Generated by Unlimited AI
Financing
AIGC's live broadcast large model company "Lingxi Deep Wisdom" completes a 50 million yuan angel round of financing
According to a report by Jijin on December 22, AIGC's live broadcast large model company "Lingxi Deep Wisdom" completed an angel round of financing, with investors including Quwan Technology, Cheetah Mobile, and Zero2IPO, totaling 50 million yuan. The new financing will be used for technical research and development, business expansion, and team building.
Founder of Niu Technologies starts a business in the AI vertical model field, plans to reach a user scale of 1 million within a year
According to Interface News on December 22, Hu Yilin, the founder of Niu Technologies, has started a new company "Shiyanjia," which integrates AI technology into the watch industry and has completed two rounds of financing, with a total financing amount of 50 million yuan and a post-investment valuation of 300 million yuan.
Legal AI startup Harvey secures $80 million in Series B financing, with a valuation of $7.15 billion
According to the webmaster's home on December 22, the legal artificial intelligence startup Harvey announced that it has secured $80 million in Series B financing, with a valuation of $7.15 billion, from investors including Sequoia Capital, Redpoint Ventures, and OpenAI Startup Fund.
AI company rabbit secures tens of millions of dollars in financing
According to Geek Park, AI company rabbit has recently secured tens of millions of dollars in financing, with the latest investment coming from American venture capitalist Vinod Khosla. The total amount of the three rounds of financing is $30 million. Rabbit is a company that develops the next-generation operating system based on Large Action Model (LAM).
AI company AutoAgents.ai completes tens of millions of yuan in angel round financing, led by Innovation Works
According to IT Orange on December 25, AI company AutoAgents.ai recently completed tens of millions of yuan in angel round financing, led by Innovation Works, with Qingshui City Seven Xi Investment participating. The funds from this round of financing will mainly be used for product development, market expansion, and team expansion. AutoAgents.ai is committed to providing autonomous intelligent agents (AI Agents) and intelligent assistant (Copilot) software services for enterprises in multiple countries and regions worldwide to improve work efficiency.
Large Models
The first official large model evaluation results in China are released, with four large models including Tongyi Qianwen and Tencent Huan Yuan passing the evaluation
According to a report by New Beijing Shell Finance, on December 22, it was learned from informed sources that the first official "large model standard compliance evaluation" results in China have been released. The large model products of four enterprises including Alibaba Cloud, Baidu, 360, and Tencent have passed the evaluation, representing that they have met the national standards in terms of universality and intelligence. According to public information, the large models owned by the four enterprises are Tongyi Qianwen, Wenxin Yiyuan, 360 Zhinao, and Huan Yuan large model, with Tongyi Qianwen being the only open-source model.
The "large model standard compliance evaluation" was initiated by the China Electronics Technology Standardization Institute, aiming to establish a Chinese large model standard compliance list to lead the healthy and orderly development of the artificial intelligence industry. The evaluation collected opinions from dozens of leading units in the academic and industrial sectors, covering 38 specific evaluation dimensions for assessing the universality and intelligence of language large models, and is an authoritative evaluation based on the official large model testing benchmark.
Zhipu AI open sources the visual language model CogAgent, supporting GUI graphical interface Q&A
According to the webmaster's home on December 21, Zhipu AI has open sourced CogAgent, a visual language model with 180 billion parameters. The model has shown outstanding performance in GUI (graphical user interface) understanding and navigation, achieving SOTA general performance in multiple benchmark tests. In addition, it supports high-resolution visual input and dialogue Q&A, and can answer questions based on any GUI screenshot.
Zhiyuan Research Institute releases a 37 billion parameter multimodal large model Emu2
AI New Intelligence World News, on December 21, the Beijing Zhiyuan Research Institute announced the release of Emu2, a multimodal large model with 37 billion parameters.
Emu2 significantly surpasses mainstream multimodal pre-training large models such as Flamingo-80B and IDEFICS-80B in tasks including few-shot multimodal understanding, visual question answering, and subject-driven image generation, achieving optimal performance in tasks including VQAv2, OKVQA, MSVD, MM-Vet, and TouchStone.
Emu2 demonstrates strong multimodal context learning capabilities, and can even solve tasks requiring real-time reasoning, such as visual prompts and object-based generation. Emu2-Chat, fine-tuned based on Emu2, can accurately understand graphic and textual instructions, achieving better information perception, intent understanding, and decision planning. Emu2-Gen can accept sequences of images, text, and locations interleaved as input, achieving flexible, controllable, and high-quality image and video generation. The research team also stated that Emu2 can serve as a basic model and general interface for various multimodal tasks.
Meta releases a new AI translation large model, achieving real-time voice conversion in under 2 seconds
According to IT Home on December 25, Apple collaborated with researchers from Columbia University in October 2023 to release an open-source multimodal LLM named Ferret, which did not attract much attention at the time. Therefore, many people in the artificial intelligence community missed the release of Ferret.
Bart de Witte, who operates a European non-profit organization focused on open-source artificial intelligence in the medical field, recently posted on X: "I somehow missed this, Apple joined the open-source artificial intelligence community in October. The launch of Ferret proves Apple's commitment to influential artificial intelligence research and consolidates its position as a leader in the field of multimodal artificial intelligence…ps: I look forward to the day when local large language models (LLLMs) run as integrated services on my redesigned iOS on my iPhone."
According to a report by Pionex on December 22, Meta has recently released a series of AI translation large models, achieving real-time voice conversion with a delay of no more than 2 seconds. These models support multiple language translations and have the ability to mimic intonation, speech rate, emotions, and other characteristics. This series of models is called Seamless Communication, including SeamlessExpressive, SeamlessStreaming, SeamlessM4T v2, and Seamless, with the first three already open-sourced on GitHub.
To ensure translation accuracy and prevent abuse, Meta has adopted toxicity mitigation technology to filter "toxic content" before training and automatically detect and adjust the generation of toxic words during the translation process. Additionally, watermarks are added to the audio to track the source. To mitigate abuse risks, Meta has also added watermarks to the audio, embedding imperceptible signals in the audio to accurately trace its source and counter various attack methods.
Applications
Midjourney opens alpha version testing for the V6 model
AI New Intelligence World News reported on December 21 that Midjourney announced the opening of alpha version testing for the V6 model in the Discord community.
Stable Diffusion introduces commercial paid subscription plans
AI New Intelligence World News reported that Stability AI recently announced the launch of a membership subscription plan for its text-to-image model Stable Diffusion. The non-commercial level membership allows free use of the core model for personal and research purposes. The professional version is priced at $20 per month and is suitable for creators, developers, and startups. The enterprise version is mainly tailored for large enterprises and can be customized for large-scale operations, with pricing based on the customization.
Nubia announces that the Z60 Ultra phone is equipped with the industry's first vertical image AI large model
According to IT Home on December 19, at the ongoing Nubia new product launch event, Nubia Z60 Ultra announced that it is equipped with the industry's first vertical image AI large model. The trained AI large model enhances Nubia's imaging system. Since its establishment in 2012, Nubia has accumulated a large amount of professional image data in more than 30 image fields, enabling customized training of AI for exclusive image scenes. In various aspects such as starry sky and humanities, Nubia deeply integrates AI technology to create the industry's first vertical image AI large model.
Xiaohongshu internally tests AI chatbot "Davinic"
According to Tech Planet on December 25, Xiaohongshu has internally tested an AI feature called "Davinic" in its main app. This feature, which has been in testing since September, provides intelligent Q&A and is more inclined to provide Q&A related to good living, covering travel guides, food guides, geographical and cultural knowledge, life skills, personal growth and psychological advice, as well as event recommendations in multiple aspects. "Davinic" is trained based on Meta's LLAMA large model.
Major Companies
Microsoft integrates AI music creation platform Suno into Copilot, allowing music generation through text
AI New Intelligence World News reported on December 20 that Microsoft announced on its official website a collaboration with the AI music creation platform Suno and its integration into Copilot, allowing users to generate various types of music through text.
Google: AI coding access provided to all levels of Colab users in 175 countries/regions
AI New Intelligence World News reported on December 20 that Google announced that AI coding access has been provided to all levels of Colab users in 175 countries/regions. Colab was initially built by a small team at Google Research and now has over 10 million monthly active users, including millions of students worldwide, making it Google's largest AI coding tool.
Samsung and Naver showcase the latest AI chip, with about 8 times higher efficiency than Nvidia's chip
Citing Businesskorea, the Science and Technology Board Daily reported that Samsung Electronics and Naver showcased a jointly developed artificial intelligence (AI) semiconductor over the past year. The product's efficiency is about 8 times higher than that of competitors' chips like Nvidia's, and it is expected to support Naver's large-scale AI model HyperCLOVA X.
Meta releases a new AI translation large model, achieving real-time voice conversion in under 2 seconds
According to a report by Pionex on December 22, Meta has recently released a series of AI translation large models, achieving real-time voice conversion with a delay of no more than 2 seconds. These models support multiple language translations and have the ability to mimic intonation, speech rate, emotions, and other characteristics. This series of models is called Seamless Communication, including SeamlessExpressive, SeamlessStreaming, SeamlessM4T v2, and Seamless, with the first three already open-sourced on GitHub.
To ensure translation accuracy and prevent abuse, Meta has adopted toxicity mitigation technology to filter "toxic content" before training and automatically detect and adjust the generation of toxic words during the translation process. Additionally, watermarks are added to the audio to track the source. To mitigate abuse risks, Meta has also added watermarks to the audio, embedding imperceptible signals in the audio to accurately trace its source and counter various attack methods.
Perspectives
LinkedIn Vice President: In the AI era, the value of education has "significantly diminished"
According to Business Insider on December 19, Aneesh Raman, Vice President of LinkedIn, recently stated in a podcast that in the era of generative artificial intelligence, having a bachelor's degree from an Ivy League school may no longer be the key to success in one's career, as the value of education has "significantly diminished." Adaptability is considered a key soft skill today, and using AI in the workplace can not only help employees improve efficiency but also facilitate more effective cross-cultural, cross-lingual, and cross-departmental communication, as well as enhance empathy.
Amazon Founder Bezos: ChatGPT is not an "invention" but a "discovery"
According to iFanr on December 21, Amazon founder Bezos shared his insights on generative AI on the popular tech podcast Lex Fridman Podcast.
Regarding generative AI such as ChatGPT, Bezos has proposed an interesting definition: "Today's large language models are not inventions, they are discoveries." In Bezos' view, only something deliberately designed and clearly understood in its operation is an invention. For example, a telescope is an invention, but seeing Jupiter through the telescope and knowing it has its own satellites is a discovery. Large language models are more like discoveries. We are often amazed by their capabilities. They are not products of deliberate design.
As for the potential harm of AI to human survival, Bezos has shown an optimistic attitude: "We humans have many ways to bring about our own destruction. These technologies may help us avoid doing these things, and may even save us."
AI Professor Dou Dejing: Domestic large models reach GPT-3.5 level, narrowing the gap with GPT-4 technology
According to Global Times, on December 23, Dou Dejing, a renowned expert in artificial intelligence and big data and a part-time professor in the Department of Electronic Engineering at Tsinghua University, stated that domestic large models have currently reached the level of GPT-3.5, with a certain gap from GPT-4, but the gap is narrowing. Professor Dou, who has been engaged in AI research for over 20 years, mentioned that the term "artificial intelligence" was coined in 1956. In the past, the technology community expected to achieve general artificial intelligence around 2050, but the emergence of generative artificial intelligence and large models has greatly accelerated this process. It is now expected that this goal will be achieved within 5 to 10 years.
People's Daily Commentary: AI Customer Service Driving People Crazy? Technology is good, but not omnipotent
AI New Intelligence World News reported on December 22 that People's Daily published a commentary titled "AI Customer Service Driving People Crazy? Technology is good, but not omnipotent." The article stated that a recent media report titled "AI Customer Service is Driving People Crazy" has attracted attention and once again sparked widespread ridicule and complaints from netizens about AI customer service. The benefits of AI customer service are well known: for consumers, it provides 24-hour availability, instant response, and efficient handling of procedural transactions. However, at present, AI customer service has not yet reached the level where it can completely replace human customer service. In particular, in services that require more emotional value, AI often not only fails to play a role but may also have a counterproductive effect. "I feel like being served by AI customer service gives a sense of being ignored." This is also the sentiment of many consumers. It is evident that while AI is good, it is not omnipotent, and in its application, it is not a one-size-fits-all solution. Businesses need to adapt to different scenarios and be flexible.
Cheetah Mobile Chairman Fu Sheng: 2024 will be the wave year for the application of AI large models
According to Interface News on December 22, Fu Sheng, Chairman and CEO of Cheetah Mobile and Chairman of Cheetah Lab, stated at the 2023 Exploration Conference that 2023 is the year of AI and the wave year. In the first half of the year, everyone was investing in billion-scale large models. 2024 will definitely be the wave year for the application of AI large models, and many applications that have never appeared in the past, like Didi and Meituan in the era of mobile intelligence, will emerge.
Fu Sheng stated that as an entrepreneurial enterprise, it is important to realize that using general large models cannot completely solve the company's own problems. The core competitiveness of a startup company lies in private data, not the data previously stored in ERP systems, but the private data that encompasses various decision-making processes and cognitive iterations within the company.
Xia Yiping, CEO of Xpeng Motors, on "Getting on the Big Model": A car without a big model is a functional car
According to IT Home on December 24, Xia Yiping, CEO of Xpeng Motors, discussed the topic of "Will 2024 be the year of the new energy exam?" on Weibo, believing that after this year, the "year of intelligent cars," next year will be the arena of smart cars. Xia Yiping believes that next year will undoubtedly usher in a revolution in intelligent cars, and large models will be the watershed. A car without a large model is a functional car, while a car with a large model is an intelligent car, and over 90% of car owners who replace fuel cars will choose intelligent cars. "With a large model on board, cars truly become intelligent." He also stated that the next 3-5 years will be the era of large models, and large models will be the OS of intelligent cars.
Research Reports
Report: Global shipments of generative AI smartphones to reach 522 million units by 2027
According to IT Home on December 21, a report titled "Insight into the Shipment of Generative AI Smartphones" released by market research firm Counterpoint Research estimates that 2024 will be a critical year for generative AI smartphones, with estimated shipments reaching 100 million units. The organization forecasts that by 2027, global shipments of generative AI smartphones will reach 522 million units, with a compound annual growth rate of 83%.
The organization defines "generative AI smartphones" as falling under the category of AI smartphones that can use generative AI to create original content and can run AI models locally. The organization considers Samsung and Qualcomm to be direct leaders in this field.
5% to 8% of enterprises in China will see a leap in large model parameters to the trillion-level
According to China National Radio, a research report released by the Institute of Industrial Economics of the Ministry of Industry and Information Technology predicts that by the end of 2024, 5% to 8% of enterprises in China will see a leap in large model parameters from the hundred-billion level to the trillion-level, with a 320% increase in computing power demand. Currently, there are five companies in China with parameter scales reaching the trillion-level, and as the parameter scale continues to break through, the speed at which AI large models empower various industries is also increasing. The top 50 AI large model industry applications in China cover 13 fields, mainly concentrated in the financial industry, followed by the industrial, government, and transportation industries. Among the top ten AI large model enterprises in China, 100% of them have independent computing resources.
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。
