After foreign media reported that ByteDance used GPT to train an artificial intelligence model, the account was suspended by OpenAI. Subsequently, ByteDance responded that it had taken steps to ensure that API calls to GPT met the requirements of the specification. Google gemini revealed that Chinese uses Wenxin Yiyan for training, and has not yet responded to this matter.
Generated by Unbounded AISources: 01000000000 plans to raise 200 million US dollars, and Alibaba Cloud may participate
According to Reuters on December 14**, two sources familiar with the situation revealed that the AI company founded by Sinovation Ventures Chairman and CEO Kai-Fu Lee is in the process of a new round of financing, with a maximum of $200 million in funding. Alibaba Cloud had earlier pledged to fund the round but did not respond to a request for comment at this time, the sources said. Last month, Zero One Everything was valued at $1 billion.
Mistral AI Complete 3A new €8.5 billion funding round will see the launch of new large language models
According to Bloomberg on December 11, French artificial intelligence startup Mistral AI raised 38.5 billion euros (approx. 4.5 billion.)$1.5 billion) to advance its open source software. The round was led by Silicon Valley venture capital firm A16z (Andreessen Horowitz) and initial investor Lightspeed Ventures, valuing the company at about $2 billion. Mistral AI said it will launch new large language models and open up its business platform.
Glean, an AI company founded by the former Google search team, is looking to raise at least $200 million at a $2 billion pre-investment valuation
Glean, a startup founded by a former Google search engineer, is in talks with investors to raise at least $200 million at a $2 billion pre-investment valuation, according to The Information on December 14.
Kleiner Perkins, which has also invested in GLEAN's first three funding rounds, is in discussions to co-lead the latest funding round, according to one of the people familiar with the matter. The funding will double GLEAN's valuation from May 2022, and the company's annual recurring revenue has increased to more than $30 million as of the quarter, at least three times higher than the same period last year.
People familiar with the matter: Legal AI startup Harvey is in talks for a new round of funding at a $700 million valuation
Legal AI startup Harvey is in talks for a $70 million to $80 million funding round that will be valued at $700 million, including investments, according to The Information on December 14, according to two people familiar with the funding situation. It is reported that the financing is more than four times the valuation it announced in April.
Venture capital firm Kleiner Perkins and individual investor Elad Gil, who has previously invested in the company's seed and Series A rounds, will co-lead the round, according to people familiar with the matter. In addition, existing investors Sequoia Capital and OpenAI Ventures** will also participate in this round of investment.
AI recruitment platform HelloBoss has raised millions of dollars in Pre-A round of funding, led by Yunqi Capital
According to the "Science and Technology Innovation Board**" on December 12**, Japan's AI recruitment platform Helloboss completed a multi-million dollar Pre-A round of financing, led by Yunqi Capital, followed by Golden Egg, and Silver Oak Capital served as the exclusive financial advisor. This round of financing will be used for product research and development, market expansion and team building. HelloBoss is aiming at AI+HR services, which help companies and job seekers quickly match and complete recruitment through AI.
Artificial intelligence medical equipment developer "Ritu Biotech" completed a B2 round of financing of over 100 million yuan
According to the National Business Daily, recently, Ritu Biotech completed a B2 round of financing of over 100 million yuan, led by Guoke Investment and well-known domestic **, with Shenzhen Investment Holding Capital participating in the investment, and Dianshi Capital serving as the exclusive financial advisor. This round of financing will be mainly used for the production, commercialization and R&D iteration of core products. RETO Biotech is a developer of artificial intelligence medical equipment, the company is mainly engaged in providing human body full specimens, full process, fully automated, fully intelligent inspection products, and has developed a new generation of deepcell based on artificial intelligence.
AI startup Arcane raised $5 million in funding led by Accel
London-based AI startup Arcane has raised $5 million in funding, led by U.S. venture capital firm Accel, with participation from Cocoa, FirstMinute Capital, Kima Ventures and Seedcamp, according to Business Insider on December 14. In addition, angel investors, including Connor Hayes, VP of generative AI products at Meta, Tom Blomfield, founder of Monzo, Ilkka Paanen, CEO of Supercell, and Anthony Yazbeck, president of WeWork, also backed the round.
Foreign media: ByteDance is secretly using OpenAI technology to build its own large model
According to The Verge, quoted by Jin Shi on December 16, ByteDance is secretly using OpenAI's technology to develop its own large language model (LLM). In the field of artificial intelligence, this practice is often considered inappropriate. This is also a direct violation of OpenAI's Terms of Service. When ByteDance used GPT to train their AI model, OpenAI suspended their account. However, most of ByteDance's GPT usage is done through Microsoft's Azure platform, not directly through OpenAI. It's unclear whether Microsoft will follow OpenAI's lead and suspend access to ByteDance.
ByteDance responded that it has taken steps to ensure that API calls to GPT comply with the specification.
Gemini revealed that Chinese uses Wenxin Yiyan for training
According to Qubit on December 18**, Google's gemini Chinese corpus comes from Wenxin Yiyan. After testing, when using the model for Chinese dialogue on Google's Vertex AI platform, Gemini-Pro directly stated that it was a large model of Wenxin. In Google AI Studio, Gemini-Pro shows that it uses Wenxin on training data in Chinese. There has been no response to this matter.
Microsoft has released phi-2, a 2.7 billion parameter language model that matches or outperforms models that are 25 times larger
On December 13, Microsoft released a language model with 2.7 billion parameters—phi-2. According to reports, due to innovations in model scaling and training data wrangling, PH-2 can match or outperform models that are 25 times larger than it in complex benchmarks.
It is reported that PHI-2 is a transformer-based model, which is in 14 t chunks were trained on synthetic and network datasets from NLP and encoding. Training on PHI-96 on 100 A2 GPUs took 14 days. PHI-2 is a foundational model that has not been adjusted by human feedback reinforcement Xi (RLHF) and has not undergone guided fine-tuning. Nonetheless, PHI-2 has better performance in terms of toxicity and bias compared to aligned existing open-source models.
In a variety of comprehensive benchmarks, PHI-2 outperformed the Mistral and LLAMA-2 models at 7 billion and 13 billion parameters. Notably, on multi-step inference tasks (i.e., coding and math), PH-2 outperformed the LLAMA-2-70 B model. In addition, the performance of the PHI-2 is on par with, or even better than, the recently released Google Gemini Nano 2.
Mistral AI's first open-source MOE model, Mistral-8x7b, has been launched
According to the webmaster's home on December 14, Mistral AI recently released the first open-source MOE model Mistral 8x7b, and announced that it will be launched in the Moda community.
Mictral-8x7B is a hybrid expert model (MiXeTrueof Experts), composed of 8 expert networks with 7 billion parameters, in terms of capabilities, Mictral-8x7B supports 32K token context length, supports English, French, Italian, German and Spanish, has excellent generation capabilities, and can be fine-tuned as a instruction-following model. The model achieves an MT-bench score of 83 points, with GPT35 quite.
Alibaba's research department launched Seallm, a large AI model tailored for Southeast Asia
According to the webmaster's home on December 12, Alibaba Group's research institute DAMO Academy launched Seallm, an artificial intelligence large language model tailored for Southeast Asian languages, highlighting the company's ambition to develop the market in a wider region. Southeast Asia LLM (Seallm) is pre-trained on Vietnamese, Indonesian, Thai, Malay, Khmer, Lao, Tagalog, and Burmese datasets and outperforms other open-source models in terms of language and security tasks, according to Alibaba Research.
Zhongke Wenge released Yayi 20 large models: 30 billion parameters, support for 128k contextual window input
According to the "Science and Technology Innovation Board" on December 15, Wenge Company, which was incubated by the Chinese Academy of Sciences, released Yayi 2 today0 large models, 30 billion scale parameters, support 128k contextual window input, ** multimodal capabilities, more than 10 language dialogues, and intelligent tool calls. Wang Lei, chairman of Zhongke Wenge, said that Yayi 20 is one of the few domestic native large models that are pre-trained from scratch, focusing on safe and reliable government and enterprise large models, supporting commercial use by open source, and building a large model ecology in the fields of security, finance, and public opinion. It is reported that Zhongke Wenge released the Yayi model in June this year.
The Midjourney v6 model will be released next week, allowing you to draw more complex content
According to the webmaster's home on December 18, the Midjourney V6 model will be released next week, and Midjourney officially began the community style evaluation today. According to some of the community reviews**, the v6 model has improved significantly in reproducing complex picture information and content, and is also more daring to draw more complex content.
According to the news, Google's Pixel 9 series phones will have a new and exclusive AI assistant "pixie" built-in
According to the IT Home on December 16, comprehensive The Information, 9to5 Google and other foreign media reports, Google is developing a new AI assistant, which is based on its latest Gemini model, will be the exclusive assistant for Pixel devices, called "pixie".
According to the report, Pixie will become a "more personalized" Google Assistant, with the goal of performing complex multimodal tasks, using Gmail, Maps and other data from Google products on Pixel series phones. For example, suggest the nearest store to the user.
Lenovo AI PCs will be the first to carry a large personal model next year
According to the "Science and Technology Innovation Board**" on December 14**, Wang Chuandong, vice president of Lenovo Group and chief marketing officer of China, said that in the AI era, AI PC will become the preferred terminal for AI inclusiveness, and Lenovo AI PC will be the first to carry a personal large model next year. In the future, the combination of AI and PC will form a new hybrid of "computing platform + personal large model + AI application".
Nanjing Agricultural University developed the country's first ancient AI model "Xunzi".
According to the official website of the "Nanjing Agricultural University Alumni Association", the research team of Professor Wang Dongbo of the School of Information Management of Nanjing Agricultural University released the "Xunzi" ancient book large language model in Beijing.
According to reports, the "Xunzi" large model includes the vast majority of ancient books including the "Siku Quanshu", with a large corpus of more than 2 billion words, which is an intelligent tool for the processing and research of ancient books launched by the school's team in conjunction with the Gulian Company of Zhonghua Book Company with the support of the national social science major project "Construction and Application Research of Cross-language Knowledge Base of Ancient Chinese Classics". The model consists of two parts: the pedestal model xunziallm and the dialogue model xunzichat, and users can also use the local training corpus to fine-tune the pedestal model according to their own needs, so as to achieve better processing performance in the downstream processing of ancient books.
The Wenxin Yiyan plug-in** is launched, providing functions such as mind map and audio extraction
AI New Intelligence News, on the evening of December 13th, AI official *** issued an announcement that the Wenxin Yiyan plug-in has now been officially launched, including plug-ins covering various use scenarios such as office efficiency improvement, multi-modal content understanding and generation, and professional information query.
Users only need to use simple instructions to realize the needs of PPT generation, audio extraction, mind map production and other multi-scene and multi-modal needs. The plug-ins currently available include mind mapping tools, ppt generators, and "assistants" who can refine key information by input.
Meitu launched WHEE, an AI image generator for mobile, based on the MiracleVision model
On December 13, Meitu announced the launch of its mobile AI image creation tool WHEE, which provides one-stop AI visual creation services based on Meitu's MiracleVision model.
Xuhui District, Shanghai issued the "Several Opinions of Xuhui District on Supporting the Development of Artificial Intelligence Large Models (Trial)".
According to the **Times E Company on December 15**, Xuhui District, Shanghai issued the "Several Opinions of Xuhui District on Supporting the Development of Artificial Intelligence Large Models (Trial)", which proposes to strengthen the overall planning of urban computing power, implement the computing partner plan, and support the computing power cost of innovative subjects who carry out the training and optimization of general large models and vertical large models. If the relevant innovative entity purchases or leases computing power, it is determined that it can give a subsidy of no more than 10 million yuan at a maximum of 30% of the actual expenses incurred.
Analysis: Intel's Gaudi AI chip orders have reached $2 billion
According to the "Science and Technology Innovation Board**" on December 14**, American brokerage Raymond James recently said that although PC sales are declining, Intel may face difficulties in the short term, but investors should pay attention to the company's long-term opportunities, such as artificial intelligence. Analyst Srini Pajjuri reiterated Intel's "outperform" rating on Dec. 12 and raised his price target to $48 from $42. Analysts point out that orders for Intel's Gaudi AI accelerator, which Intel has developed specifically for AI, are expected to have the ability to give these chips** to enterprises for use in smaller industry-specific AI models, analysts said.
Da Mo: Microsoft reduced orders for H100 AI servers to 80,000 units, and transferred some orders to B100
According to the latest news from Da Mo, Microsoft may reduce orders for H100 AI servers from 120,000 to 80,000 units in 2024, and transfer some orders to B100, which is expected to ship in the fourth quarter of 2024, because the computing power of B100 is at least twice that of H100, but ** is only 50 60% higher. In addition, some orders are transferred to the Supermicro MI300.
Robin Li: The 100-model war is a great waste of social resources, and more resources should be placed in super applications
According to Yicai report, on December 16, Robin Li, founder, chairman and CEO, pointed out in a conversation with Zhang Peng, founder and president of Geek Park, that "the 100-model war is a great waste of social resources, and more resources should be placed in super applications." "There are three key points in the development of large models: industrial policies encourage AI-native application development;Existing enterprises can use large models to have a positive effect on core indicators;The advent of super apps.
According to Sina Technology, Robin Li also said that the value of the large model may be to generate new super applications, or it may be a transformation of existing products. For example, he said, Microsoft launched Copilot based on Office 360, which costs $30 a month, and the annual revenue is several times higher than OpenAI. Therefore, enterprises need to think more about how to integrate AI with their existing business.
Kai-Fu Lee: Five or six large models will survive in the end, and the opportunity to do AI applications has just begun
According to the first financial report, on December 14, in an AI-related conversation, Kai-Fu Lee, CEO of 010000 Things, said when asked how many companies would survive in the end of the large model track, "In addition to the big manufacturers, China and the United States together should be able to support about five or six." In his view, the opportunity to do huge pre-trained models cannot be said to be closed, but it will become more and more difficult, and new technologies may be needed to have another chance, but the opportunity to do other aspects such as AI Infra (artificial intelligence basic framework technology), AI applications, etc., has just begun.
Sam Altman: AI development is like a tornado, and we understand the concerns of the outside world that OpenAI needs to push the limits
According to The Paper, on December 11, local time, at the HOPE Global Forum held in Atlanta, USA, OpenAI CEO Sam Altman appeared on stage and accepted an interview. Altman said AI will "lift the whole world" and drive potential advancements in areas such as healthcare and education. Altman also said that in literary and artistic creation, he is shocked by the ruthless image of AI such as "Terminator", and he can understand people's concerns about AI;However, in order to realize the full potential of this technology, companies including OpenAI need to work to "push the envelope."
Commenting on the technological revolution brought about by AI and the difference between previous technologies, Altman admits that the speed at which AI technology is evolving is "a little scary" and that OpenAI has "jumped into this tornado that won't stop." Altman also announced that he will be working with John Hope Bryant, founder of the U.S.-based organization Hope, to create a new AI ethics committee in Atlanta.
Report: Generative AI technology is poised to add 1. to India's GDP over the next seven years$5 trillion
According to a report by Ernst & Young India, generative AI has the potential to add 1.0 percent to India's GDP over the next seven years, according to a report by Ernst & Young India2-1.$5 trillion
The report, titled "India's AIDEA: The Potential of Generative AI to Accelerate India's Digital Transformation," provides insights into the industry's readiness and challenges in embracing AI general. According to the report, the new generation of AI could add a cumulative 1. to India's GDP by FY29-302-1.$5 trillionNearly 69% of the overall impact is expected to come from sectors such as business services (including IT, legal, consulting, outsourcing, machinery and equipment leasing, etc.), financial services, education, retail, and healthcare.
Institutions: AI talent is in short supply ChatGPT researchers earn an average monthly salary of 670,000 yuan
According to the **Times on December 16**, a few days ago, the "2023 Talent Migration Report" released by Maimai Gaopin shows that in 2023, the new economy will continue to be saturated with talents, and the talent supply and demand ratio will increase from 0 in 202132 rises to 2 in 202304, an average of 2 people compete for 1 position. However, AI-related positions are in short supply, and among the top 20 high-paying positions, there are 10 AI positions represented by ChatGPT researchers, algorithms, and deep learning Xi. ChatGPT researchers earn an average monthly salary of 670,000 yuan tops the list of high salaries.
Report: In 2022, gaming technology contributed 31 percent to AI's cross-industry technological advancement37%
According to the news on December 15, at the 2023 China Game Industry Annual Conference held today, the report "Super Digital Scene Source Power - Research on Game Technology and Innovation Application" jointly released by the China Audio and Digital Association and the Industrial Culture Development Center of the Ministry of Industry and Information Technology proposed the concept of super-digital index for the first time, which quantified the cross-field contribution rate of game technology. According to the report, in 2022, the contribution rate of game technology to cross-industry scientific and technological progress in chips, AI, cloud computing, mobile phones, XR, autonomous driving and other industries is as follows: 45% and 5719%。
CCID Research Institute of the Ministry of Industry and Information Technology: This year, the adoption rate of generative AI enterprises in China has reached 15%, and the market size is about 144 trillion yuan
According to the CCID Research Institute of the Ministry of Industry and Information Technology, this year, the scale of China's generative artificial intelligence market is expected to exceed 10 trillion yuan. Generative AI is accelerating its penetration into four major industries: manufacturing, retail, telecommunications, and healthcare. Data shows that this year, the enterprise adoption rate of generative AI in China has reached 15%, and the market size is about 144 trillion yuanThe adoption of generative AI technology in four major industries, namely manufacturing, retail, telecommunications, and healthcare, has all seen rapid growth. Experts**, generative AI is expected to contribute nearly 90 trillion yuan of economic value to the world in 2035, of which China will exceed 30 trillion yuan, accounting for more than 40%.