Babbitt丨Daily Must Read: Shanghai Launches 11 Measures for the Innovation and Development of Large Models; Alipay launched a free AI photo function

Abstract:Shanghai has recently launched 11 measures to promote the innovation and development of large-scale pre-trained models and accelerate the construction of a world-class industrial cluster for artificial intelligence, including the implementation of the large model innovation support plan, the establishment of a large model testing and evaluation center, and the implementation of the large model intelligent computing power acceleration plan. Alipay recently launched a free AI photo function, which can generate a personal photo album with just one click with just 1 selfie. At present, it is a limited-time free stage, and the generation styles include space explorer, magazine photo, antique style, etc.

Image source: Generated by Unbounded AI

Hot Info:

Shanghai has released 11 measures to promote the innovation and development of AI large models: it will create benchmark large model products and services

The Shanghai Municipal Commission of Economy and Information Technology and other departments recently issued a notice on the issuance of the "Several Measures for Promoting the Innovation and Development of Artificial Intelligence Large Models in Shanghai (2023-2025)", introducing 11 measures to promote the innovation and development of large-scale pre-trained models and accelerate the construction of a world-class industrial cluster of artificial intelligence, and formulate the following measures: 1. Implement the large-scale model innovation support plan; 2. Establish a large model testing and evaluation center; 3. Implement the large-scale model intelligent computing power acceleration plan; 4. Build a collaborative ecosystem of software and hardware for smart chips; 5. Co-construction and sharing of corpus data resources; 6. Implement the large-scale model demonstration application promotion plan; 7. Promote the application of large models of scientific intelligence; 8. Create a large-scale model innovation highland for enterprises and talents; 9. Promote the ecological construction of large-scale model applications; 10. Establish a normalized service mechanism; 11. Increase investment.

Zhou Hongyi: It is said that beyond ChatGPT, the real competition of large models is to combine them with user scenarios

The 2023 World Internet Conference Wuzhen Summit opened today, and Zhou Hongyi, founder of 360 Group, said that 360's AI model has recently been recorded, proving that it has passed the security audit. But he mentioned that there are too many large models now, and everyone is numb, and they all say that they have surpassed ChatGPT. Now it's not a problem to release large models, the real competition is to combine with user scenarios.

NVIDIA has partnered with communications services provider Amdocs to accelerate the adoption of generative AI in the telecommunications industry

On November 7, NVIDIA announced that it is partnering with communications services provider Amdocs to optimize large language models (LLMs) to accelerate the adoption of generative AI in the $1.7 trillion telecommunications industry.

The two companies will customize enterprise-grade LLMs running on NVIDIA accelerated computing as part of the Amdocs amAIz framework, enabling communications service providers to efficiently deploy generative AI use cases across businesses ranging from customer experience to network configuration. Amdocs will use NVIDIA DGX Cloud AI supercomputing and NVIDIA AI Enterprise software to support flexible adoption strategies and help ensure that service providers can use generative AI applications simply and securely. According to reports, Amdocs' customers include more than 350 of the world's leading telecommunications and media companies, including 27 of the world's top 30 service providers.

SmartMore released IndustryGPT V1.0, an industrial multi-modal model

Recently, Jia Jiaya, a tenured professor at Chinese University of Hong Kong and founder and chairman of Smartmore, officially released the world's first industrial multi-modal large model IndustryGPT V1.0. According to reports, the training data of IndustryGPT V1.0 covers a total of 200+ different industrial scenarios and more than 50 billion tokens. In addition, on November 7, SmartMore also signed strategic cooperation with Panasonic and Johnson Controls at the CIIE, and will cooperate in depth in the field of intelligent manufacturing and digital intelligence innovation in the future.

Google's new paper says that "AI is on the verge of surpassing humans" is unrealistic and has its own limitations

On November 8, three Google researchers pointed out in a paper recently submitted to the Preprint Document Library (ArXiv) that deep neural network transformer, the underlying technology in the field of artificial intelligence, is not good at generalization. The three authors wrote: "When a task or function needs to be outside the scope of the pre-trained data, transformers exhibit various failure modes, and even simple task extensions are found to reduce their inductive power." ”

According to the paper, deep neural network transformers are good at performing tasks related to training data and are not very good at handling tasks beyond that. For those looking to achieve Artificial General Intelligence (AGI), this issue cannot be ignored.

Y Combinator has partnered with Microsoft to provide AI startups with services such as Azure Credits

On November 8, Y Combinator, a well-known startup incubator, announced a partnership with Microsoft to provide AI startups with Azure credits and priority access to the ND A100 Azure VMs configuration service for model training and fine-tuning. Y Combinator said that AI startups can take advantage of this partnership opportunity through an application from YC W24.

Samsung showcased Samsung Gauss, a generative AI model that consists of three parts: language, code, and images

On November 8, Samsung introduced Samsung Gauss, a generative AI model developed by Samsung Research, at its AI Forum. It is reported that the model consists of Samsung Gauss Language, Samsung Gauss Code and Samsung Gauss Image, which is currently being used to improve the work efficiency of the company's internal employees, but will be extended to product applications in the future.

Researchers Create Large Models That Recognize ChatGPT-Generated Content with 99% Accuracy

On November 8, according to the latest paper included in sciencedirect, AI Detector researchers created a new model through 20 text features and XGBoost, which can effectively distinguish between human collaboration and GPT-3.5/GPT-4 generated content. The researchers said the model showed 99% accuracy in distinguishing between human writing and ChatGPT-generated text, significantly outperforming GPT-2 Output Detector, an AI detection software provided by ChatGPT maker OpenAI.

Insider: Amazon is developing "Olympus" AI with the aim of closing the gap with Microsoft and OpenAI

Amazon is working on new conversational AI software aimed at catching up with OpenAI and Microsoft in selling such software to enterprise customers, according to a person familiar with the matter. The new large language model, codenamed Olympus, could also power its online retail stores, Alexa voice assistant on devices like Echo, and new features from AWS.

IBM will launch a $500 million venture fund to invest in artificial intelligence companies

As reported on November 7, IBM announced that it will launch a $500 million venture fund to invest in a range of generative AI companies. Rob Thomas, senior vice president and chief commercial officer of IBM software, said the company plans to invest in startups in phases, but has not set an annual target number of investments or a timeline for capital deployment.

Northwestern Polytechnical University and Huawei Collaborate to Develop the First Large-scale Model of Fluid Mechanics for Aircraft "Qinling Soaring"

As reported on November 7, the Convergence Media Center of the Propaganda Department of the Yanta District Party Committee recently announced that the first large-scale model of fluid dynamics for aircraft, "Qinling Soaring", jointly developed by Northwestern Polytechnical University and Huawei, was officially released.

According to reports, the "Qinling Soaring" model is an intelligent model for aircraft fluid simulation jointly developed by the International Joint Institute of Fluid Mechanics and Intelligence of Western University of Technology and Huawei AI4Sci Lab on the basis of the domestic open-source fluid computing software Fenglei, relying on the surging computing power of Ascend AI and the MindSpore AI framework.

Alipay launched the AI photo function, and selfies can generate personal photo albums

Alipay has recently launched a free AI photo function, and you can directly search for "AI photo gallery" on the homepage to enter the mini program. The main information of the Mini Program is Alipay's subsidiary, Squeak (Shanghai) Network Technology Co., Ltd. According to the official introduction of the applet, 1 selfie can generate a personal photo album with one click. At present, it is a limited-time free stage, and the generation styles include space explorer, magazine photo, antique style, etc.

Good Article:

"Don't learn from OpenAI's startup, it's already a monopoly giant"

OpenAI has found a way to get along with a large number of startups that have grown in its ecosystem.

The method is that it itself officially turned into a monopoly giant.

OpenAI made it clear to the world at its annual convention that I would not compete with developers anymore because I myself had become a monopoly. Ultraman solved these doubts through the way of giants - he established the GPT app store and the GPT application distribution system, like Big Brother, and did profit sharing with these entrepreneurs.

"The first AGI will come in 2028?" Google DeepMind proposes 6 AGI standards and defines 5 AGI levels》

Humanity is getting closer and closer to the appearance of the first AGI!

Shane Legg, co-founder of DeepMind and chief AGI scientist, said in a recent interview that there is a 50% chance that humans will develop the first AGI by 2028. The DeepMind research team led by him published a paper on Arxiv, which directly released the roadmap and timeline of AGI.