Collection of daily reports on large models on November 16

News7months agorelease AIWindVane
36 0
Collection of daily reports on large models on November 16

[Collection of daily reports on large models on November 16] Microsoft’s late-night amplification move: GPT-4, DALL·E 3, GPTs are free to use, self-developed AI chips dedicated to large models; Kingsoft Office WPS AI will open public beta today and will be gradually open to all users Experience; OPPO ColorOS 14 is connected to a large AI model for the first time, and the new Xiaobu has been connected to AndesGPT

Open the Android phone, and the screen is filled with three words: Big Model丨Model Power Era



At this stage, Android manufacturers will not hold mobile phone launch conferences easily without a large model. OPPO has just upgraded its voice assistant with a large model, and vivo has officially announced its own large-scale mobile phone AI model. Xiaomi directly inserted the large model into the mobile phone system at the press conference… The competition is as fierce as competing for the first chip. what is going on? The reason is that smart terminals have become the “new beachhead” for various AIGC applications. First, large image generation models were inserted into mobile phones one after another. From Stable Diffusion with one billion parameters, a golden retriever puppy was quickly generated on the mobile phone. Run ControlNet with 1.5 billion parameters on your mobile phone to quickly generate an AI landscape photo with a limited image structure. Subsequently, large text generation models also rushed to launch new mobile applications – domestically there are Wenxin Yiyan and Zhipu Qingyan APP, while abroad there is OpenAI’s mobile version ChatGPT, and the Llama 2 mobile version is also being prepared urgently. Now, in this wave of large-scale smart terminal models, the lowest level of software and hardware technology gears are starting to turn. From Qualcomm to Apple, the latest chip manufacturers’ press conferences all emphasize the support of software and hardware for machine learning and large models.

Microsoft’s late-night amplification strategy: GPT-4, DALL·E 3, GPTs free of charge, self-developed AI chip dedicated to large models



Early this morning, Microsoft held its latest Ignite conference. CEO Satya Nadella introduced the release and update of more than 100 products and technologies at the conference, covering a wide range of applications, productivity, security, etc. aspects.

OPPO ColorOS 14 is connected to AI large models for the first time, and the new Xiaobu has been connected to AndesGPT



At today’s press conference, OPPO brought the new ColorOS 14, which is smarter, smoother, and safer; at the same time, OPPO also launched its personal large model service AndesGPT, and brought Pantanal smart cross-terminal The system is implemented. According to reports, the new ColorOS Xiaobu has introduced large model support, such as support for call assistants. The smart summary function can extract important content of call records, and can also handle complex mobile phone operations, helping you generate customized content, summary summaries, and more.

Kingsoft Office WPS AI starts public beta today and will be open to all users for experience.



Kingsoft Office announced that WPS AI will open public beta starting today and will be open to all users for experience. The functions of WPS AI cover text, tables, and presentations.

Making development and operation and maintenance easy in the world, DevOps finally has a dedicated large model, jointly released by Ant and Peking University



Large language models have made significant progress in various NLP downstream tasks, and large models in various vertical fields are springing up like mushrooms after a spring rain. However, in the field of DevOps, no relevant large model has been released. To fill the gap in this area, Ant Group and Peking University released the first open source large language model for the Chinese DevOps field, namely DevOps-Model. This large model is designed to help developers improve efficiency throughout the entire life cycle of software development and operation. The ultimate goal is to achieve solutions to any problems faced in the DevOps process by asking questions to the DevOps-Model.

When data becomes “production data”, three papers summarize how to protect the copyright of AI training data with watermark technology



Currently, some existing data protection technologies, such as encryption, digital watermarking, differential protection, etc., are mainly designed to prevent unauthorized users from using protected data. However, these methods are not suitable for protecting publicly released datasets on which DNN training relies. Specifically, encryption and differential protection processing will affect the normal functionality of protected data sets, and digital watermarking technology has little effect in this scenario, because unauthorized users will only publish their trained models without Their training samples will be made public. How to protect publicly released datasets remains an important open question. This problem is challenging because the attacker has access to the compromised data set. The security of data sets is a key issue that must be faced in the promotion and application of AI. Therefore, it has attracted widespread attention from the industry.

The Bank of Korea says 4 million jobs in South Korea may be replaced by AI in the next 20 years



A research report released by the Bank of Korea on Thursday stated that artificial intelligence may replace as many as 4 million jobs in the next 20 years, accounting for 14% of the total number of jobs in South Korea. The findings further warn of impending disruption in labor markets around the world. A Bank of Korea research team said: “As artificial intelligence can handle analytical and cognitive tasks more easily, workers with higher education and higher incomes face greater threats.” They said pharmacists, doctors, lawyers, accountants and asset management companies are most at risk, while jobs in religious fields, food services, teaching and singing are least vulnerable.

© Copyright notes

Related posts

No comments

No comments...