[Collection of Big Model Daily on November 14] Nvidia releases a new generation of AI chip H200, with performance increased by up to 90%; Microsoft AI assistant Copilot has been integrated with Edge browser PDF files; Google joins hands with UC Berkeley to throw out Wangbo IGN: to replace diffusion Model to generate realistic images in a single step

China AIGC data annotation panoramic report: market size of tens of billions, employment gap of one million

Link: https://news.miracleplus.com/share_link/11702

Data annotation is facing a critical moment of reshuffle. The era of large models has arrived, and the data-centered AI development model is accelerating to the forefront. The value of data has never been fully exploited as it is today – large models are iterated from training to deployment and application, AIGC has been implemented in many vertical scenarios, and general intelligence and tools have been implemented. Exploration in cutting-edge fields such as personal intelligence is inseparable from high-quality, professional scene data. As the underlying basic service, data annotation has never received as much attention as it does today, but at the same time, opportunities and challenges arise.

Kai-fu Lee’s AI company “Zero One Everything” open source Yi large model was accused of plagiarism LLaMA

Link: https://news.miracleplus.com/share_link/11703

Finnish artificial intelligence startup Silo AI launches open source large model Poro

Link: https://news.miracleplus.com/share_link/11704

Silo Al, an artificial intelligence startup headquartered in Helsinki, Finland, this week launched a new open source large language model Poro, which has 34.2 billion parameters and is based on training on part of the 21 trillion token multilingual data set, covering English, Finnish and programming languages such as Pvthon and Java.

Microsoft AI assistant Copilot has been integrated with Edge browser PDF files

Link: https://news.miracleplus.com/share_link/11705

Recently, some netizens discovered on Twitter (now known as X) that Microsoft’s AI assistant Copilot is now integrated with PDF files. This means that users can directly use Copilot’s features when opening PDF files using the Edge browser, making it easier to get help with work and creation to improve efficiency.

NVIDIA releases new generation AI chip H200, performance improved by up to 90%

Link: https://news.miracleplus.com/share_link/11706

NVIDIA today released a new generation of artificial intelligence (AI) processor H200. This is a graphics processing unit (GPU) designed to train and deploy various artificial intelligence models. The new H200 GPU is an upgrade from the current H100 and integrates 141 GB of memory. When used for reasoning or generating answers to questions, performance is improved by 60% to 90% compared to the H100.

NVIDIA: B100 will be launched next year and its AI performance is more than twice that of H200

Link: https://news.miracleplus.com/share_link/11707

Nvidia recently revealed that the Blackwell architecture B100 GPU, launched in 2024, beats A100, H100 and H200 in the GPT-3 175B inference performance benchmark, and its AI performance will be more than twice that of the Hopper architecture H200 GPU.

Google joins hands with UC Berkeley to throw out IGN: To replace the diffusion model, realistic images can be generated in a single step

Link: https://news.miracleplus.com/share_link/11708

Google recently teamed up with the University of California, Berkeley (UC Berkeley) to develop a new generative AI method that can replace diffusion models (Diffusion Models) – Idempotent Generative Network (IGN). Including Generative Adversarial Networks (GAN), Diffusion Models, and Consistency Models released by OpenAI in March this year, current mainstream generative AI models are all random noise, sketches, or low resolution or other damage. An input, such as an image, is mapped to an output (usually a natural image) corresponding to a given target data distribution to generate an image.

It is reported that Samsung’s Gaussian AI large model supports Chinese and is used in the Galaxy S24 series of mobile phones of the National Bank

Link: https://news.miracleplus.com/share_link/11709

At the Samsung Artificial Intelligence Forum 2023 earlier this month, Samsung officially announced its self-developed generative AI product Gauss, named after the legendary mathematician Gauss. Samsung says Gauss is currently used to improve employee productivity, but will be expanded to various official Samsung apps in the near future to provide new user experiences. According to blogger @Tech_Reve, Samsung Gauss’s language model supports not only Korean, but also English, French, Spanish, Chinese and Japanese. In other words, domestic users are also expected to experience the Gaussian large model in the future.

Introducing Agent into RPA, Tsinghua University and Wall-facing Intelligence release new automation paradigm APA

Link: https://news.miracleplus.com/share_link/11710

Recently, Facewall Intelligence and researchers from Tsinghua Natural Language Processing Laboratory and other institutions jointly released a new generation of process automation paradigm Agentic Process Automation. This paradigm not only realizes the automation of workflow construction, but also introduces dynamic decision-making during workflow execution. automation. This innovation will bring a higher level of efficiency and flexibility to the field of future automation, freeing humans from heavy labor.