Big Model Daily, January 12

News4months agorelease AIWindVane
17 0
Big Model Daily, January 12

[Big Model Daily, January 12] To supplement mathematics for large models, the open source MathPile corpus was submitted, with 9.5 billion tokens, and it can also be used commercially; Nat.Commun. | The Shandong University team developed a rare cell group based on graph transformer Analysis method; Byte allows Da Vinci and Mona Lisa to “quarrel across the air” with just a picture, an audio clip, and an emotional clip; Luo Yonghao suspends AR business and turns to large models, Thin Red Line Company’s software, hardware, and algorithm teams Half of the layoffs

To supplement mathematics for large models, submit the open source MathPile corpus with 9.5 billion tokens, which can also be used commercially

In the current development of intelligent dialogue models, powerful underlying models play a crucial role. The pre-training of these advanced models often relies on high-quality and diverse corpora, and how to build such a corpus has become a major challenge in the industry. In the high-profile field of AI for Math, the relative scarcity of high-quality mathematical corpus limits the potential of generative artificial intelligence in mathematical applications. In order to meet this challenge, the Generative Artificial Intelligence Laboratory of Shanghai Jiao Tong University launched “MathPile”. This is a high-quality, diverse pre-training corpus specifically targeted at the field of mathematics, which contains approximately 9.5 billion tokens and is designed to improve the capabilities of large models in mathematical reasoning. In addition, the laboratory also launched a commercial version of MathPile – “MathPile_Commercial” to further broaden its application scope and commercial potential.

Nat.Commun.|The Shandong University team developed a rare cell group analysis method based on graph transformer

Rare cell populations are key to tumor progression and treatment response, providing potential targets for intervention. However, their computational identification and analysis often lag behind major cell types. To fill this gap, a research team from Shandong University introduced MarsGT: multi-omics analysis using single-cell graph transformer for rare population inference. It uses a probability-based heterogeneous graph transformer to identify rare cell populations on single-cell multi-omics data. MarsGT outperforms existing tools in identifying rare cells in 550 simulated datasets and 4 real human datasets.

AI methods for describing liquids and soft matter open a new chapter in density functional theory

Scientists at Universität Bayreuth have developed a new method for studying liquids and soft matter using artificial intelligence, opening a new chapter in density functional theory. We live in a highly technological world, and in this dense and complex interconnected network, basic research is the engine of innovative development. The new method here could have a huge impact on a wide range of simulation techniques, allowing for faster, more precise and deeper study of complex matter on the computer. In the future, this may have implications for product and process design. The fact that the newly formulated neuromathematical relations can represent the structure of liquids well is a major breakthrough, opening up a range of possibilities for gaining deep physical insights. “In this study, we show how artificial intelligence can be used for fundamental theoretical physics to solve the behavior of fluids and other complex soft matter systems,” says Professor Matthias Schmidt, Head of the Department of Theoretical Physics II at the University of Bayreuth. We have developed an advanced scientific approach to study matter at the atomic and (macro) molecular level, combining machine learning and mathematical methods to calculate complex physical properties.”

Byte allows Leonardo da Vinci and Mona Lisa to “quarrel across the air” with just a picture, an audio, and an emotional clip

AIGC shows extraordinary potential in the field of video generation. Recently, the Bytedance Intelligent Creation Team and the University of Texas at Dallas proposed a diffusion model-based framework called DREAM-Talk: receiving a driven audio sequence, a given portrait picture and an example of an emotional style ( A video of an emotional talking face) as input can generate a lifelike, lip-synchronized video of a talking face, which contains high-quality emotional expressions, even portraits can be performed lifelike, and supports multiple languages.

Behind the popular imaging tool Magnific: a two-person startup

Recently, an image super score & enhancement tool called Magnific has become popular. Magnific can improve the clarity and sharpness of the image, and can complete the details in the picture through your prompts. It can even use AI to help you complete your imagination in some subtleties. This not only upconverts the image, but also completes the re-creation of the image.

Luo Yonghao postpones AR business to develop large-scale models, and lays off half of the software, hardware and algorithm teams of Thin Red Line Company

Luo Yonghao’s thin red line AR business software and hardware research and development has slowed down, AR-related business layoffs, may shift to the large model industry. An industry executive revealed to Sohu Technology that Luo Yonghao no longer makes AR hardware and may plan to make large models.

AI reasoning performance increased by 7 times. Alibaba Cloud’s eighth-generation enterprise-class instance g8i is here!

Today, Alibaba Cloud officially released the eighth-generation enterprise-class general computing instance ECS g8i. Based on Alibaba Cloud’s self-developed “Feitian+CIPU” architecture system and the fifth-generation Intel Xeon scalable processor, the overall performance of the g8i instance has been improved by up to 85% %, AI reasoning performance is improved by up to 7 times, and can support large language models with up to 72B parameters, helping to reduce the cost of building small and medium-sized models by 50%. At the same time, the new instance also provides end-to-end security protection, providing powerful privacy-enhanced computing power support for enterprises to build trusted AI applications.

OpenAl’s ChatGPT Enterprise version gains 260 enterprise customers within 4 months of launch

According to Bloomberg, OpenAI COO Brad Lightcap said in an interview that the 260 enterprise customers currently using ChatGPT Enterprise have a total of more than 150,000 employees registered to use the product. ChatGPT Enterprise, launched in August, offers enhanced functionality and privacy protection to meet the specific needs of enterprises. These include data encryption and technology to ensure that customer information is not used to develop OpenAl. ChatGPT Enterprise’s pricing is flexible and can be customized to suit each company’s requirements.

© Copyright notes

Related posts

No comments

No comments...