[March 1 Big Model Daily Collection] The first batch of Sora-like models appeared, Salabu launched Snap Video, the effect is better than Pika, not inferior to Gen-2; ControlNet author’s new work: AI painting can be divided into layers! The project earned 660 stars without being open source; the open source code generation model StarCoder 2 is newly launched; AI Pin manufacturer Humane signed an investment agreement with South Korea’s SK Telecom, and the first batch of products is expected to be delivered in mid-April

The first batch of Sora-like models appeared, Sarabu launched Snap Video, the effect is better than Pika and not inferior to Gen-2

Link: https://news.miracleplus.com/share_link/19903

Recently, the popularity of the OpenAI video generation model Sora has brought a new wave of heat to diffusion models based on Transformer, such as the DiT (Diffusion Transformer) proposed by William Peebles, one of Sora’s R&D leaders, and Xie Saining, an assistant professor at New York University, last year. Of course, as the AI trend in video generation continues to evolve, there will be more and more models with similar architectures. Just yesterday, Snap Company, which developed the SnapChat picture sharing software, the University of Trento and other institutions jointly released Snap Video, a text generation video model similar to Sora. This time they used the scalable spatio-temporal Transformer. The related paper “Snap Video: Scaled Spatiotemporal Transformers for Text-to-Video Synthesis” has been released.

New work by the author of ControlNet: AI painting can be divided into layers! The project received 660 stars without being open source

Link: https://news.miracleplus.com/share_link/19904

“It’s by no means a simple cutout.” A recent study launched by the author of ControlNet has received a lot of attention – just a prompt, you can use Stable Diffusion to directly generate single or multiple transparent layers (PNG)! This is the new method proposed by the authors of ControlNet – LayerDiffusion, which allows large-scale pre-trained latent diffusion models (Latent Diffusion Model) to generate transparent images.

Completed 6 months of experiments in 5 days, accelerating catalysis research, and the “autonomous driving” catalysis laboratory Fast-Cat was published in the Nature sub-journal

Link: https://news.miracleplus.com/share_link/19905

At the end of January this year, the University of Amsterdam in the Netherlands developed an independent chemical synthesis AI robot “RoboChem”. Within a week, the synthesis of approximately 10 to 20 molecules can be optimized. This takes a PhD student several months. There is also the independent laboratory A-Lab developed by the DeepMind team last year, which independently synthesized 41 new compounds in 17 days. Coscientists at Carnegie Mellon University independently design, plan and execute complex scientific experiments…Now, in collaboration with North Carolina State University and global specialty materials company Eastman Chemical Company, An “autonomous driving” catalysis laboratory: Fast-Cat was developed to use AI and automated systems to conduct in-depth analysis of catalytic reactions used in chemical research and manufacturing. Fast-Cat can provide more information in five days than a traditional test takes six months. Fast-Cat is fully autonomous, using artificial intelligence and automated systems to continuously perform high-temperature, high-pressure, gas-liquid reactions. Autonomous technology also analyzes the output of each reaction to determine (without human intervention) how different variables affect the results of each experiment.

The “national team” has entered the game, and Lianhui Technology, a large multi-modal model company, announced the completion of a new round of strategic financing of hundreds of millions of yuan

Link: https://news.miracleplus.com/share_link/19906

Recently, Hangzhou Lianhui Technology Co., Ltd. (hereinafter referred to as “Lianhui Technology”) announced the completion of a new round of strategic financing of several hundred million yuan. The investors include China Mobile Industry Chain Development Fund China Mobile and Venture Capital, Qianhai Ark (Qianhai It is composed of several leading state-owned assets and market-oriented institutions including Zhongyuan Qianhai Fund and Qilu Qianhai Fund, a parent fund management institution). The leading investor, China Mobile Industry Chain Development Fund, China Mobile and Venture Capital, was established by China Mobile, the Beijing Municipal Government, and the Shanghai Municipal Government to implement the requirements of the modern industrial chain length of central enterprises and serve the digital economy and modern mobile information industry. national strategies such as chain development and strategic emerging industries.

The open source code generation model StarCoder 2 is newly launched!

Link: https://news.miracleplus.com/share_link/19907

StarCoder2 is a series of open large language models for code, providing models in 3 sizes, including 3 billion (3B), 7 billion (7B) and 15 billion (15B) parameters respectively. In particular, the StarCoder2-15B model was trained on more than 4 trillion tokens and more than 600 programming languages, based on The Stack v2 dataset. All models adopt the Grouped Query Attention mechanism (Grouped Query Attention), with a context window of 16,384 tokens and a sliding window attention of 4,096 tokens, and use “Fill-in-the-Middle” Technology training. StarCoder2 includes models at three scales: a 3 billion parameter model trained by ServiceNow(), a 7 billion parameter model trained by Hugging Face, and a 15 billion parameter model trained by NVIDIA on NVIDIA accelerated infrastructure using NVIDIA NeMo.

AI Pin manufacturer Humane signed an investment agreement with South Korea’s SK Telecom, and the first batch of products is expected to be delivered in mid-April

Link: https://news.miracleplus.com/share_link/19908

It is reported that SK Telecom has become Ai Pin’s exclusive telecommunications partner in South Korea. The company announced that it has obtained Humane’s AI technology and CosmOS system authorization. In the future, it will integrate SKT’s mobile network, Internet of Things, cloud, AI and 5G technologies to jointly develop AI applications. According to reports, Humane is a startup company founded by former Apple designer Imran Chaudhri and software developer Bethany Bongiorno. The company launched the wearable artificial intelligence device Ai Pin in November last year. The device itself does not have a screen, but uses a projection display : It can project a 720p monochrome UI interface on the palm of the hand, and users can interact through voice and gestures. In addition, Ai Pin also comes with 2 breathing lights to notify users of incoming calls, text messages, etc.