“ChinAI #257: Can Chinese companies keep up with Sora?” by Jeffrey Ding
Mar 11, 2024
auto_awesome
The podcast delves into Chinese companies trying to match Sora's AI advancements, focusing on Shengshu-AI and its Tsinghua professor chief scientist. It explores China's progress in video generation, comparing methods with Sora. Additionally, it covers private hacking contractors in China, leaked documents revealing practices, and the tech-government alliance in surveillance.
Chinese companies are striving to close the technological gap with Sora by leveraging innovative DIT method from Xing Xiuai.
Transitioning to longer video generation tasks requires significant technical challenges and expertise beyond computing power, emphasizing the role of engineering optimization.
Deep dives
Chinese Companies' Efforts to Keep Up with Sora
Chinese companies, particularly Xing Xiuai, are striving to close the technological gap with Sora, an AI model that surprised the industry with its video generation capabilities. The emergence of Sora has raised concerns about the increasing divide between Chinese and Western AI advancements. Despite challenges in computing power and resource limitations, Xing Xiuai remains optimistic due to their work on a model similar to Sora's innovative DIT method. By highlighting the importance of engineering technology and leveraging experiences from 2D image generation, Xing Xiuai aims to navigate the complexities of advancing to longer video generation tasks.
Technical and Resource Challenges in Developing Sora-Like Capabilities
The journey towards achieving Sora-like capabilities involves significant technical and resource challenges, as outlined by Xing Xiuai's CEO and team. The industry estimates that substantial resources, equivalent to around 10,000 chips and Vidya A-series, are required to reach a satisfactory level. While Xing Xiuai has expertise in training large-scale models efficiently, transitioning from short videos to longer formats demands robust training data and engineering optimization. OpenAI's prior models and accumulated experiences play a crucial role in bridging the gap, emphasizing the complexities beyond sheer computing power in advancing video generation technologies.
…***We’ve hit a bit of a lull in paid subscriptions lately, so please consider subscribing here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay support access for all AND compensation for awesome ChinAI contributors). As always, the searchable archive of all past issues is here.
Feature Translation: Can Chinese companies make Sora? This Tsinghua large model team gives hope
Context: Thanks for exercising your right to vote. In a close contest, the #6 option (link to original jiqizhixin article) won out: How are Chinese companies trying to keep up with Sora? This week we zoom in on Shengshu-AI (生数科技), a Beijing-based generative AI startup that has raised 14 million USD and boasts Jun Zhu (Tsinghua professor who co-directors Tsinghua's AI lab) as chief scientist. There are hopes and expectations that Shengshu [...]