At the Apsara conference in Hangzhou, hosted by Alibaba Cloud, the Chinese AI startup highlighted its efforts to develop large-scale language models.
The company’s efforts follow Openai’s latest LLM announcement, including a Microsoft-backed O1 generation pretraining transformer model. The model is intended to tackle difficult tasks and pave the way for advances in science, coding and mathematics.
During the meeting, Kunal Zhilin, founder of Moonshot AI, highlighted the importance of the O1 model, adding that it could rebuild a variety of industries and create new opportunities for AI startups.
Zhilin said that supplementary learning and scalability can be crucial for AI development. He spoke about scaling methods. This states that the performance of larger models with more training data is improving.
“This approach pushes the AI capabilities upper limit,” Zhilin said, adding that Openai O1 could disrupt the sector and create new opportunities for startups.
Openai also emphasizes the ability of models to solve complex problems, saying it works in a similar way to human thinking. By improving strategies and learning from mistakes, the model improves problem-solving capabilities.
Zhilin said that companies with sufficient computing capabilities can innovate not only with algorithms but with basic AI models. He considers this to be crucial, as AI engineers are increasingly relying on augmented learning to generate new data after exhausting available organic data sources.
Stepfun CEO Jiang Daxin agreed to Zhilin, but said computing power is a major challenge for many startups, particularly due to US trade restrictions that hinder Chinese companies’ access to sophisticated semiconductors.
“The calculation requirements are still substantial,” Daxin said.
Baichuan AI insiders say only a small group of Chinese AI startups, including Moonshot AI, Baichuan AI, Zhipu AI and Minimax, are in a position to make large investments in reinforcement learning. Collectively known as “AI Tigers,” these companies are deeply involved in LLM development and are promoting the next generation of AI.
Details of the Apsara meeting
Also at the conference, Alibaba Cloud made several announcements, including the release of the Qwen 2.5 model family, featuring advances in coding and mathematics. The model ranges from 5 billion to 72 billion parameters and supports around 29 languages, including Chinese, English, French and Spanish.
Special models such as the QWEN2.5-CODER and QWEN2.5-MATH have already gained some traction, with over 40 million downloads on the platform holding faces and model copes.
Alibaba Cloud was added to its product portfolio and provided image generator Tongyi Wanxiang with a text-to-video model. This model allows you to create videos in a realistic, animated style that could be used in advertising and filmmaking.
Alibaba Cloud has announced the latest version of the Vision Language model, the Qwen 2-VL. It handles videos for over 20 minutes, supports video-based questions, and is optimized for mobile devices and robotics.
For more information about the meeting, click here.
(Photo: @guy_ai_wise via x)
Want to learn more about AI and big data from industry leaders? Check out the AI & Big Data Expo in Amsterdam, California and London. The comprehensive event will be held in collaboration with other major events, including the Intelligent Automation Conference, Blockx, Digital Transformation Week, and Cyber Security & Cloud Expo.
Check out other upcoming Enterprise Technology events and webinars with TechForge here.