- The newly published dialogue covers an October 16 conference in Shanghai.
- Speakers describe technical ambitions and company strategy, without announcing a model release.
Learning beyond a context window
Chinese AI founders set out priorities for continuous learning and enterprise model training in an edited dialogue published on 36Kr on October 28. The discussion took place at the AI Cloud 100 China conference in Shanghai on October 16; today's publication is a report of those remarks, rather than a new product launch.
Zhang Muhan, founder of Wuya Zhiyuan and an assistant professor at Peking University's Institute for Artificial Intelligence, argued that longer context windows cannot alone solve continuous learning. His stated goal is to let models absorb new knowledge and experience through changes to their parameters. These are research ambitions, not independently verified performance results.
Enterprise training strategy
Mindverse founder Chen Kaijie said the company began training models in late 2023 to strengthen its own applications. He described turning its accumulated post-training capabilities and infrastructure into a commercial platform as more enterprises sought their own models.
Chen linked that demand to domain performance, cost and retaining experience in company-trained models. The dialogue presents the speakers' strategic views; it provides no new model benchmark or release timetable.
Sources & context
Go to the original material. Company claims remain attributed to their sources.
01
