Tencent has recently undergone personnel adjustments within its large multimodal team, focused on its core large language model. Just weeks after establishing a dedicated department for foundational models, changes in leadership roles are taking place.
The former head of the multimodal division is expected to see their responsibilities change. Meanwhile, a senior researcher who recently left a major tech company’s AI division has rejoined Tencent, primarily working within the gaming sector. This researcher had previously contributed to developing vision-language models and now reports directly to the chief scientist leading the AI efforts.
Additionally, a Ph.D. graduate from a top U.S. university, who previously collaborated with the chief scientist at another major AI research organization, joined the team last month. This individual specializes in text-to-video and text-to-image algorithms. The company has indicated that this new team member might assume significant leadership responsibilities moving forward, especially in multimodal projects.
The company combined its large language model and multimodal model teams into a single department led by the chief AI scientist. Despite these organizational updates, progress in the multimodal space has been somewhat limited. The company released an image generation model in September and a video generation model in November, but has yet to debut a new multimodal model since then.
The most recent video generation model boasts 8.3 billion parameters but does not support fully synchronized audio and video creation within one process. This means it cannot generate videos with matching sound and visuals seamlessly based only on a single prompt, and lip-syncing issues remain unresolved.
Competitors like Alibaba, ByteDance, and Kling AI have achieved notable breakthroughs this year, creating tools capable of producing synchronized videos and advancing multimodal reasoning abilities. For instance, ByteDance has released successive updates to its Seedance models, showcasing progress in this competitive landscape.