Unique Research · Company / funding · June 2, 2023
On May 22, 2023, multimodal LLM developer Zhizi Engine (智子引擎) closed a ¥10M angel round. CEO Gao Yizhao is a Renmin University PhD under advisor Lu Zhiwu (also Softcom Power chief AI scientist and Zhizi advisor).
Lu and Gao have worked multimodal LMs since 2020. Before ChatGPT, BAAI’s Wudao gathered four forces (Tsinghua Tang Jie / Liu Zhiyuan / Huang Minlie + Renmin Wen Jirong). Wen’s Gaoling AI lab led multimodal “Wenlan”; Lu was a core modeller, Gao on data/train/eval. After Wudao, Tang/Liu/Huang spun out companies — Renmin’s entry completed the “four kings” entrepreneurship arc. Lu’s team was among China’s earliest multimodal LM groups with strong results.
Product: application-grade multimodal ChatGPT “Yuan Chengxiang ChatImg” (Mar 8, 2023) — ~tens-of-billions parameters; trained on image–text pairs + VQA for match/retrieval/caption/text-gen. Input image or text → chat, stories, ads. Wenlan 2.0 (BriVL-2, Jun 2021): weak image–text correlation assumption + DeepSpeed multimodal train; 6.5B weakly correlated pairs → 5B params — then largest Chinese general image–text pretrained model.
Strategy: general multimodal generative model “Yuan Chengxiang” — multimodal in/out. ChatImg is the dialog slice; team already has text-to-image, text-to-video, multimodal search to fold in. ChatImg 2.0 launch with Softcom Power / Softcom Wisdom / Fujian AI Computing Center for industrial landing.
— END —