ChainCatcher, citing Beating AI, reports that an interview summary circulating in the investment community states that Zhipu's next-generation foundation model may have 2500000000000 to 3500000000000 parameters, and that its training issues may have been resolved.
Reports say that several Chinese large-model companies are strengthening data governance because of training-data quality issues. Tencent Hunyuan has rebuilt its pretraining and reinforcement-learning infrastructure, while Alibaba and Baidu have also upgraded their cleaning processes.
