a half times more than gpt 3.5, on which chad gpt is based and 13 times more than the fallen fudai university neural network, another question, the researchers talk about some difficulties with the date set, that is, with typing in chinese, in which this neural network needs to be trained, this and the knowledge base. in the sense of facts and the basis for understanding the machine and the intricacies of the structure of the language is what allows large language models to conduct a dialogue. with a person in a way that is understandable to him, and here they name two problems, firstly, the chinese language construction is significantly different from english and, accordingly, use the developments of western researchers. not so easy. well, and secondly, they say, but the insufficient number of chinese texts in the public domain, including the restrictions associated with the publication of what goes against the accepted rules and norms for understanding in the gpt3 training dataset. hundreds of billions of english-language texts included books, magazines, websites, wikipedia in chinese, as t