At 14 o'clock on April 23,Yang Yiming, Professor of the School of Language Science and Art of Jiangsu Normal UniversityAs a guest 367th "New Apricot",Academic lectures entitled "Different Language Models and Human Natural Languages" in the A606 of Yuzhixinlou A606。The lecture was chaired by Professor Yue Lijing,Professor Wang Hui、Professor Kou Xin、Associate Professor Liu Wen、Associate Professor Zhang Yanfen attended,attracted nearly a hundred teachers and students to come to listen。
At the beginning of the lecture,Professor Yue Lijing briefly introduced Professor Yang Yiming's research direction and main academic achievements,Speaking and Professor Yang Yiming involved widely,Research on Yin Rhymes in the early days,The "Perspective Separation method" it created is known as "a new method of modern sound research"。In recent years, it has mainly devoted itself to the study of neurological bet365 live casino bettingBet365 gameslinguistics,It is the main pioneer and founder in the field of neurological research。
First,Professor Yang Yiming reviews the evolution process of the language research object。He pointed out that every change laid a solid foundation for the birth and development of big language models。With the rapid progress of technology,The development of big language models is also about the same daily。Professor Yang specifically pointed out,The parameters of the large language model have achieved significant growth,The continuous integration of new technologies makes the function of language models more powerful。At the same time,The concept of technical sharing and opening up further promotes the extensive landing of large language models in practical applications。Then he systematically sorted out the development process of large language models,and divide it into three stages for detailed interpretation,In -depth analysis of the functions and characteristics of the language model in processing language。
Next,Professor Yang Yiming passed a series of in -depth experimental research,Significant differences between large language models and human languages in detail。He mainly conducts comparative analysis Bet365 gamesbet365 live casino bettingfrom two aspects: language acquisition mechanism and language processing mechanism。When exploring the language acquisition mechanism,Professor Yang Yiming pointed out,Language learning of large language model,That is its training process,It depends entirely on language data and languageFeatures,The imitation process similar to the "parrot learning tongue"。Relatively,Human language habits have an inherent genetic mechanism,Make us only need a small amount of language stimulation to master and develop the corresponding language ability。At the level of language processing,The difference between the big language model and humans is reflected in the input of information、Processing method、Information understanding and content output and other dimensions。Professor Yang Yiming emphasized,Large language models can imitate human language behavior,But due to the lack of human language biological mechanisms,It does not really understand the deep meaning of language,Therefore, there is a essential difference between innovative and human language。
Last,Professor Yang Yiming is sure that the big language model is generated in text、Translation、Powerful Bet365 gamesBet365 Casino loginfeatures in the fields of programming and creation,But also pointed out its existing problems。For example,The preferences of big data model for big data have led to high resource consumption。At the same time,Because its training data mainly comes from the open Internet,A large number of uncomfortable verifications contained in them、false or even wrong information,Even in the data pre -processing stage,It is also difficult to ensure its complete compliance。other,The internal organizational structure of the large language model still needs to be further improved,and it lacks self -supervision mechanism,Sometimes an unrealistic "hallucination",A serious nonsense。For these problems,Professor Yang Yiming proposed two improvement paths: First, to promote the upgrading of AI technology based on language processing and language cognition;The second is to build a "language brain" based on the concept of "pre -installation"。These cutting -edge viewpoints indicate the direction of future big language model development。
In the interactive link,On -site teachers and classmates are conceptualization of psychological process、Baby -oriented language and other questions, ask Bet365 gamesBet365 gamesProfessor Yang Yiming,Professor Yang Yiming answered patiently and carefully。
Professor Yue Lijing expressed his gratitude to Professor Yang Yiming's wonderful speech on behalf of the Teachers and students of the School of Arts。Professor Yue Lijing pointed out,Professor Yang’s lecture not only broaden our field of vision,The vast space for us to show us linguistic research,Remind us,Although the big language model is performing well in some ways,But there are still many jobs that need to be completed。Therefore,While we cherish and use the convenience brought by the big language model,Should maintain independent thinking and creativity。The lecture ended successfully in warm applause。