Associate Professor of Tsinghua University: ChatGPT cannot realize non-0 or 1 accurate calculation, and there are security problems–Fast Technology–Technology Changes the Future

Fast Technology learned on April 8 that the main forum of the “Artificial Intelligence Large Model Technology Summit Forum” sponsored by the Chinese Association for Artificial Intelligence opened in Xiaoshan, Hangzhou. At the forum, Huang Minlie, tenured associate professor of Tsinghua University, founder of Lingxin Intelligent, and national outstanding youth winner, delivered a speech.

Huang Minlie said that there are deficiencies and defects in ChatGPT, for example,Mathematical calculations and symbolic reasoning cannot be realized, specifically manifested as the inability to realize precise calculations that are either 0 or 1.

Since ChatGPT is essentially a language model, its ability to embody comes from fitting the training data,There is no explicit logical reasoning and judgment in its operating mechanism.There are issues with accuracy, security, etc.

The mayor of Hepburn, west of Melbourne, Australia, has accused ChatGPT of defaming him after the chatbot falsely claimed he was a guilty party to certain scandals when it answered questions.

In addition, many users who have experienced it before complained that ChatGPT will forcibly fabricate an answer when encountering a question that cannot be retrieved. It seems that the grammar is smooth, but the information is all fabricated.