(Source: CCTV13 2025-02-20)

Recently, a set of data on the mortality rate of those born in the 1980s has attracted widespread attention on the Internet.CCTV reporters found that there are many articles on the Internet citing this data about the mortality rate of the post-80s generation, and many self-media articles are constantly exaggerating it, such as: "The 5.2% mortality rate of the post-80s generation ushered in the sad departure of the golden generation"; another example: "When the post-80s generation began to collectively call a curtain call - behind the 5.2% mortality rate, the survival dilemma of a generation and the sorrow of the times." There are even pictures on the Internet that make a histogram of the mortality rate of the so-called post-80s generation, saying: "One in every 20 people has passed away", and labeling comparative data, saying that "the mortality rate of the post-70s generation is lower than that of the younger post-80s generation." Is such shocking data true?

The source of the wrong mortality rate may be "AI operation bias"

Where did this sensational and flawed mortality data come from? Li Ting, a professor at the w88 casino of Population and Health of Renmin University of China, found out that it is likely that the large AI model made errors in the question answering process. Professor Li Ting entered the question "What are the mortality rates of these generations in the 50s, 60s, 70s, and 80s" into the large AI model. The large model showed that based on network information, it was concluded that "there are 212 million people born in the 1980s, with a survival rate of 94.8% and a mortality rate of 5.2%."

Illusion problem and generation mechanism for large models,Professor Zhang Xin of Law w88 casinoAccepted a telephone interview with CCTV13 News Channel to analyze the operating mechanism of large models, the reasons for the occurrence of large model illusions in the data layer, model layer and application layer, my country's current network information governance ideas, and possible legal liabilities when intentional dissemination of false information causes adverse effects.

In view of the technical causes of the "illusion" of large models,Zhang Xinpointed out that the data layer, model layer and application layer may cause the "illusion" of a large model. At the data level, timeliness deviations and quality defects in training corpus are the primary causes. In particular, false information hidden in web crawled data can easily form systematic cognitive biases after model parameterization. At the algorithm level, she pointed out that the autoregressive generation characteristics and probabilistic sampling mechanism of the Transformer architecture objectively lead to the emergence of content illusion. At the application layer, insufficient prompts and scene adaptation in the human-computer interaction process, as well as the multi-modal trend of the model, may increase the possibility of the model producing hallucinations.

my country has previously issued a series of departmental regulations represented by the "Regulations on the Management of Deep Synthesis of Internet Information Services" and the "Interim Measures for the Management of Generative Artificial Intelligence Services". Recently, it has also been working hard to establish and improve methods and standards for identifying artificial intelligence-generated synthetic content to provide a clear compliance framework for generative synthesis service providers and content dissemination service providers. If false and harmful information is deliberately fabricated and disseminated through the Internet to cause public panic, disrupt public order, and cause adverse social impact, you may also bear corresponding criminal liability.

The reporter found that after fermentation on the Internet, the death rate has become a gimmick for some self-media to promote sales.It can be seen that although large AI models have super data storage and computing capabilities, due to insufficient training corpus and limited data sources, it may lead to erroneous calculations. In order to monetize traffic, some self-media will also take quotes out of context and use the topic to promote the hype and spread of these outrageous data.

Attachment: Original link

https://content-static.cctvnews.cctv.com/snow-book/index.html?item_id=2084160556267115271&t =1740056794180&toc_style_id=feeds_default&share_to=wechat&track_id=23ae2058-645c-47f2-b968-4321ae119902