early December 2022，American Artificial Intelligence Lab (Openai) released a natural language processing tool called ChatGPT。ChatGPT quickly attracted attention，It can "catch" a series of "tricky questions" proposed by netizens，For example, you can write a review in English、Generate papers in specific fields, etc.，Dialogue robot that is closest to human level so far。As an application featured as a characteristic in AI language，It represents the rise of the new type of information dissemination in the future of "human intention+machine combination content" in the future，AIGC (AI Generated Content，Artificial intelligence content production) Social risks are worth thinking about，and layout risk prevention and control measures in advance。

Bet365 app download  

ChatGPT via a WeChat dialog box，Answer the questions raised by the user，It is a more "like man" dialogue robot in the AIGC field，Its technical characteristics represent the basic technical characteristics and trend of AIGC。In the past few years，Openai research has achieved a lot of achievements，Among them, the natural language processing model GPT-3、Visual recognition model Dall-E and enhanced learning model AlphaGo，Chatgpt depends on the natural language processing model GPT-3。

GPT-3 model is a very powerful natural language generation model，It can be used to generate text、Answer questions、Translation text、Summary text、Text classification and other tasks，Based on the Transformer model architecture，A large bet365 Play online games amount of text data through pre -training，Repeat special task，To achieve efficient natural language processing ability。At present, it seems，GPT-3 represents the development direction of AIGC content generation，Its technical features are manifested in four aspects。One is a large scale，GPT-3 has tens of billions of parameters，So far the biggest natural language generation model，​​This constitutes the basis of accurate content。Second is automation，The model can automatically adjust the language style and content of generating text based on the content of the text，Improve the shortcomings of ordinary AIGC "the same" by adding personalized factors。Third is universal，GPT-3 can be used for a variety of natural language processing tasks，and perform well in these tasks。Fourth is scalability，You can learn new tasks by fine -tuning，and maintain higher performance with the expansion of the model，New information by adding new information，You can train the language model，To achieve the purpose of the user's own use。

These technical characteristics are integrated，Compared with other chat robots on the market，Chatgpt's function is more diverse，For example, it can write praise poems according to specific requirements、Write a speech、Press release and even write physical simulation procedures。other，The answer to the chat robot is more rigorous，Being able to take the initiative to admit your deficiencies，Facing many difficult problems，Will answer "Lack of Information" or "I'm just a language model"，Try to answer after that then。It can be said，Through purposeful setting conditions，AIGC can be a lot、Quick production content，Especially the conditional restrictions bet365 Play online games generated by refining content，It can produce content that is highly similar to human language。

AIGC's social communication risk 

AIGC's social communication risk problem mainly refers to it may be used for immoral，Even illegal information dissemination activities，Its core is to generate "it looks like that"，But it is essentially the content of false information。For details,，AIGC applications represented by ChatGPT have the following social communication risks。

Generate fake news。ChatGPT can be based on one or more keywords given，Generation looks real，Actually it is completely fabricated news。It can achieve all automated output from news title to content and even comments。For example，When a given keyword "American Leader" and "Diplomacy Crisis"，plus conditions limited within 500 words，ChatGPT will generate a news article，Report a diplomatic crisis in a certain country。

For information fraud。Since ChatGPT can produce high -quality human spoken text，It can be used for network deception，For example, the implementation of property fraud by writing the legal e -mail content，Policy documents through manufacturing fake、Government document to make information fraud。Its core is to automatically imitate the text content with a fixed mode according to the user's requirements。For example，You can generate a very "decent" government investment document。

Imitation of a specific discourse style for content generation。ChatGPT while obtaining a large number of network texts，You can also input artificial corpus，Post -training for artificial intelligence。Such as an Internet celebrity、Big V, etc. There are greater influential opinion leaders，Its language style is often very distinctive，ChatGPT can imitate specific language styles for content bet365 live casino games production，Its discourse structure、The characteristics of words are similar to the height of the real person，It is easy to mislead the audience。

Looking at it overall，AIGC applications have a high degree of intelligence，Pulling down the technical use threshold of artificial intelligence production。It can handle the discourse style、Words、Text types and other languages ​​and other languages ​​are required，It can be said that the more specific the conditioning conditions are set，The closer to the content of its production, the closer to the real。For false、Unprecedented information，Similar AIGC technology will directly reduce the technical requirements for making false information in batches，Accelerate the production speed of false content，further challenges to the future information content governance。

AIGC Social Communication Risk Coppering Strategy 

In recent years，With the rapid development of natural language processing (NLP) technology，Chatgpt and other high -level language models have made significant progress in language understanding and generation。，These technologies also bring risk of false information。should prevent AIGC from becoming the birthplace of false information from the following three aspects。

Implementation Content Responsibility System: Who generates，Who produces，Who is responsible。In terms of network space content governance，Clear responsibility system is the basis for building a clear network space。For a new generation of artificial intelligence production applications such as ChatGPT，It should continue the existing content responsibility system in my country，Continuing the "Administrative Measures for Internet Information Services", "Regulations bet365 Play online games on the Management of Internet Following Post Comment Service", etc.，Who generates it，Who produces，Who is responsible，Many parties jointly assume responsibility for governance。Model developers should be responsible for the risk of the model to generate false information。Platform provider and content publisher should review the information generated by ChatGPT，To prevent the spread of false information。The government should formulate regulatory regulations on network information dissemination，To ensure the authenticity and accuracy of the information，Punishment of false information providers。

AI AI: Establish an artificial intelligence generation content identification and disposal mechanism。Facing artificial intelligence content generation，Only AI can use AI's strategy to counter。Layout should be made in advance for the technical characteristics of different language models，continuously iterate artificial intelligence content monitoring system。At present, it seems，Chatgpt and other AIGC applications have some obvious limitations，Including but not limited to: grammar and semantic errors，When generating a text, there will be a syntax error and semantic inconsistencies。Publicization ability is limited，Weak's generalization ability when dealing with new tasks，It may not be able to accurately answer new questions or deal with new text data。Lack of logical reasoning ability，Can't get conclusions through reasoning，Mistakes may occur when dealing with complex problems。Dependent input text，It has a very strong dependence on the content of the input text。These problems largely represent the general defects of the production of artificial bet365 live casino games intelligence content，Repeatability of network content、logic、Grammar Symons and other aspects of development monitoring tools，Identify false information through analysis of language characteristics and context information，The method of using pre -training data for data driving to detect the generated information，Constructing an artificial intelligence generation content identification and disposal mechanism。

Increase the efforts to cultivate autonomous and controllable natural language generation models。At present, artificial intelligence language models on the market are not uncommon，Such as Google Bert model、Alibaba's plug model、Tsinghua University Computer Department Natural Language Treatment and Social Humanities Computing Laboratory "Nine Songs"，Can also use anthropomorphic methods to talk to users with free dialogue.。But compared to the above model that can generate language，ChatGPT has more powerful features，Behind it is the development logic of "stronger" in the field of artificial intelligence production。ChatGPT mainly optimizes the conversation of dialogue by using human feedback (RLHF)，This is a method of using human demonstration to guide the model to implement the required behavior，Its parameters have increased from the initial 117 million to 175 billion，Pre -training data volume increases from 5GB to 45TB，Among them, the cost of GPT-3 training is about $ 4.6 million，Total training cost reaches $ 12 million。Popular artificial intelligence language models continue to obtain information feedback in people's use，Continuous evolution，Finally let the competitors "discouraged"。This prompts us to strengthen the research and development investment and promotion application of autonomous and controllable natural language generation models，Finally, promote the development bet365 best casino games of Chinese artificial intelligence through market means。

(Author Unit: Institute of Journalism and Communication of the Chinese Academy of Social Sciences)