A photograph of a slide from Qihoo 360’s recent launch event for their latest large language model (LLM) went viral on Weibo, demonstrating the model’s internal censorship mechanisms. The photograph depicted a flowchart on how Qihoo 360 filters sensitive words in user inputs, chat disconnects if sensitive words were detected, and the model updated every 10 minutes. The Chinese security software also shares blacklists of sensitive words with public security bureau, further monitored by censors. The Qihoo 360 slide was similar to a leaked internal flowchart revealing how Xiaohongshu censors sudden incidents.
The Chinese government has previously shown concerns on the control of information generated by LLMs. Qihoo 360’s LLM passed the Ministry of Industry and Information Technology’s security review, being the first model to do so. Qihoo 360 is a widespread security software in the Chinese PC industry. The slide, however, did not mention what information Qihoo 360 deems sensitive, but reports suggest that China’s generative AI chatbots are programmed not to answer certain types of questions.
It must be noted that the photograph was removed by Weibo’s censors, raising concerns about censorship by Chinese companies. The chatbots’ ability to restrict access to certain keywords and phrases, as well as the practice of sharing the sensitive word blacklists with public security agencies, sparked debates on online censorship in China.
The use of LLM’s in China has been increasing in recent years, with several tech giants investing in the research and development of these models. However, the Chinese government has also implemented stricter regulations on their use, particularly in terms of data privacy and information control.
Overall, the photograph has given insight into how Qihoo 360 operates their LLM’s internal censorship mechanisms. It has also raised concerns about censorship practices and the use of chatbots in online communications.
Frequently Asked Questions (FAQs) Related to the Above News
What is Qihoo 360?
Qihoo 360 is a Chinese security software company, widely used in the Chinese PC industry.
What does the leaked slide from Qihoo 360 demonstrate?
The slide demonstrated the company's internal censorship mechanisms for its latest large language model (LLM), including filtering sensitive words in user inputs and disconnection if sensitive words are detected. The model is also updated every 10 minutes and shares blacklists of sensitive words with public security bureaus.
Has Qihoo 360's LLM passed security reviews in China?
Yes, Qihoo 360's LLM has passed the Ministry of Industry and Information Technology's security review, making it the first LLM to do so.
What concerns has the Chinese government expressed regarding LLMs?
The Chinese government has expressed concerns about the control of information generated by LLMs.
What do reports suggest about China's generative AI chatbots?
Reports suggest that China's generative AI chatbots are programmed not to answer certain types of questions.
Why was the photograph of the Qihoo 360 slide removed by Weibo censors?
The photograph was removed by Weibo censors, raising concerns about censorship by Chinese companies.
What debates has the Qihoo 360 slide sparked about censorship practices?
The Qihoo 360 slide has sparked debates on online censorship in China and the chatbots' ability to restrict access to certain keywords and phrases, as well as the practice of sharing the sensitive word blacklists with public security agencies.
How are LLMs being used in China?
LLMs are being increasingly used in China, with several tech giants investing in their research and development, although the Chinese government has implemented stricter regulations on their use, particularly in terms of data privacy and information control.
Please note that the FAQs provided on this page are based on the news article published. While we strive to provide accurate and up-to-date information, it is always recommended to consult relevant authorities or professionals before making any decisions or taking action based on the FAQs or the news article.
