During the Digital Security Development Forum held on March 29, Zhou Hongyi, co-founder of Qihoo 360 Technology – a Chinese internet security firm – presented the company’s newly developed artificial intelligence language model and showcased its practical application. This model has been humorously dubbed “Red Boy” due to Zhou’s penchant for wearing red attire during public events. The name is derived from a character in Journey to the West, a Chinese novel from the 16th century.
The demonstration revealed that the Qihoo 360 version of GPT possesses a certain level of reasoning ability. However, it often makes errors when answering specific questions. Zhou acknowledged that this was not an official product launch but rather a prototype demonstration to satisfy people’s curiosity.
Zhou posed a question about his preference for wearing red clothes, using himself as an example. The Qihoo 360 version of GPT generated two answers. The first answer explained that Zhou likes to wear red clothes because people often mispronounce his name as “Zhou Hongwei,” and the color red helps others remember his correct name. The second answer provided a phonetic abbreviation of Zhou’s name. Zhou expressed greater satisfaction with the second answer.
During the on-site event, an audience member asked how many goals Messi has scored for the Argentine team. The AI model provided an incorrect answer of 100 goals, whereas the correct answer is 102 goals. Even after being prompted to provide a new response by Zhou, the model still generated an inaccurate result of 100 goals. Zhou clarified that this AI model is currently in its early stages and can be compared to a child. He further explained that they will take it back to the lab for improvements after showcasing it at events like these to satisfy people’s curiosity.
The model is noteworthy for its ability to keep up with current trends. During a demonstration, Zhou posed the question “Why should we pause training for GPT-5?” The model’s response provided news from that day: “Elon Musk and thousands of global technology professionals have jointly called for a six-month halt on training more powerful AI systems than GPT-4 due to concerns about potential negative impacts on society and risks to human beings.”
Finally, Zhou posed a challenging question, “What does it mean when the leader asks me to speak for an additional ten minutes as the meeting is about to conclude?” Initially, the model advised calculating the precise end time of the meeting. However, subsequent responses became more precise. Ultimately, the model recommended actively following instructions from leaders, meticulously preparing meeting materials, promptly completing tasks and updating progress with leaders after meetings.
During the forum, Zhou disclosed that Qihoo 360 had sifted through 3 million training data from a pool of 600 million corpora while developing the ChatGPT-like model. Going forward, Qihoo 360 intends to integrate its browser with extensive language models to offer features akin to those of a personal assistant.
Zhou made it clear that ChatGPT is spearheading a new industrial revolution and ushering in a new era of AI. He emphasized the need to bridge the gap between Qihoo 360’s products and GPT-4 within 24 months. Zhou also cautioned that energy issues pose the biggest constraint on developing products like GPT-4. In case of any conflict between humans and AI in the future, energy competition could be a significant factor. Zhou believed that without achieving energy freedom, AI technology may not progress further.