During a livestream on March 15, Zhou Hongyi, the founder of Qihoo 360, a Chinese internet security company, disclosed the company’s artificial intelligence development plan amidst the ChatGPT hype. The Beijing-based company intends to build a generative large language model and release related products and services at the earliest opportunity.
Zhou Hongyi suggested that in order to advance GPT technology in China, developers must prioritize identifying application scenarios and simultaneously develop the core algorithms. He used China’s J-10 fighter jet as a metaphor, explaining that since the R&D of the aircraft’s fuselage, avionics, and radar systems are highly complex, developers cannot wait for the engine to be completed before working on the rest of the aircraft. Similarly, GPT technology requires sophisticated engineering and commercialization capabilities, as well as data cleaning and manual annotation. If Chinese developers wait for algorithm development, they may miss out on opportunities.
Zhou Hongyi also announced that Qihoo 360 will provide services for individuals, small and medium-sized companies, local governments, and large enterprises in the Chinese market. The company aims to leverage a generative large language model to drive digital transformation across the country.
Qihoo 360’s plans for various customer segments include launching a new intelligent search engine similar to Microsoft’s new Bing for individual consumers, as well as artificial intelligence personal assistants based on their search requirements. The company intends to provide small and medium-sized companies with SaaS applications that rely on generative large language models, while integrating core digital security capabilities into the AI model for local governments and large enterprises. Additionally, Qihoo 360 plans to introduce private AI services to help customers index assets according to their custom needs.
Qihoo 360, as a company that offers internet services to millions of Chinese netizens, possesses significant advantages in big data, knowledge mapping, content comprehension, and other areas. It boasts vast arrays of algorithms that enable audio, video, image, and text comprehension, and its AI services are called upon several billion times per day to support its users.
In a report of Chinese entrepreneurs, Zhou expressed his belief that ChatGPT is about to usher in a new era of information revolution. He added that China’s previous AI research and development was too pragmatic, causing the opportunity to develop a general model like ChatGPT to be missed.
Following the industry buzz generated by ChatGPT, OpenAI has released its latest deep learning model, GPT-4, which the company dubs as “the latest milestone in its effort in scaling up deep learning”. Unlike its predecessor, GPT-3.5, which only accepts text input, GPT-4 can accept both image and text inputs. It performs at a “human level” on various professional and academic benchmarks, including passing a simulated bar exam with a score in the top 10% of test takers.
Despite the system messages and other enhancements, OpenAI acknowledges that GPT-4 is far from perfect. The model is prone to generating incorrect facts and makes errors in reasoning, sometimes with high confidence. OpenAI also mentioned that “GPT-4 generally lacks knowledge of events that have occurred after the vast majority of its data cuts off (September 2021) and does not learn from its experience.”