In late November 2022, the AI service ChatGPT was launched for trial use, attracting 1 million users in just one week. Some even suggested it could replace Google Search. ChatGPT is a chatbot service based on the GPT-3 language model. It can understand multiple languages, including Chinese, and provide unexpected answers, rekindling many people's interest in artificial intelligence.
Taking medical applications as an example, ChatGPT allows users to ask questions about complex illnesses through a realistic conversational experience. Unlike reading Wikipedia or other content websites where you have to search through long articles for relevant information, or searching on Google where you have to browse and judge multiple search results, ChatGPT provides an experience like a caring personal health assistant, allowing you to obtain the information you need using natural language.
In 2017, when artificial intelligence was just beginning to take off, research institutions predicted that natural language chatbots would be among the top three business opportunities in medical AI. In addition to personal health and medical assistants for consumers, ChatGPT's conversational technology can also be used in emerging medical services such as telemedicine, giving medical institutions the opportunity to provide realistic care services.
ChatGPT originates from GPT (Generative Pre-trained Transformer), a general pre-training technique proposed by OpenAI in 2018. It eliminates the need for extensive manual data labeling, a task requiring significant expertise. Its strength lies in its ability to build a general model from unsupervised (unlabeled) data and then fine-tune it using supervised (labeled) data. The first-generation GPT technique had approximately 100 million parameters, requiring about 5GB of data; while the new generation of GPT technology boasts 175 billion parameters, requiring 45TB of data.
This technology requires massive amounts of data and enormous computing power; using conventional computing systems, it might take decades. This means that future artificial intelligence will be data-intensive and capital-intensive innovations, making it virtually impossible to build a universal model single-handedly.
While ChatGPT is powerful, it also requires the integration of more suitable local Taiwanese language and medical data to realize its medical applications. Excitingly, in 2018, the Ministry of Science and Technology built an AI supercomputer ranked 20th globally. If the strengths of academia and the medical community can be integrated, Taiwan has a great opportunity to establish its own universal medical model in the new field of large language models, leading innovation in precision health services in the post-pandemic era!
Source: Financial News Issue 675