The recent meteoric rise of generative AI technologies like ChatGPT has left an indelible mark on the global stage. Acknowledging the far-reaching significance of these advancements, Wu Tsung-tsong?—Minister without Portfolio and Minister of the National Science and Technology Council (NSTC), Executive Yuan?—unveiled plans the other day for Taiwan to develop a trustworthy generative AI dialogue engine. The ambitious endeavor aims to integrate the collective expertise of domestic academia, research institutions and the industrial sector. To showcase the project's initial outcomes, the NSTC held the "TAIDE Project Phase One Results Press Conference" today (June 14), providing an overview of the project's structure and direction. The "Trustworthy AI Dialogue Engine" (TAIDE) 7B model took center stage at the event.
AI Chatbot Tailored for Taiwan
Minister Wu highlighted the untapped potential of generative AI and its proven value when applied to a diversity of fields. In an effort to adhere to ethical guidelines, he hopes to harness its capabilities while ensuring the content generated by the model aligns with the linguistic nuances of the majority of its Taiwanese users. Therefore, TAIDE will seek to embody Taiwanese culture and incorporate local language, cultural values, and customs. This ensures that the AI engine comprehends and responds to the unique requirements of its users. Additionally, the gradual introduction of reliable Traditional Chinese texts from various domains will further enhance TAIDE's adaptability and performance in those specific domains.
Collaboration Across Multiple Agencies
The NSTC has assembled a team of esteemed Taiwanese professors specializing in AI-related disciplines to spearhead the development of the core models and language learning technology. Moreover, the NSTC oversees the operations of affiliated entities such as: the National Center for High-performance Computing (NCHC), which is responsible for building the next-generation supercomputers to enhance computational resources and provide application service platforms; and the Science and Technology Policy Research and Information Center (STPI), which oversees the collection and processing of training data. In addition, the Ministry of Digital Affairs (MODA) is dedicated to crafting testing environment architecture and conducting evaluations of the project models to ensure the trustworthiness of the models.
TAIDE's Progress and Phase One 7B Model Performance
The 7B model is TAIDE’s maiden medium-sized model. Based on the LLaMA prototype, the model underwent further refinement and fine-tuning by the experts mentioned earlier. Despite the project only being in its nascent stages and constrained by limited data, the TAIDE 7B has exhibited exceptional performance in automatic summarization, text translation, letter writing, and article composition in real-world testing. It has demonstrated notable consistency and accuracy in how it handles and responds with Traditional Chinese.
Minister Wu acknowledged that the legal acquisition and organization of high-quality data in Taiwan will require time. Future plans for the project involve the introduction of larger models and an expansion of its functionalities. Anticipated next steps for the coming six months include demonstrations of specific tasks and leveraging the medium-sized model in collaboration with TAIDE partners to develop additional features. By the end of the year, the large model (13B) and its ability to block inappropriate responses will be unveiled. The model will then be made available to TAIDE partners.
Public-Private Collaboration to Foster Development and Application of AI
Minister Wu underscored TAIDE’s mission to provide a platform for government bodies and enterprises to continue developing proprietary internal application systems and value-added services. This approach allows organizations to choose models that align with their specific needs, while also factoring in the required computational resources and associated costs to utilize their value-added services, thereby accelerating Taiwan’s diverse sectors on their journey towards self-sufficiency. The NSTC stated that in addition to collaboration between the academic and research institutions, early involvement and support from industry stakeholders plays a pivotal role in subsequent technological transitions and the advancement of the reach of AI technology. The press conference witnessed the participation of information service providers, financial institutions, and startups, all of whom demonstrated keen interest in leveraging the TAIDE base model. They also shared insights regarding their requirements for generative AI requirements and their current capabilities in this domain.
Minister Wu emphasized that talent cultivation is key to ensuring the continued development of technology and boosting competitiveness. The TAIDE project has already established a successful framework for cooperation and talent development. The NSTC remains committed to supporting AI research projects, fostering exceptional scientific teams, and nurturing professionals with innovative thinking and practical expertise. NSTC will also uphold scientific ethics and social responsibility and collaborate with relevant entities to establish norms and regulatory mechanisms, ensuring the fair, transparent, and reliable utilization of AI technologies in compliance with regulations.