2023-11-27
With the release of ChatGPT, the concept of large language models (LLMs) has entered the public eye. In just one year, major companies have launched their own large language models. What impact will the development of large language models have on the tourism industry? How do we understand the opportunities and challenges of the era of large language models? Let’s delve into the perspectives of our guests through this tech dialogue.
Moderator: JIANG Lieyi, CEO of Future Plus Technologies (Beijing) Co., Ltd.
Panelists:
WANG Qing, Vice President of Technology of Trip.com Group
NI Shenghua, Vice President and Chief Technology Officer of Fliggy
PENG Lun, General Manager of Travel, China, Google
WANG Yuan, Vice President of NetEase and Executive Director of Hangzhou Research Institute
XIN Yian, Chairman of Beijing Quanshi World Online Network Information Co., Ltd.
Dialogue Transcript:
Moderator: What impact will large language models have on tourism industry?
WANG Qing: The most important feature of large language models is their “large size”. Models have been applied in various scenarios for a long time. With the emergence of ChatGPT, it’s realized that models can be massively scaled up, trained with larger datasets, leading to a surge in intelligence. The leap in model capabilities impacts various industries, including tourism. It greatly enhanced intelligent applications in various aspects, improving quality while controlling costs.
NI Shenghua: The tourism industry is labor and cost-intensive, relying heavily on knowledge accumulation. It is a high barrier to leverage professional knowledge in tourism and tourist attractions to form rational travel plan. Large language models make it possible to handle order changes and refunds, client consultations, etc., with lower costs, aiding in business decision-making.
PENG Lun: The core impact of AI large language models is the enhanced speed and efficiency of information dissemination. AI stimulates and promotes the tourism industry through its ability to capture user needs accurately using big data, improving travel planning. It significantly enhances the speed of information dissemination and efficiency in capturing diverse tourist needs.
WANG Yuan: I believe the impact of large language models on the tourism industry is twofold: First, they enhance our ability to access language services, allowing us to go to unfamiliar places. Second, they provide better personalized guidance, helping us discover less popular but more personally suited destinations. Ultimately, this could lead to a more dispersed and balanced distribution of the tourism industry.
XIN Yian: Tourism is merging with culture, sports, and various other aspects. AI’s impact on tourism isn’t isolated but should interconnect with related industries. For consumers, AI applications should be seamless, even simplistic, to truly integrate large language models into the industry and unleash their significant commercial value.
Moderator: How can large language models help the tourism industry provide personalized experiences? What kind of progress are you seeing?
WANG Yuan: Personalization can be viewed from two perspectives: from the user’s perspective, the most significant personalization occurs in choosing destinations, itineraries, and routes; from the operator’s perspective, it is about considering how to offer personalized experiences. Due to physical space limitations, each tourist spot can only present limited information. Large language models can step in timely in aspects like mining stories behind attractions based on user interests, and offering varied experiences.
Moderator: In addition to creating travel destination products, digital marketing plays a crucial role. Mr. Xin, could you share your thoughts on this?
XIN Yian: Large language models have the most significant impact on the marketing industry. With short video streaming, a professional used to edit 4-5 videos per day. Now, with AI tools, hundreds can be edited in a minute, drastically reducing costs. For web series or TV dramas, compared with the traditional method involving watching and selecting exciting scenes by humans, AI tools automatically capture and edit scenes from an entire series, creating hundreds or even thousands of clips. The current strategy is quantity-driven, and AI enables the creation of countless advertising accounts for clients, testing vast amounts of material to find the most impactful content. The influence of AIGC and AI tools is essentially about cost reduction and efficiency enhancement.
To develop large language models, we’ve launched a digital assistant, or “digital employee”, in partnership with Xiong Hua. This AI assistant is integrated into enterprise WeChat, websites, and mini-programs, serving as a primary application for businesses. The product is already put online.
Digital marketing requires significant operation and maintenance. Explaining usage to B2B clients takes time and team effort. So, even as large language models become better and more specialized, a substantial number of service providers and maintenance staff are needed to effectively train company information and data, helping businesses reduce costs and increase efficiency.
Moderator: What are Fliggy’s plans and practices in this area?
NI Shenghua: First, large language models provide a better way of interaction, allowing users to clearly express their demands. Traditional recommendations are a one-time try, guessing what you like. But tourism demands are diverse, so this new type of interaction offers possibilities for consumers to freely express their needs. The challenges of large language models include ensuring the timeliness, as the route for visiting the same spot varies with time; addressing the precision of space; providing the playbook at each spot; matching large language models with the commercial resources of merchants and tourism practitioners; and lastly, addressing personalized consumer needs.
We’ve categorized the issues into two sides. In the consumer dimension, we should tell whether consumers want to travel independently or need tourism practitioners to help complete their travel plans. In the B2B dimension, we should decide whether to develop product lines first, or focus on offering customization on consumer demands. Based on these dimensions, Fliggy has formulated a plan: The first phase addresses consumers traveling independently; the second phase makes existing routes better fit consumer demands, internally termed as “Making Routes More Vivid and Flexible”; the third phase combines tourism practitioners’ resources with consumer demands to freely assemble differentiated products. The function of the large language models is to bring decision-making closer to consumers. For practitioners, the hope is that large language models can empower the industry ecosystem, allowing every practitioner to enjoy the benefits of technology.
Moderator: Looking at the technological trends, what are your visions for the future of large language models?
WANG Yuan: In pre-travel planning, large language models can play a significant role, mainly due to two capabilities: understanding vast input content and correcting errors in real-time through dialogue, significantly improving over existing technologies. Secondly, their problem-solving abilities are more flexible and robust. We should combine these strengths to enhance pre-travel services. Additionally, actions taken before travel, like booking, pose greater challenges for large language models, requiring high accuracy and credibility, along with the flexibility to handle information changes. I think the major part for large language models to play is in pre-travel stages, especially in choosing destinations and arranging itineraries and bookings.
Moderator: In terms of cross-cultural barriers and challenges, what assistance can large language models provide? Are there any successful cases or practices you can share?
PENG Lun: Before the rise of large language models, AI had already been strategically deployed. Google focuses on two aspects: First, for Chinese tourists and businesses going abroad, we concentrate on tools like navigation, such as Google HD Maps, for precise guidance, and translation, for quick language communication in European countries. Second, for domestic travel agencies promoting culture abroad, like intangible cultural heritage, Google has a dedicated art and culture product that guides users to understand Chinese culture and tell China’s story. AI and large language models provide complete solutions for precise targeting.
Post-pandemic, overseas tourists come to China mainly for two purposes: personal or group tours, and business travel, including visiting various exhibitions. Domestic and overseas businesses collaborating require AI large language models to understand how to precisely reach these users, informing them of the best travel methods, attraction choices, and exhibition information in China, providing entry solutions.
Moderator: What practical applications can Xiecheng Wendao provide in terms of cross-language and cross-cultural aspects?
WANG Qing: Cultural exchange is a crucial part of travel, and overcoming multiple language barriers is essential. A key ability of large language models is multilingual communication. For example, ChatGPT has excellent capabilities in English and other languages, facilitating cross-language communication and cross-border travel. Using large language models, we developed a tourist guide project for foreign tourists entering China, enabling multilingual interaction, with fairly good accuracy. Essentially, AI large language models possess cross-language communication skills, supporting us in better understanding the cultures behind languages, enhancing the travel experience.
Moderator: Amidst the opportunities and challenges, how do you view the prospects of large language models in the cultural and tourism industry?
WANG Qing: Large language models are still in the early days, and they are evolving rapidly. A clear benefit is enhanced natural language interaction, improving the efficiency when searching for products and information. Multilingual capabilities enhance the experience for outbound and inbound travelers.
NI Shenghua: Don’t underestimate the development speed of large language models. In just six months, they have progressed far beyond traditional technologies. Tourism practitioners need to consider how to restructure their organizations and work methods through technology. It’s certain that large language models will significantly reduce industry cost structures.
PENG Lun: Opportunities far outweigh challenges. The key is to maintain the best control over compliance risks. In niche large language models, as long as fine-tuning is done well, datasets are properly configured, and compatibility with databases is ensured, the application of large language models is very promising.
WANG Yuan: There is significant room for large language models to excel in single tasks. The greatest challenge lies in whether these models can reach their limits when tasked with complex assignments composed of multiple tasks.
XIN Yian: Large language models and AI are not just for leading companies; they should find use across the entire upstream and downstream ecosystem. Tech companies should share their AI technology or products, applying them throughout the ecosystem to enhance overall efficiency.