(Yicai) March 7 -- It is premature for artificial intelligence to design fully personalized travel itineraries, industry insiders said after Chinese flight and hotel booking platforms Tongcheng Travel and Fliggy announced integration with DeepSeek's ChatGPT-like large language model.

Although LLMs can respond quickly, they struggle to fully comprehend the nuances of tourist experiences and personalized needs, similar to how AI-produced videos may contain logical errors, Wei Changren, founder of tourism consultancy Ctcnn.Com, explained to Yicai.

The application of LLMs in the tourism industry remains in its nascent stages. Alibaba Group Holding's flight booking platform Fliggy acknowledged that AI can hallucinate, but noted that response accuracy can be enhanced by incorporating extensive user experience insights and supply chain information.

Hangzhou-based AI startup DeepSeek launched DeepSeek R1, a powerful and cost-effective alternative to OpenAI's ChatGPT, in January. This open-source model is transforming several industries, including tourism, which would otherwise face challenges developing in-house LLMs due to high costs.

On Feb. 28, Suzhou-based Tongcheng became China's first online travel agency to announce integration with DeepSeek R1, enhancing its homegrown LLM Chengxin. Moreover, Fliggy introduced its smart itinerary assistant this week, integrating DeepSeek R1 and Alibaba Cloud's LLM Tongyi Qianwen to generate personalized travel plans. Even hotel management giant H World Group is reportedly upgrading its systems and considering collaboration with DeepSeek.

DeepSeek is driving a new wave of advanced trip planners, but smart itinerary assistants have been trending for a while. Trip.Com launched its LLM Wendao as early as 2023 to cater to users' diverse needs.

After the first hype of general-use LLMs, it will take time to develop industry-specific models. Vertical models specialized in tourism continue to face challenges with ambiguous requests and niche services, Tongcheng Chief Marketing Officer Bai Zhiwei said, adding that combining Tongcheng's Chengxin with DeepSeek R1 is expected to improve the accuracy of trip recommendations.

Fliggy emphasized that developing vertical LLMs requires deep adaptation of business logic and thorough integration of industry norms and regulations.

Editors: Xu Wei, Emmi Laine