登錄
選單
登錄
海擇短評 Haize Comment：
近期日本LINE Yahoo公告已提供"AI旅遊線路模型"功能，當用戶在Yahoo!搜尋引擎的入口搜尋"地名+觀光"時，會觸發生成式AI的旅遊模型。具體來說，與Yahoo!綁定的生成式AI目前提供概述、推薦景點、示範線路、活動列表、主題地圖等五個功能，當用戶觸發模型時，生成式AI會根據這五個功能提供對應內容。目前市場上以生成式AI提供旅遊行程規畫的產品很多，大體可以分為大語言模型本身、OTA的聊天機器人及新創公司的相關工具三類；而Line Yahoo的行程規劃工具比較特殊，是以海擇資本做簡單的評測，目的是理解功能良窳，並推估成熟型態的樣貌。謹將評測的用戶體驗部分說明於下：
用戶體驗不足部分：
1. 功能僅限於手機使用：誠然全球智慧型手機的覆蓋已臻完整，但仍有不少人習於在電腦進行詳細計劃，認為操作更方便，螢幕更大，鍵盤輸入更快，僅推出手機板帶來很大的限制。
2. 缺乏下達提示詞功能：一般的生成式AI會按照用戶的提示詞需求生成內容，LINE Yahoo的做法是僅提供已生成的內容，用戶只能閱讀系統已產出的內容，無法提出自己的需求，限制了工具的實用性；比如，搜尋北海道的行程，只有"適合女性旅行的北海道路線推薦"、"北海道適合與朋友一起遊覽的景點"、"適合情侶享受的北海道約會套餐"三項。
3. 無法微調已產出的內容：好的生成式AI應能根據用戶的需求進行互動和調整，不能進行細節提問和微調限制了工具的靈活性。
用戶體驗優越部分：
1. 能查詢旅遊目的地的文化展覽/活動慶典：一般OTA的當地玩樂產品以(須付費的)景點門票/活動體驗/租車包車/目的地參團為主，鮮少涵蓋在目的地有限定時間的文化展覽/活動慶典，特別是免費的部分。LINE Yahoo很好的整合了這層次的用戶需求，原本用戶可能要通過網站工具(比如：Eventbrite、Meetup)，並在社交屬性的平台(比如：Facebook、X)查詢資料，LINE Yahoo整合了相關資源，雖然可能還不夠多，因為北海道應該不會只有216個可選項，但是不錯的開始。
不過，所有基於生成式AI的行程規劃，也在不停的進展，近期我們測試ChatGPT 4o，它已經可以做到：
1. 將行程對應特定的活動賽事慶典：比如要求ChatGPT在7月16日到7月23日間，安排觀賞一場日本東京有巨人隊參加的職棒球賽，它能比對出7月17日巨人隊對阪神虎隊在晚上6點的比賽，並加入行程。
2. 能將行程更精細化：雖然不是每個人都有"特種兵"的需求，但在部分行程的時間段必須切分到小時或是分鐘，而這已經是Gemini Advanced版本的功能了。
3. 能提供對應景點的地圖：ChatGPT已能以OpenStreetMap作為地圖數據來源，使用Leaflet.js庫顯示地圖，並將景點以Folium工具製作出地圖，使用上更為直觀。
海擇資本認為，像ChatGPT這類大語言模型，在旅遊規劃功能上已漸趨成熟，剩下的是怎麼快速甚至提前理解用戶，最有效率的完成甚至"挖掘"不同用戶的定制需求；這個領域對新創公司任務艱鉅，OTA的聊天機器人未來似乎只能繼續好好賣貨，靜觀ChatGPT、Gemini、Claude等大語言模型一較長短。
-------
Recently, Japan's LINE Yahoo announced the availability of the "AI Travel Route Model" feature. When users search for "place name + tourism" on the Yahoo! search engine, it triggers a generative AI travel model. Specifically, this AI, integrated with Yahoo!, provides five functions: summaries, recommended attractions, sample routes, activity lists, and thematic maps. When activated, the AI delivers content based on these features. There are many products on the market for travel planning using generative AI, generally categorized into large language models, OTA chatbots, and tools from startups; however, the LINE Yahoo tool is distinct. Haize Capital has conducted a preliminary evaluation to understand its functionality and estimate its maturity. Here is a brief on the user experience evaluation:
Shortcomings in the user experience:
1. Limited to mobile use: Despite the widespread coverage of smartphones globally, many people prefer planning on computers due to the convenience of larger screens and faster keyboard input. Offering only a mobile version significantly limits accessibility.
2. Lack of prompt-driven functionality: Unlike typical generative AIs that produce content based on user prompts, LINE Yahoo's approach only provides pre-generated content. Users can only read what the system has produced, without the ability to submit their specific requests, reducing the tool's practicality; for example, when searching for Hokkaido itineraries, only limited options like "Hokkaido routes suitable for female travelers," "attractions in Hokkaido for friends," and "Hokkaido date packages for couples" are available.
3. Inability to adjust generated content: Effective generative AIs should allow for interaction and modification based on user needs. The inability to ask detailed questions or make fine adjustments limits the tool's flexibility.
Advantages in the user experience:
Ability to search for cultural exhibitions and festival events at travel destinations: Typical OTA local activity products mainly focus on paid attractions, activity experiences, car rentals, and group tours at destinations, rarely covering time-limited cultural exhibitions or festival events, especially free ones. LINE Yahoo effectively integrates this level of user demand. Previously, users might have needed to use website tools like Eventbrite or Meetup and search on social platforms like Facebook or X. LINE Yahoo consolidates these resources, and although the options may still be limited—since it's unlikely that Hokkaido has only 216 options—it's a commendable start.
However, all AI-generated travel planning is continuously evolving. Recently, we tested ChatGPT 4o, which can now:
1. Match itineraries to specific events: For instance, if requested to arrange a viewing of a baseball game in Tokyo involving the Yomiuri Giants between July 16 and July 23, it can match a game on July 17 where the Giants play against the Hanshin Tigers at 6 PM, and add it to the itinerary.
2. Refine itineraries in more detail: While not everyone needs "commando" precision, some parts of the itinerary may need to be broken down to hours or minutes, a feature already available in the Gemini Advanced version.
3. Provide maps for specific attractions: ChatGPT can now use OpenStreetMap as a data source, display maps using the Leaflet.js library, and create maps of attractions with the Folium tool, making the interface more intuitive.
Haize Capital believes that large language models like ChatGPT are becoming increasingly mature in travel planning capabilities. The challenge now is to quickly and even proactively understand users to efficiently meet or "uncover" their tailored needs. This area presents a formidable task for startups, while OTA chatbots may have to stick to their core function of facilitating transactions. It remains to be seen how models like ChatGPT, Gemini, and Claude will compete in this space.
標籤 Label： AI Line Travel Itinerary ChatGPT Gemini