Baidu has created the world's first AI-native new map in the era of large models

In the era of big models, what capabilities do products and applications of AI-native thinking reconstruction have? On October 17, Robin Li, founder, chairman and CEO of Baidu, released a series of AI native applications at Baidu World 2023. Among them, Baidu Maps was comprehensively reconstructed from the two dimensions of "new interaction" and "new navigation". Becoming the world's first AI native map.

Robin Li said: “In the past 18 years, Baidu Maps has accumulated a very rich set of functions. It can navigate, hail a taxi, find a driver, book hotels, buy air tickets, and even meet POIs under more complex conditions. Find needs. In the past, these operations were complicated and often required six, seven steps or more to complete; now, as long as users state their needs, the map can mobilize the API interfaces of thousands of services, and multi-level menus become one-step direct, quick response .”

Reconstruct interaction: Not only navigation, but also your travel guide

How to reconstruct AI native applications in Maps? Robin Li once mentioned: "Today people can interact with AI using natural language. As long as the ideas are active and the expressions are clear, the machine can do the work for you."

For Baidu Maps, the first step in reconstruction is to solve the problem of understanding users' natural language open input. Based on the intensive training method, the Wenxin large language model makes the semantic understanding of complex sentences a piece of cake. Even when recognition errors occur in local inputs, it can still give ideal answers and successfully achieve multiple rounds of natural language interaction, making it closer to "real people". comminicate".

Based on this, the newly launched AI native map has launched the "AI Guide" function. Users can awaken various functions and services buried in the map in one step simply through natural language conversations, further improving travel and decision-making efficiency and giving users a new interactive experience. At the same time, it also has a personalized digital human image. In addition to the default image recommended by the system, users can generate their own exclusive digital human by uploading pictures. Based on user feedback and deep learning capabilities, the AI ​​guide can learn and grow independently during continuous use, gradually deepening its understanding of users' travel preferences, and truly becoming a powerful assistant in travel.

Multi-level menu direct access in one step

Today’s Baidu Map is equivalent to an AI native map agent that can evolve itself. Just put forward your needs in a chat, such as buying snacks in Daoxiang Village on the way home, recommending a restaurant that is moderately close to three different communities and suitable for bringing children, or even calling the restaurant directly to reserve a private room, etc. Baidu Maps can listen Understand your every word and accurately "understand and generate" the answers you need. In the future, with the widespread use of AI guides, a steady stream of real feedback information will build a better question and answer system data flywheel, making this "guide" smarter and smarter as it is used, and may even exceed your imagination. Baidu Maps V19 will be updated from now on, and you can experience it early by searching for "AI Guide".

Multiple rounds of natural language interaction to meet travel needs

Reconstruct navigation: urban lane-level navigation3.0Accelerate nationwide implementation

Baidu Maps relies on its self-developed Beidou high-precision positioning technology, intelligent lane recommendation algorithm and 3D precision model-level high-definition rendering technology to newly upgrade and release urban lane-level navigation 3.0, which can provide clearer and three-dimensional global lane planning and guidance. It recommends lanes with fewer lane changes and smoother lane changes throughout the process, helping users see the road clearly, walking in the right lane, easily coping with many "lane changing anxiety" scenarios, and improving driving safety and experience.

Urban Lane Level Navigation 3.0

At the same time, Baidu Maps has also self-developed and implemented the industry's first large-scale map generation model. Through the new paradigm of end-to-end lane network, it has reduced the cost of map production by 95%, significantly leading the industry level, and greatly improving the entire map process. Production efficiency. In fact, the production of lane-level maps requires higher "freshness" of data so that it can better adapt to road changes, such as solid lines changing to dotted lines, turning left and going straight, etc. The large map generation model independently developed by Baidu can effectively solve industry problems such as high manual dependence, low data production efficiency, and poor scene generalization capabilities under the traditional model, and accelerate the large-scale mass production application of lane-level map data. It is reported that Baidu Maps lane-level navigation will cover more than 100 cities by the end of 2023 and achieve nationwide coverage in 2024.

As Robin Li said in May this year, Baidu wants to be the first company to reconstruct all its products. Baidu Maps, which has been born for 18 years now, has also handed over the world's first AI-native map answer sheet. When it comes of age, it uses a large model as a base to radiate vigorous creativity and move towards the AI-native era. Baidu Vice President Shang Guobin said: "We hope that in the future, users can continue to discover that AI native maps bring us more surprises when using Baidu Maps."

The above is the detailed content of Baidu has created the world's first AI-native new map in the era of large models. For more information, please follow other related articles on the PHP Chinese website!