Baidu's Long Text Understanding Model "Orange Text" Launched: Another Addition to the Large Model Matrix
In Baidu's Q4 2023 earnings call, Baidu's founder, chairman, and CEO Robin Li revealed that Baidu Intelligent Cloud's total revenue for the fourth quarter was 8.4 billion yuan, with large models contributing approximately 660 million yuan in incremental revenue to the cloud business.
On May 30, at the 2024 Baidu Mobile Ecosystem Conference, Baidu unveiled the comprehensive AI native application "Orange Text." This application is positioned as a long text AI understanding model, enabling users to understand, summarize, and query documents with "extremely large volumes, multiple formats, and very long content." It also supports the generation of "extremely long" texts, deep editing, and multimodal free creation.
According to the introduction, "Orange Text" is based on the accumulation of 1.2 billion contents from Baidu Wenku, 200,000 finely tuned data points, behavior data feedback from 140 million users, and the development of hundreds of AI capabilities. On Orange Text, users can perform AI-powered intelligent searches across the web and academic searches, leveraging Baidu Wenku, Baidu Scholar, and tens of billions of professional information and resources from across the web.
Since 2022, Baidu has successively released several important large language models, including the well-known Wenxin Yiyan, ERNIE 3.5, and the Wenxin Large Model. Among them, Wenxin Yiyan's training data includes trillions of web data, billions of search data, and images, making it Baidu's flagship product in large models; ERNIE 3.5 is a large language model built on the foundation of ERNIE 3.0, released in June 2023, positioned as a "flagship large-scale language model," with leading advantages in processing text information; the Wenxin Large Model is an industry-level knowledge-enhanced large model released by Baidu, rated as the top in comprehensive evaluations.
In October 2023, Baidu released the Wenxin Large Model 4.0, claiming it is "on par with GPT-4 in comprehensive capabilities," marking a significant improvement in the company's large language model capabilities. Baidu stated that the Wenxin Large Model 4.0 has further optimized knowledge graphs and knowledge embeddings, enabling more accurate understanding and generation of information containing complex knowledge. In the field of multimodal fusion, the Wenxin Large Model 4.0 has further strengthened its ability to process multimodal data such as text, images, and videos, making it perform better in cross-modal tasks.
The newly released "Orange Text" application is another pioneering achievement by Baidu in the field of large models. Wang Ying, Vice President of Baidu and Head of the Wenku Business Unit, said today: "As an AI practitioner and witness, it is a great honor to deeply participate in such a technological revolution."
At the conference, Baidu also announced the latest progress of Baidu Wenku after its reconstruction by the large model. Currently, the platform has gathered 1.2 billion high-quality document resources and released hundreds of multimodal AI functions, attracting over 140 million AI users, with the usage of AI functions exceeding 1.5 billion times.
In Baidu's Q4 2023 earnings call, Baidu's founder, chairman, and CEO Robin Li revealed that Baidu Intelligent Cloud's total revenue for the fourth quarter was 8.4 billion yuan, with the large model bringing about 660 million yuan in incremental revenue to the cloud business. The daily call volume of the Wenxin Large Model has exceeded 50 million times, with a quarter-on-quarter growth of 190%; in December, about 26,000 enterprises called the Wenxin Large Model, with a quarter-on-quarter growth of 150%. Currently, well-known enterprises such as Samsung, Honor, and Autohome have all reached cooperation with Baidu.
·Original
Disclaimer: The views in this article are from the original author and do not represent the views or position of Hawk Insight. The content of the article is for reference, communication and learning only, and does not constitute investment advice. If it involves copyright issues, please contact us for deletion.