Covering Scientific & Technical AI | Saturday, December 28, 2024

Baidu Launches ERNIE 4.0 Foundation Model, Leading a New Wave of AI-Native Applications 

BEIJING, Oct. 17, 2023 -- Baidu, Inc. today hosted its annual flagship technology conference Baidu World 2023 in Beijing, marking the conference's return to an offline format after four years.

Credit: Shutterstock

With the theme "Prompt the World," this year's Baidu World conference saw Baidu launch ERNIE 4.0, Baidu's next-generation and most powerful foundation model offering drastically enhanced core AI capabilities. Baidu also showcased some of its most popular applications, solutions, and products re-built around the company's state-of-the-art generative AI.

"ERNIE 4.0 has achieved a full upgrade with drastically improved performance in understanding, generation, reasoning, and memory," Robin Li, Co-founder, Chairman and CEO of Baidu, said at the event. "These four core capabilities form the foundation of AI-native applications and have now unleashed unlimited opportunities for new innovations."

In his keynote, Li showcased the new ERNIE Bot powered by ERNIE 4.0 through extensive demonstrations of its four core AI capabilities. ERNIE Bot is able to "Understand" and provide answers to complex and even disorganized human requests, as well as interpreting hidden messages. ERNIE can also "Generate" a range of content, including text, images, and videos, in just a few minutes, based on one simple text prompt and image input. Li further demonstrated ERNIE Bot's ability to "Reason" by having it solve complex geometry problems. Lastly, ERNIE Bot's ability to "Memorize" and integrate incremental inputs was demonstrated by writing a short story and regularly adding new information as ERNIE was in process of writing the story.

According to Dr. Haifeng Wang, Chief Technology Officer of Baidu, ERNIE 4.0 started beta version testing in September based on a limited amount of user input, and has since seen its overall performance improved by almost 30%. ERNIE 4.0 is now accessible to invited users on ERNIE Bot, and the API will be available upon application to enterprise clients via Qianfan foundation model platform.

The emergent intelligence brought by foundation models and large language models (LLMs) is unlocking unprecedented avenues for innovative applications. Over the past few months, Baidu has spearheaded this effort by adopting an AI-native approach to re-build its family of apps, solutions, and products, including Baidu Search, Baidu GBI, Infoflow, Baidu Wenku, Baidu Drive and Baidu Maps. This revamp has not only significantly enhanced productivity but has also sparked creativity, leading to delightful and diverse user interactions.

Key updates on Baidu AI-native applications, solutions, and products:

  • New Baidu Search: Next-gen search experience powered by Generative AI: Generative AI and search products are by nature highly complementary. The addition of Generative AI has transformed the search experience, changing the way users engage with information and providing more in-depth information from a single search. The new Baidu Search can now aggregate and summarize information from diverse web sources and present a consolidated answer through a variety of modals, including text, image and dynamic graph, ensuring answers that are both vivid and concise. Baidu search will also suggest more content streams that are relevant to previous requests while staying highly customized based on the user's unique profile. Additionally, Baidu Search offers an interactive chat interface for more complex searches, enabling users to iteratively refine their queries until reaching the optimal answer, all in one search interface.
  • Baidu GBI: China's first generative business intelligent solution: As a business intelligence solution now powered by Generative AI abilities, Baidu GBI plays the role of an all-knowing, fast-reacting assistant in the business world. It can quickly execute data inquiries, output business analytics, and deliver results through easy-to-understand text and graph summaries. It reduces the heavy lifting of having to go through multiple spreadsheets to retrieve and analyze data, drastically accelerating the process of business analytics and decision making.
  • Infoflow: The intelligent workplace APP with super assistant: Infoflow integrates enterprise communication, collaborative work, knowledge management, and enterprise services all in one platform. Boosted with ERNIE's AI capabilities, Infoflow can execute an array of office tasks through natural language processing - from orchestrating meeting agendas and summarizing key takeaways from complicated chat history to preparing meeting briefings based on users' previous meeting documents and even securing flight bookings based on their agenda.
  • Baidu Wenku: From sharing documents to supercharging productivity: With the addition of foundation model capabilities, Baidu Wenku has evolved from a content documenting platform to a content productivity platform. It can now assist users in conducting academic research, summarizing key information, creating documents, and generating slideshow presentations, drastically reducing the time spent in content production in scenarios such as preparing a keynote speech.
  • Baidu Drive YunYiduo: World's first cloud drive intelligent assistant: With the addition of YunYiduo to Baidu Drive, users can now use text prompts or voice prompts to retrieve files, images, videos, and more formats, even for very specific requests. Powered by the multi-modal capabilities, YunYiduo can summarize key information even from different figures appearing in a video, precisely pinpoint the exact frame in the video, render transcriptions, and add subtitles to drastically enhance users' efficiency to extrapolate, summarize, and generate content across various media formats.
  • Baidu Maps: World's first AI-native map product making modern travel effortless: The new Baidu Maps integrated AI capabilities into its extensive functions, including navigation, ride-hailing, restaurant recommendations, hotel booking, and more. The newly integrated AI guide will take the initiative to suggest next step of action to users during their trip.
  • Plugin Matrix AI Platform: Offering low-threshold access and productivity tools to develop AI plugins, this platform allows both individual developers and enterprises to build AI plugins. In just one month since its launch, 27,000 developers have applied to join the Plugin Matrix AI Platform, covering a variety of sectors ranging from legal counseling and resume creation to brain mapping and spoken language practice.
  • Qianfan Foundation Model Platform: As the world's first one-stop enterprise-level foundation model platform, Qianfan consolidates a wide range of pre-installed foundation models (both from the Baidu ERNIE foundation model family and third-party foundation models), datasets, and a full set of toolchains to enable seamless development and application of foundation models. Qianfan has now become China's largest, most inclusive foundation model development platform, encompassing 42 existing foundation models, and has empowered enterprises from different sectors to develop their own foundation models and solutions covering more than 500 use scenarios.

Baidu World 2023 also demonstrated Baidu's recent major strategic progress across multiple business sectors. At the event, Dr. Dou Shen, Executive Vice President of Baidu and General Manager of AI Cloud Group, announced a full upgrade of Baidu AI Cloud's strategy, aiming to provide infrastructure such as Baidu Qianfan Foundation Model Platform to empower enterprise customers in the development low-threshold AI application and foundation models.

In his keynote speech at the event, Jackson He, Senior Corporate Vice President of Baidu and General Manager of the Mobile Ecosystem Group, outlined a future where Baidu's mobile ecosystem, linking billions of users and millions of ecosystem partners, will stand at the cutting-edge of developing and introducing AI-native applications.

Ying Li, Corporate Vice President of Baidu and CEO of Xiaodu Technology, revealed a series of smart products built on incorporating AI in intelligent devices. The new releases include Tiantian Home Robot, Xiaodu Qinghe Learning All-in-one Machine, as well as Smart Speaker Series Tiantian Casa.

Zhenyu Li, Senior Corporate Vice President of Baidu, explained how the deployment of foundation models has accelerated the scaling of fully driverless robotaxis by drastically improving operational efficiency. Today, Baidu possesses the world's largest autonomous driving ride service platform, Apollo Go, with cumulative rides provided to the public reaching more than 4 million. Apollo Go's fully driverless robotaxi fleet, operating without any driver or safety operator in the car, currently serves major cities in China, including Beijing, Wuhan, Chongqing, Shanghai, and Shenzhen. Baidu soon expects a future where lower-cost robotaxi services will be more accessible to the public, bringing the industry closer to profitability.

"China possesses rich resources in terms of AI application scenarios, as well as a wealth of users who are passionate about embracing new technologies," Robin Li said. "Now, with the emergence of advanced foundation models, I am confident that we can build a thriving AI ecosystem and create a new round of economic growth together."

Watch the entire recording on Baidu's YouTube page.


Source: Baidu

AIwire