- The new iRAG technology
mitigates hallucinations in image generation, boosting the
applicability of AI-generated images.
- Miaoda is a no-code tool that
enables users to build applications using simple natural
language.
- ERNIE now handles 1.5 billion
daily API calls, marking an exceptional 30-fold growth over the
previous year.
SHANGHAI, Nov. 12,
2024 /PRNewswire/ -- Baidu, Inc.
(NASDAQ: BIDU and HKEX: 9888), a leading AI
company with strong Internet foundation, today hosted its annual
flagship technology conference Baidu World 2024 in Shanghai. At the event, the company announced
a range of new AI technologies and solutions to accelerate the boom of AI
applications, featuring iRAG (Image-Based
Retrieval-Augmented Generation), a new
technology designed to tackle hallucinations in image
generation, and Miaoda, a no-code tool that empowers
businesses and individuals to create applications.
The launch comes amid surging demand for Baidu's AI offerings,
as evidenced by the ERNIE
foundation model's daily API calls reaching 1.5 billion
by early November. This represents a substantial 30-fold increase
from the 50 million announced a year ago.
"The growth rate exceeded my expectations," said Robin Li, Co-founder, Chairman, and CEO of
Baidu, who described the steep increase as a reflection of the
explosive growth in generative AI applications in China over the past two years.
Looking ahead, Li emphasized that agents will serve as the
predominant form of AI applications and are approaching a tipping
point of explosive growth. Underscoring this point, Li introduced
the Top 100 Agents and Top 100 Industry Applications on the
ERNIE AgentBuilder platform.
Baidu World 2024 also highlighted the latest user growth for
ERNIE Bot, ERNIE's expanding role in enterprise applications, and
featured the debut of Xiaodu AI Glasses by Xiaodu Technology.
New iRAG
Technology to
Mitigate Hallucinations in Image Generation
Hallucinations, a phenomenon in which AI generates false or
misleading information, have remained one of the most intractable
barriers to the widespread adoption of generative AI. In text
generation, RAG technology has largely resolved the problem of
hallucinations, greatly enhancing the accuracy of generated
answers. However, in the field of multimodality, hallucinations
remain a key obstacle, often manifesting as inaccurate depictions
of people or landmarks.
Baidu's newly launched
iRAG technology can mitigate hallucinations in text-to-image
generation. Leveraging Baidu Search's vast collection of hundreds
of millions of images and the company's strong foundation model
capabilities, the new technology enables text-to-image models to
deliver hyper-realistic visuals while also significantly reducing
the cost of image production. The ability of the iRAG to reduce
hallucinations boosts the applicability of images generated by
text-to-image models across visual mediums, including comics,
storyboards, posters, among others. Li described the reduction of
hallucinations as laying the groundwork for the coming boom in AI
applications.
Miaoda: Building Applications with Natural Language
Baidu also unveiled Miaoda, a no-code tool that makes it
possible to build entire applications simply by describing them in
natural language. Miaoda provides no-code programming, multi-agent
collaboration, and multi-tool invocation. No-code programming
allows anyone to generate code without writing a single line,
lowering barriers to AI development and making it accessible to
all. Its multi-agent collaboration leverages ERNIE's thinking and
planning capabilities to coordinate and manage different agents
effectively, while its multi-tool invocation taps into ERNIE's tool
invocation abilities, extensively utilizing web search, iRAG, maps
API, and other tools for a seamless workflow.
"Baidu isn't aiming to launch a 'super app'; instead, we aim to
help more people and businesses create millions of 'super useful'
applications," Li said.
Mirroring the real-world product development process, Miaoda
draws on the abilities of different agents across multiple domains
such as project management and planning, content editing,
programming and quality control. Miaoda can even automatically
identify bugs and use a range of tools. Li called it "the most
complex application case of multi-agent collaboration to date".
Miaoda empowers everyone with the capabilities of a programmer –
anyone who can speak, can create applications, greatly enhancing
human productivity.
Agents as the Next Frontier in AI Application
"Today, while all leading global tech firms are paying attention
to agents, few have made them as central to their strategy as Baidu
has," Li said at the event. He likened the potential of agents to
websites in the PC era and social media accounts in the mobile
age.
"Agents are more human-like, more intelligent, and act like your
sales, customer service representatives, or assistants. Agents will
become a new vehicle for content, information and services," Li
added.
Four types of agents – company, character, tool, and industry –
were highlighted at the event to demonstrate the potential of
agents in the AI era. Company agents, for example, differ from
traditional websites that only show static company and product
information, and can instead proactively recommend products based
on customer needs and respond promptly to service requests, greatly
improving the efficiency of interactive marketing. After deploying
an agent powered by ERNIE, automaker BYD saw a 119% increase in
lead conversion.
Li also offered the example of the tool agent Free Canvas,
developed in collaboration between Baidu Wenku and Baidu Drive. It
allows users to drag and drop documents, audio, video, and other
materials onto a canvas-like interface to quickly generate
multimodal content. Similarly, the industry agent Faxingbao was
specifically developed for the legal field. It has answered over
16.6 million legal questions and can also calculate compensation,
draft legal documents, and recommend suitable lawyers.
To date, Baidu's ERNIE AgentBuilder has garnered the interest of
150,000 businesses and 800,000 developers. The top
100 agents encompass character-based agents like the "Farmer
Academician", as well as agents for tools, industries, workplace,
emotions, and entertainment.
Expanding ERNIE's Reach with Growing Developer Community,
Enterprise Applications and AI Glasses
Haifeng Wang, Chief Technology
Officer of Baidu, announced that ERNIE Bot has amassed 430
million users, while the PaddlePaddle deep learning platform
and ERNIE have a total of 18.08 million developers. Wang
also mentioned that the ERNIE foundation model is still in
training, with an even more powerful version worth looking forward
to.
Dou Shen, Executive Vice
President of Baidu and President of Baidu AI Cloud Group,
highlighted that the Qianfan platform has helped customers
fine-tune 33,000 models, and develop 770,000 enterprise
applications. Shen noted that the explosion in new AI
applications has begun on the enterprise front, supported by a new
AI infrastructure comprising an enterprise-level foundation model
engineering platform and a heterogeneous computing platform, poised
to replace traditional cloud computing.
Ying Li, CEO of Xiaodu
Technology, introduced Xiaodu AI
Glasses, the first native AI glasses powered by a
Chinese language foundation model. Powered by ERNIE and equipped
with visual, audio, and location-based capabilities, these smart
glasses serve as a versatile AI assistant across everyday
scenarios. They function as a
personal tour guide, offer information via Baidu Maps and
Search, and excel in instant translation and content summarization
from photos. Ideal for both academics and casual readers, they also
assist with intelligent note-taking and can personalize music to
enhance the user's surroundings. Li revealed that Xiaodu AI Glasses
will be available for sale in the first half of 2025.
About Baidu
Founded in 2000, Baidu's mission is to make the complicated
world simpler through technology. Baidu is a leading AI company
with strong internet foundation, trading on the NASDAQ under "BIDU"
and HKEX under "9888." One Baidu ADS represents eight Class A
ordinary shares.
Media Contact
intlcomm@baidu.com
View original content to download
multimedia:https://www.prnewswire.com/news-releases/baidu-unveils-irag-and-no-code-tool-miaoda-to-drive-ai-application-boom-302303079.html
SOURCE Baidu, Inc.