Baidu Rolls Out Ernie 4.0 Generative AI Model Rivaling GPT-4
Chinese tech giant Baidu introduced an upgraded version of its generative AI foundation model called Ernie 4.0 at its annual Baidu World conference in Beijing this week. Ernie 4.0 incorporates several improvements to Ernie’s features, which the company claims rival anything that OpenAI’s GPT-4 model can do.
Ernie AI
Baidu CEO Robin Li demonstrated Ernie 4.0 through the Ernie Bot conversational assistant powered by the model as ChatGPT is powered by GPT-4. He highlighted several ways that Ernie 4.0 improves on the previous iteration, including an ability to grasp nuance and subtext in conversation, even spotting hidden messages. The AI model can also synthesize text, images, and video from an initial prompt in minutes. While on stage, Li guided Ernie through developing the outline for a novel about martial arts, solved complex math problems, and even produced a car commercial in minutes. The overall performance of Ernie 4.0 is 30% above the last iteration, Baidu claimed. Ernie 4.0 is built on Mandarin but can operate at a less advanced level in English as well.
“Ernie 4.0 has achieved a full upgrade with drastically improved performance in understanding, generation, reasoning, and memory,” Li said in his presentation. “These four core capabilities form the foundation of AI-native applications and have now unleashed unlimited opportunities for new innovations.”
Access to Ernie 4.0 is currently invite-only through Ernie Bot and Baidu’s foundation model platform. The company plans to widen availability to enterprise clients and eventually the general public. Baidu product users will start having access to Ernie 4.0 through the company’s family of products, however. The generative AI model is augmenting Baidu Search with interactive chat and deeper search results pulled from web sources, including images and “dynamic graphs.” Baidu Maps will get its own AI guide, offering proactive advice during trips. Ernie 4.0 will also power a new generative AI assistant for cloud storage service Baidu Drive named YunYiduo, enabling users to ask by voice or text command for files to open or search for data within documents and videos.
On the enterprise front, Baidu’s Infoflow workplace app will use Ernie 4.0 for scheduling events and summarizing meetings. The summarization feature is also coming to the Google Workspace-equivalent Baidu Wenku hub, where it will also produce documents and slideshows from text prompts. The model will be used to bring generative AI capabilities to the Baidu GBI analytics platform.
All of these services come following the initial Ernie Bot release, which had to demonstrate compliance with China’s new generative AI regulations. Authorities mandated training data and outputs align with government-approved values, leading to events like Apple pulling over 100 apps from its Chinese App Store for not meeting the rules. And while Ernie 4.0 may be capable of competing with GPT-4, Baidu still has local rivals to face, like Alibaba and Huawei. Still, the company has come a long way from the initial Ernie Bot demo that relied on videos and drew criticism for its caution.
“China possesses rich resources in terms of AI application scenarios, as well as a wealth of users who are passionate about embracing new technologies,” 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.”
Follow @voicebotaiFollow @erichschwartz
Baidu Releases Generative AI Chatbot Ernie Bot to General Public in China With Government Approval
Alibaba Adds Visual Understanding to Open-Source Generative AI Large Language Models