Saturday, June 15, 2024

Baidu releases ERNIE Bot today (with invitation link)

Baidu held a press conference at its Beijing headquarters, focusing on its new generation large language model and generative AI product, ERNIE Bot (文心一言). Baidu’s founder, chairman, and CEO, Robin Li, demonstrated the performance of ERNIE Bot in five use cases: literary creation, commercial copywriting, mathematical calculations, Chinese understanding, and multimodal generation.

In literary creation, ERNIE Bot summarized the core content of the famous sci-fi novel “The Three-Body Problem” based on dialogue questions and proposed five suggestions for continuing the story, showcasing its capabilities in dialogue, summary analysis, and content generation.

ERNIE Bot also accurately answered factual questions about the author of “The Three-Body Problem” and TV series actors. Unlike other generative AI, which often fabricates answers, ERNIE Bot improved the accuracy of factual questions significantly, following Baidu’s knowledge-enhanced large model concept.

In commercial copywriting, ERNIE Bot successfully completed tasks such as company naming, slogan writing, and press release creation. Thanks to its vast training data, including trillions of web pages, billions of search and image data, billions of daily voice calls, and a 550-billion fact knowledge graph, Baidu’s ERNIE Bot has a unique position in Chinese language processing.

ERNIE Bot also possesses cognitive abilities, learning to perform relatively complex tasks such as mathematical deductions and logical reasoning. Faced with classic logic problems like “chickens and rabbits in the same cage,” ERNIE Bot understood the problem, followed the correct steps, and calculated the correct answer.

As a large language model rooted in the Chinese market, ERNIE Bot has advanced natural language processing capabilities in Chinese and performs well in Chinese culture. It demonstrated an understanding of the idiom “Luoyang paper is expensive,” its corresponding economic theory, and even created an acrostic poem using the four characters.

In multimodal generation, Robin Li showcased ERNIE Bot’s ability to generate text, images, audio, and video. Interestingly, it can even generate regional dialects like Sichuanese. Video generation capabilities are not yet available to all users due to high costs but will be gradually integrated in the future.

Robin Li stated that multimodal generation is a clear development trend for generative AI, and as Baidu’s multimodal unified large model capabilities improve, ERNIE Bot’s multimodal generation capabilities will also continue to advance.

Baidu also announced an invitation testing plan for ERNIE Bot, with the first batch of users able to access the platform via an invitation code. The API will soon be available for enterprise customers through Baidu’s Intelligent Cloud.

Register here (mobile only)

Leave a Reply

Your email address will not be published. Required fields are marked *