Let us know about free updates
The founder of Baidu says that demand for the type of text-based model developed by generative AI Sensation Deepseek is “reducing” as his search group is about to reestablish his position as a Chinese artificial intelligence leader.
In a prominent criticism of the limitations of China’s AI darling, Robin Lee told Baidu’s developer meeting on Friday that models like Deepseek, which generate only text-based content, are constrained.
“The market for text models is shrinking,” Li said as he released two new multimodal models (Ernie 4.5 Turbo and X1 Turbo) that feature not only text but also audio, image and video capabilities. He added that Deepseek’s model was more likely to be misleading and “hastised” and was slower and more expensive than other domestic offerings.
Deepseek did not respond immediately to requests for comment.
Li said the competitive landscape of new models is constantly changing, with a stream of “strong new models offering more options.”
His comment involves Baidu trying to reposition himself as an AI leader after being forced to pivot by dropping the subscription service into a chatbot and making the model freely available as “open source”. Baidu faces tough domestic competition from Pia Alibaba, which released a competitive open source multimodal model.
Baidu has presented several use cases for the multimodal model. This includes an update to the AI avatar platform that allows merchants to create human-like figures to host live streams and promote their products.
Li pointed out the restrictions on Deepseek, but nonetheless, the internet company has adopted rival models since the startup was catapulted into pole position as China’s leading major language model player following the release of the R1 Reasoning model in January.
Over the past few months, Baidu has added DeepSeek to the Qianfan Enterprise platform and integrated it into its maps and search applications.
Charlie Dai, Vice Chairman of Forrester Research, said Baidu’s announcement on Friday would “accelerate the adoption of AI in Chinese industries, reduce developer barriers, and continue to strengthen competition with other major vendors such as Alibaba Cloud, Huawei Cloud and Tencent Cloud.
Hong Kong’s Baidu shares rose more than 4% on the news.
DeepSeek is focusing on developing more developed models as engineers work full steam with the release of the R2 and V4 models last month.
After ChatGpt was released in November 2022, Baidu was the first Chinese company to support Openai’s popular chatbots. In March 2023, erniebot was announced, and the mobile version was later rebranded to Wenxinyan.
Recommended
Baidu’s chatbots had early success in China, but Bytedance’s Doubao and Deepseek’s chatbots became much more popular later. This year, Baidu has removed its subscription service after tumbling intake due to the abundant free products from rivals.
After initially fiercely defending the closed-source model approach, Baidu began open sourceing the model, allowing developers to have more flexibility in creating applications.
On Friday, Baidu announced the release of a new AI agent application called Xinxiang, entering an increasingly busy market that includes products from start-ups such as Alibaba Quark app and Manus AI.
Baidu has also announced that it has built a computing cluster consisting of 30,000 Kunlun P800 AI chips from its semiconductor design subsidiary. Li added that developers don’t need to worry about a lack of computing power.
Last month, FT reported that Samsung had sold the supply of Kunlun’s logic chips for three years, a key component in the manufacture of AI products. Samsung’s ability to continue working with Kunlun products may be undermined by new US export controls.
