Keep knowledgeable with free updates
Merely signal as much as the Synthetic intelligence myFT Digest — delivered on to your inbox.
Baidu’s founder has mentioned demand for the kind of text-based fashions developed by generative AI sensation DeepSeek is “shrinking”, as his search group seeks to reestablish itself as a synthetic intelligence chief in China.
In a hanging criticism of the constraints of China’s AI darling, Robin Li advised Baidu’s developer convention on Friday that there have been constraints to DeepSeek’s main mannequin. Its well-liked R1, broadly praised by the worldwide developer group, is geared in direction of text-based duties.
“The marketplace for textual content fashions is shrinking,” Li mentioned, as he launched two new multimodal fashions — Ernie 4.5 Turbo and X1 Turbo — with not simply textual content but additionally audio, picture and video capabilities. He added that DeepSeek’s mannequin had a better propensity for deceptive “hallucinations” and was slower and dearer than different home choices.
DeepSeek didn’t instantly reply to a request for remark.
Li mentioned the aggressive panorama for brand new fashions was continually altering, with a stream of “highly effective new fashions that present extra alternative”.
His feedback come as Baidu tries to reposition itself as an AI chief after being compelled to pivot by dropping its subscription service to its chatbot and making its fashions freely obtainable as “open supply”. Baidu faces stiff home competitors from its peer Alibaba, which has launched aggressive open-source multimodal fashions.
Baidu introduced a number of use circumstances for its multimodal fashions, together with an replace to its AI avatar platform that permits retailers to create humanlike figures to host livestreams and promote their merchandise.
Whereas Li pointed to DeepSeek’s constraints, the web firm has however embraced its rival’s fashions for the reason that start-up was catapulted into pole position as China’s main massive language mannequin participant following the discharge of its R1 reasoning mannequin in January.
In latest months, Baidu added DeepSeek to its Qianfan enterprise platform, in addition to integrating it into its map and search functions.
Charlie Dai, a vice-president at Forrester Analysis, mentioned Baidu’s Friday bulletins would “speed up AI adoption in industries in China, decrease boundaries for builders and proceed to accentuate the competitors with different main distributors” equivalent to Alibaba Cloud, Huawei Cloud and Tencent Cloud.
Baidu shares in Hong Kong rose greater than 4 per cent on the information.
DeepSeek remains to be centered on additional growing fashions, with its engineers working at full steam on the discharge of subsequent variations — the R2 and V4 fashions, the Financial Times reported final month.
After the discharge of ChatGPT in November 2022, Baidu was the primary Chinese language firm to answer OpenAI’s well-liked chatbot. In March 2023, it introduced Erniebot, with the cell model later rebranded to Wenxinyan.
Baidu’s chatbot achieved early success in China, however ByteDance’s Doubao and DeepSeek’s chatbot later overtook it in reputation. This yr, Baidu dropped its subscription service after a tepid uptake because of the abundance of free merchandise from rivals.
After initially fiercely defending the closed-source mannequin strategy, Baidu additionally began open-sourcing its fashions, permitting builders larger flexibility to create functions.
On Friday, Baidu introduced the discharge of a brand new AI agent software known as Xinxiang, coming into an more and more crowded market that features Alibaba’s Quark app and choices from start-ups equivalent to Manus AI.
Baidu additionally introduced it had constructed a computing cluster composed of 30,000 Kunlun P800 AI chips from its semiconductor design subsidiary, which it mentioned might assist coaching of a number of DeepSeek-like fashions. Li added that builders didn’t have to fret a couple of scarcity of computing energy.
The FT final month reported that Samsung had offered Kunlun three years’ provide of logic chips, a vital element in manufacturing AI merchandise. Samsung’s capacity to proceed engaged on Kunlun merchandise has probably been curbed by new US export controls.
