An Alibaba Group signal is seen on the World Synthetic Intelligence Convention in Shanghai, July 6, 2023.
Aly Tune | Reuters
Alibaba on Friday launched a brand new synthetic intelligence mannequin that the corporate says can perceive photos and perform extra advanced conversations than the corporate’s earlier merchandise, as the worldwide race for management within the know-how heats up.
The Chinese language know-how big stated that its two new fashions, Qwen-VL and Qwen-VL-Chat, will likely be open supply — that means that researchers, lecturers and corporations worldwide can use them to create their very own AI apps while not having to coach their very own methods, subsequently saving time and expense.
Alibaba stated that Qwen-VL can reply to open-ended queries associated to completely different photos and generate image captions.
Qwen-VL-Chat in the meantime caters to extra “advanced interplay,” in accordance with Alibaba, reminiscent of evaluating a number of picture inputs and answering a number of rounds of questions. Some duties that Alibaba says Qwen-VL-Chat can carry out embrace writing tales and creating photos primarily based on pictures {that a} consumer inputs, in addition to fixing mathematical equations proven in an image.
One instance Alibaba gave is of an enter that includes a hospital signal within the Chinese language language. The AI can reply questions concerning the places of sure hospital departments by decoding the picture of the signal.
To date, a lot of generative AI — the place the know-how generates responses primarily based on human inputs — has targeted on responding to textual content. The most recent model of OpenAI’s ChatGPT additionally has the power to know photos and reply in textual content, very similar to Qwen-VL-Chat.
Alibaba’s two newest fashions are constructed upon the corporate’s massive language mannequin referred to as Tongyi Qianwen, launched earlier this 12 months. An LLM is an AI mannequin educated on enormous quantities of information and underpins chatbot functions.
The Hangzhou-headquartered firm this month open sourced two different AI fashions. Whereas not incomes Alibaba any licensing charges, the open-source distribution will assist the corporate get extra customers for its AI mannequin — at a time when the agency’s cloud division is seeking to reignite progress, because it prepares to go public.
