Alibaba releases two new models in the A.I. arms race

Alibaba, one of China’s largest technology companies, announced on Friday the release of two new artificial intelligence models, dramatically boosting the possibilities of artificial intelligence.

The open-source models, called Qwen-VL and Qwen-VL-Chat, are visual language models, meaning they “read” images rather than text, unlike rivals Chat-GPT and Google Bard. Qwen-VL-Chat promises complex features like scanning street signs to provide directions, solving math equations from photos, and weaving stories from multiple images. For example, it can scan images of signs written in Mandarin in hospitals and translate them into English, or help news organizations write captions for photos, the company said.

Another version released Friday, Qwen-VL, is an updated version of its existing image-reading chatbot that can now read higher-resolution images.

Alibaba declined to comment wealth beyond its public sphere announcement.

These new iterations of artificial intelligence are the latest shot fired in an arms race among developers to create increasingly sophisticated tools as the technology goes from gimmick to real game-changer. Alibaba, for example, says its new image-scanning technology has a good chance of helping the visually impaired shop, for example, by allowing them to scan items and have a chatbot recite labels to them.

Both models will be available on Modelscope, Alibaba Cloud’s proprietary model-as-a-service platform, and Hugging Face, a hot startup with a library of AI models.

Just a day before Alibaba announced the product, Meta unveiled an AI model fine-tuned for writing code, built on the open-source Llama 2 model released in July. Over the past few months, Alibaba has struggled to keep pace with Meta’s AI deployments. Earlier this month, Alibaba unveiled its first two open-source large-scale language models: Qwen-7B and Qwen-7B-Chat, which also serve as the basis for Friday’s release. In July, the two companies reached an agreement Bringing Meta’s Llama 2 model to the Chinese market through Alibaba’s cloud arm.

By open sourcing these new models, Alibaba allows users to adapt the tools to develop their own applications or conduct research. Most AI companies want users to adapt open-source models into tools for highly specific use cases, without taking on the onerous task of building large language models from scratch. In addition to open-source products, these companies are also offering their proprietary models as a service, hoping to capture market share in this nascent industry.

AI development is a priority for the Chinese government

Just last month, the Chinese government became one of the first countries to issue comprehensive regulations on artificial intelligence, a development experts say gives Alibaba and other Chinese tech companies the green light to make their products public.

Alibaba is also preparing for a sweeping restructuring that will spin off Alibaba Cloud, the cloud-computing arm of artificial intelligence research, into a separate unit, a move investors see as welcome.Putting both in the same sector will boost the development of AI as AI technology requires massive computing power which can only be properly served through cloud networks efficiency. Zhang Yong, Alibaba Cloud’s current CEO and chairman, will step down in September and be replaced by Alibaba’s two co-founders: John Woo as CEO and Joe Tsai as chairman.

The Chinese government has said more than once that it believes artificial intelligence is critical to its technological future and has launched an arms race with the United States, where even seemingly innocuous tools like those released by Alibaba on Friday could be censored. implicated Because of their underlying technology and how other developers might use them. Kerry Brown, director of the Lau China Institute at King’s College London, said AI “has become a proxy in the battle for dominance between the US and China”. wealth early this month.

So far, Chinese tech companies appear to be lagging slightly behind their American counterparts. The open-source version of Meta’s Llama 2 model is based on about 70 billion variables (called parameters in AI lingo), roughly 10 times as many bigger than Alibaba’s new version (Alibaba does say it has a larger model, but it’s not open source.) Despite the U.S. advantage, government officials worry that the Chinese government could end up adopting AI technology developed by some private companies for military or surveillance purposes, according to Axios.

Svlook

Leave a Reply

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