Alibaba has become the latest Chinese tech giant to unveil generative AI tools across its product range. This includes a new chatbot called Tongyi Qianwen that will be integrated into its apps “in the near future”. The large language model was shown to draft invitation letters, plan trip itineraries and advise shoppers on the types of make-up they should purchase as part of a controlled, pre-recorded demonstration. It comes as the Chinese cyber regulator takes steps to control the use of general-purpose AI tools like chatbots and generative AI.
Chinese companies beyond Alibaba are rushing to build generative AI and chat-based interface tools for large language models off the back of the unexpected success of OpenAI’s ChatGPT, which became one of the fastest-growing consumer products after its launch in November last year. Since then, the majority of large Western technology companies have also adopted some form of foundation model AI in workflows including Google, Microsoft and Salesforce.
Tongyi Qianwen will start out life inside DingTalk, which is a workplace messaging app from Alibaba similar to Microsoft Teams. The AI tool will be used to summarise meeting notes, write emails and draft business proposals similar to the approach Microsoft has taken with Copilot in its 365 suite.
Alibaba AI will include chat and voice integration
Unlike Microsoft’s take, Tongyi Qianwen will also be added to the Tmall Genie voice assistant, becoming the brain of the smart speech AI. This, says Daniel Zhang, CEO of Alibaba, will “bring about big changes to the way we produce, the way we work, and the way we live our lives”. He described this approach to technology and AI models as “the big picture for making AI more popular in the future”.
As well as integrating the large language model into its own apps and products, Alibaba plans to make Tongyi Qianwen available as an API to its cloud clients. This is something Google is doing with its own PaLM large language model and Microsoft is doing with ChatGPT products over Azure.
Chinese companies are going “all in” on AI after government officials said it wanted to make the technology a core component of the economy. Zhang told reporters at the launch of the new model that this was “the starting line” for AI development and “seizing this opportunity is a common wish for all of us.”
Almost all big Chinese tech companies have announced plans to release either a generative AI or chat-based AI product including Baidu, Sensetime, Tencent and China Telecom. Most recently Sensetime has unveiled a number of new AI-powered products based off its SenseNova AI model including image generators and a chatbot, although, like most of the big Chinese announcements, no details have been revealed about rollout dates.
Baidu has shown off and had a number of users test its Ernie Bot, based on its massive ERNIE large language model. It is multi-modal on both the input and output, meaning it can summarise as well as produce text, images and video. Reuters’ tests of the tool found that while it had a good command of the Chinese language, it generally avoided political questions.
Chinese general AI regulation
While Chinese companies race ahead with new generative AI tools, regulators in the country are working on draft measures to manage the technology. This includes requiring companies to submit security assessments to authorities before launching to the public.
The Cyberspace Administration of China (CAC) regulator is the latest global watchdog to investigate ways to manage the impact of such a technology. The CAC said that the aim is to support AI innovation and encourage its safe and reliable use while ensuring the content generated by generative AI aligns with Chinese values.
This will include placing requirements on companies to ensure the legitimacy of data used in training the products and taking measures to prevent discrimination from the algorithms to the training data. Users will also be required to submit real identities when using these general purpose AI tools with providers fined or facing criminal investigations for failing to comply.
If any content deemed inappropriate is generated by one of the platforms launched publicly then that company will be given three months to update the model and underlying technology to prevent it happening again.
The UK is taking a “hands off” approach in its new AI framework that includes putting the final say in the hands of individual sector regulators. The EU is also looking at how to regulate general purpose AI in its EU AI Act.