Photo credit: Shutterstock
Alibaba Group said it has integrated its large language model Tongyi Qianwen into its e-commerce platform Taobao, part of a broader rollout of the artificial intelligence-powered tool across its ecosystem.
The Hangzhou-based holding company integrated the generative-AI-powered personal assistant across the consumer-to-consumer marketplace Taobao to provide tailored recommendations and support to consumers.
The launch illustrates Alibaba’s focus on artificial intelligence as a tool to improve consumers’ experience. In his first memo as Alibaba Group’s CEO to staff this week, Eddie Wu wrote that the two core strategies for Alibaba going forward will be “User First” and “AI-driven”.
Alibaba Cloud launched the Tonyi LLM in April and plans to have all applications and products across its ecosystem run on the large language model.
Personal Assistant
Taobao users can search Taobao Wenwen on the Taobao app and then click on the chatbot for help with searches.
Based on the large language model’s understanding of text and its generative capabilities, it can help consumers narrow their searches by suggesting more relevant questions and offering detailed product recommendations in various formats, including text and short videos.
The AI-powered tool on Taobao is undergoing tests by Alibaba staff ahead of broader release. The Taobao marketplace sits within the Taobao and Tmall Group following Alibaba’s reorganization into a holding company structure.
Other than e-commerce searches, the AI assistant can also perform a wide range of tasks, from creating detailed travel itineraries to drafting product promotion posts.
Taobao and Tmall Group, in collaboration with another Alibaba unit Aicheng Technology, also released an open-source LLM training framework called Megatron-LLaMA to the mass market. The training framework simplifies training a LLaMA model in distributed training clusters.
It is often expensive for developers to try out their own designs, as training or fine-tuning an LLM requires powerful computational resources. Typically, GPUs with large memory or distributed clusters composed of multi-GPU devices are essential for training LLMs. Megatron-LM is a distributed training solution.
In a separate move, Alibaba’s cloud computing unit, Alibaba Cloud opened Tongyi Qianwen to the public via its official website and for enterprises via the Tongyi API. Previously, the proprietary model was only available to a limited cohort of users during beta testing.
Tongyi Qianwen received over 200,000 beta testing requests in the weeks after its launch in April.
