Pandaily reported that Jiang Guanjun, head of AI algorithms at Alibaba’s Qwen business unit, presented a framework for personal agents at the China National Computer Congress in Chengdu on 22 October. Jiang described chat, agents that carry out assigned tasks, and assistants that use personal context as successive stages of AI applications. Qwen says it has mapped more than 8,000 individual model capability requirements across more than 30 industries, with industry experts evaluating and tuning the model against them.
Jiang’s framework joins the model to memory and outside services. Rather than retain every conversation as a transcript, Qwen organises memories by how closely events relate, storing and retrieving them in tiers. Jiang said accumulating chat history adds noise and inference cost.
Looking for advice about a job search could draw on an earlier conversation about a pension if the memory system connected the two, as in Jiang’s example of a family seeking help.
Qwen is building an open service system connecting professional knowledge and outside agents to areas including e-commerce, healthcare and education. Alibaba Cloud’s earlier Qwen Book unveiling concerned a desktop system designed to give an agent access to applications and controls. Alibaba opened sign-ups for its personal agent Qwen Go! in October, Pandaily reported.
