AI Stops Flaunting Skills: Taobao Uses Technology to Solve Every Specific User Problem

In 2003, Clayton M. Christensen, a professor at Harvard Business School, proposed in The Innovator’s Solution that disruptive innovation often makes complex products or services simpler, cheaper, and more accessible, enabling more people to participate. Over the next two decades, Internet pioneers embraced this view, repeatedly validating it in technological changes. Entering the new AI wave, technological development accelerated—from architectures to parameters, from single-language to multimodal large models, from foundational to industry-specific large models. Behind the “fierce competition” is a steep rise in the technology evolution curve and a flood of products and services, yet this view still holds.

For “more people,” technological evolution is like the “invisible” part under the iceberg; they care more about the “visible” part above: what specific problems AI solves.

This is the starting point for innovation among major Internet companies and new model players. Taobao is a typical example, aiming to integrate AI functions into every corner of its application scenarios to solve every specific user problem. To this end, Taobao’s AI technology team continues to advance.

In March this year, Taobao fully upgraded its AIGX technology system, which covers all scenarios for e-commerce operations—AIGI (Index), AIGR (Recommendation), AIGB (Bidding), AIGA (Auction), AIGC (Creative), and AIGD (Data)—with a complete chain of basic research, application development, and industrialization. As Zheng Bo, Chief Scientist and Technical President of Alibaba’s China E-commerce Group, put it: “AI has become the gene of Taobao’s algorithms.”

AI Era: Generative AI Drives Intergenerational Productivity Leaps

Looking back, we’ve been shocked by AI’s breakthrough speed—from ChatGPT two years ago to Deepseek, Claude Code, and Sora 2 recently. The PC Internet era focused on “digital infrastructure” (portals, search engines), while the mobile era emphasized “time-space folding” (smartphones, 4G). Now, the AI era is defined by intelligent augmentation, where humans and technology collaborate symbiotically, driven by data, models, and computing power.

Unlike previous revolutions (which solved connection efficiency), the AI era’s biggest breakthrough is generative AI’s intergenerational productivity boost. Multimodal intelligence—AI’s ability to understand/generate text, images, voice, and video—enables natural human-AI collaboration and solves real-world problems.

Since 2023, Taobao has built self-developed large models for multimodal, search-promotion, and video generation, with capabilities in China’s first echelon. Products like AI Universal Search, AI Fitting, and Wanxiang Creation emerge from its AIGX system, forming a “technology creates value—value feeds back technology” cycle.

Multimodal Intelligence: Taobao’s Core AI Domain, Deeply Integrated into AIGX

In July, Taobao launched RecGPT, a 10-billion-parameter generative recommendation model, integrating it into the “Guess You Like” feed. Unlike traditional recommendation systems (which rely on historical data and risk information cocoons), RecGPT uses multimodal LLMs and world knowledge to generate personalized, discovery-driven recommendations.

For example: If RecGPT detects a user purchasing baby cribs and formula, it infers a new parent and recommends walkers or early education toys as the baby ages. During Double 11, it combines brand preferences to suggest winter clothing discounts. Pilot data shows RecGPT increased clicks by 16% and add-to-carts by 5%.

Taobao also advanced multimodal generation with its “Taobao Star・Video Generation Model 3.0,” which creates product videos from flat images—reducing merchant content costs. Its TStars-Omni full-modal model supports text/image/video/audio input/output, enabling tasks like checking if a refrigerator fits a kitchen layout.

Additionally, Taobao’s AI Agent iFlow CLI offers a flexible, open platform for developers (terminal/IDE plugin/SDK integration) with free domestic models. It excels in programming, advertising creativity, and travel planning.

Open Technology and Ecological Resonance

Taobao open-sourced two key frameworks: ROLL (reinforcement learning for large models) in June and RecIS (generative estimation for recommendations) in October. These tools, validated in Taobao’s internal scenarios, are now available to the industry to accelerate progress toward Artificial Super Intelligence (ASI).

With AI’s problem-solving complexity growing 5-10x yearly and error rates dropping 50% annually, Zheng Bo predicts narrow AGI (outperforming 95% of humans in most tasks) could arrive in 5-10 years.