How Alibaba’s Taobao AI Models Revolutionize E‑Commerce Recommendations and Bidding

Overview of AIGX Technology Suite

On March 26, Alibaba’s Taobao Group VP of Technology and CTO of Alibaba Mom, Zheng Bo, unveiled the Taobao Group’s large‑model family and the AIGX technology system that links foundational research, application development, and industrialization across e‑commerce scenarios such as indexing (AIGI), recommendation (AIGR), bidding (AIGB), auction (AIGA), creative generation (AIGC) and data (AIGD).

Generative Recommendation – RecGPT

The “Guess You Like” feature was upgraded using the RecGPT model, a multimodal LLM with hundreds of billions of parameters that can summarize up to 100 k context items and reason over billions of product attributes. It now predicts latent consumer needs (e.g., seasonal allergy medication, outdoor‑activity gear) rather than merely reacting to post‑click or post‑search behavior.

AI‑Generated Bidding – AIGB and LMA2

AIGB (AI‑Generated Bidding) is a new training paradigm introduced in 2023. Alibaba Mom open‑sourced the AIGB benchmark at NeurIPS 2024, hosted a dedicated workshop, and launched a large‑scale automatic bidding competition. The latest AIGB‑R1 version improves advertising ROI by 12 %, while the accompanying LMA2 model scales to a trillion‑parameter level, a ten‑fold increase over its predecessor.

AI‑Generated Video – Taobao Star‑Video 2.0

The group released a short‑video generation model that can produce a 10‑second video in a single inference and a full‑pipeline product‑promotion video of about one minute. Internal tests show that personalized AI videos raise click‑through rates by roughly 17 percentage points.

Research and Community Impact

Alibaba’s “Future Life Lab” consolidates top talent, compute, and data to advance large‑model and multimodal research. In just over a year the lab has published more than ten papers at top international conferences and contributed open‑source benchmarks that have attracted global attention.