How AI4Data Is Revolutionizing Large‑Model Data Production

The global surge of large‑model AI demands high‑quality, massive training data, yet traditional Data4AI pipelines struggle with efficiency and quality. Jiang Qian, a senior data‑R&D expert at Shanghai AI Lab, will present how his team redesigns data production using a "Data+AI" architecture.

Key points of the talk include:

Evolution of data production: a review of how data‑generation methods have changed with model scale and the current bottlenecks.

Paradigm shift from Data4AI to AI4Data: why AI must be embedded in the data pipeline to move from manual‑intensive to intelligent‑driven workflows.

AI toolbox across the full data chain: intelligent solutions for data collection, ingestion, classification, labeling, synthesis, retrieval, training, and storage.

Data‑warehouse thinking: applying logical layering and domain‑centric organization to make massive datasets searchable, understandable, and reusable.

Future of AI4Data: emerging directions for fully automated, self‑adapting data ecosystems.

Two major challenges are addressed:

Production efficiency: a cloud‑native architecture enables automated scheduling and elastic resource scaling, ensuring rapid, continuous data supply.

Data usability: AI‑driven governance (smart classification, labeling, quality assessment, semantic search) combined with data‑warehouse models guarantees that large volumes of data become easily accessible assets for developers, researchers, and evaluators. The session is valuable for data platform engineers, algorithm researchers, and technical decision‑makers interested in boosting data productivity and governance for large‑model training.