The Brutal Aesthetics of Data and Compute: Scaling Laws, Generative AI, and the Evolution of Advertising Systems

Jiang Jie, a Ph.D. and Vice President at Tencent, leads the advertising platform and AI labs, overseeing data, database, machine‑learning, and billing platforms. With over a decade of experience in massive computing, distributed architecture, data mining, and machine learning, he introduced the "Hunyuan" large model in September 2023.

He observes that the recent surge of generative AI is essentially a focused application of the Scaling Law: using vast data and compute with a universal transformer architecture to achieve emergent intelligence, which he describes as the "brutal aesthetics of data and compute".

The Scaling Law, highlighted by Rich Sutton’s "The Bitter Lesson", argues that algorithmic tricks matter less than more data and more compute. Jiang applied this intuition to Tencent’s next‑generation ad system (Ad System 2.0) by building larger models, feeding them more data, and providing stronger compute resources for CTR and CVR prediction.

After the rise of generative AI, he concentrated Tencent’s compute resources into a unified machine‑learning platform, enabling teams to leverage these resources and ultimately develop the "Hunyuan" model, which now powers the newer Ad System 3.0.

Ad System 3.0 aims to give the advertising platform a deeper understanding of ads. By redefining the ad ID hierarchy and consolidating similar ads, the system reduces data sparsity, improves prediction stability, and cuts the number of active ads from 7.7 million to about 700 thousand.

The system also seeks multimodal comprehension—text, images, and video—through the capabilities of the Hunyuan model, allowing it to infer what product an ad sells and which audience it targets.

Looking forward, Jiang suggests that a simple, universal architecture like the transformer should ingest not only ad‑specific signals (clicks, purchases) but also broader web data, enabling future emergent intelligence and higher performance ceilings.

From a management perspective, as models become more capable, the role of human optimizers, designers, and agencies will shift from repetitive tasks to strategic thinking about consumer needs, brand positioning, and commercial models.

He emphasizes that advertising is not an end in itself; the ultimate goal is sales. Brands and agencies must move from focusing on ad delivery to satisfying core consumer demands across product, brand, and business‑model dimensions.

Creative work will increasingly involve humans curating and co‑creating AI‑generated assets, while data completeness across the entire sales funnel becomes essential for model effectiveness.

Finally, Jiang argues that AI will become a management problem: organizations must decide which tasks AI should handle, how AI and humans collaborate, and how to structure mixed AI‑human teams to become "AI‑native" enterprises.