HuaRenCa
Back to Forum
Community

Is traditional search, ads, and recommendation heavily impacted by AI?

淡淡阳光味
淡淡阳光味

1 week ago

I mean the whole setup of feature engineering, extracting feature signals, training models, and serving online/offline recommendation signals.

I feel like frontend and backend for traditional CRUD business is definitely dead and buried. When I talked to someone about this search/ads/recommendation stack, they said this stuff relies on inspiration to get the right signals. Is that really true? Or is it just AI brute-forcing miracles?

7
1.9K

Comments (7)

Your avatar
Sign in to comment
jackmolik
jackmolik1 week ago

First, define whether "impact" means job cuts or changes in the nature of the work—the conclusions will be entirely different.

jyh0426
jyh04261 week ago

The feature engineering part did get squeezed, but engineering constraints like online-offline consistency, latency budgets, and cold starts are still there—no matter how strong the model is, someone still has to have its back. The so-called "relying on inspiration" is more like experienced judgment in making trade-offs within limited signals; it doesn't conflict with brute-forcing miracles—they're two different things.

overseaservice
overseaservice1 week ago

Which specific scenarios do you mean by traditional search, advertising, and recommendation? The signal sources for e-commerce and content distribution differ greatly, so it would be best to add more details before making a judgment.

nex4d
nex4d1 week ago

Short-term costs may not necessarily fall; the compute overhead of tuning models and retraining could be even higher.

jia
jia1 week ago

What actually gets replaced is repetitive parameter tuning and hand-crafted features; business understanding and goal definition remain scarce. But in scenarios with poor data quality and sparse feedback, AI struggles to deliver stable returns.

dalong
dalong1 week ago

The hiring pace for this type of role can differ significantly between major domestic tech companies and smaller overseas teams, so it's hard to generalize.

liujibran
liujibran1 week ago

When you go live, feature backfilling and aligning historical data are where things most often go wrong—don't underestimate the effort involved.