In an invited keynote at the 2022 Workshop on Online Recommender Systems and User Modeling (ORSUM), Eugene Yan posed a pointed question to the recommender systems community: is the juice worth the squeeze? The talk pushes back against the assumption that ever-more-sophisticated models are inherently better, asking teams to account for the engineering effort, maintenance burden, and infrastructure costs that come with complexity.
Yan's argument centers on the gap between offline performance and real-world value. A model that looks better in benchmarks may not translate into a better user experience, and the cost of deploying and serving it can outweigh the benefit. He suggests that simpler approaches, including well-tuned baselines, often deliver most of the value at a fraction of the cost.
Because the source is a single keynote, there is no opposing view to compare. The talk is best read as a challenge to the field's default instincts: before adding another layer of complexity, measure what it actually buys you—and be honest about whether the squeeze is worth it.