Live intelligence

Latest AI updates

2 updates
Research Editor selected

RL Post-Training Improves Ranking but Hurts Recall in Generative Recommenders

A new arXiv paper studies generative recommenders that decode a user's next item semantic ID with a language model and are post-trained with reinforcement learning after supervised fine-tuning. Across five datasets, the authors find RL post-training raises hit@K for small K but lowers hit@K when K is in the hundreds, meaning the model becomes a better ranker and a worse retriever. They trace the loss to negative advantages assigned to sampled rollout candidates, since only one next item is logged per user, and propose out-of-rollout anchoring, an added loss term that preserves the SFT model's ordering of items outside the rollout. The remedy restores candidate-list recall to SFT levels on all five datasets while keeping top-of-list gains, and also works under other RL configurations.

Why it matters: It exposes a concrete, previously unmeasured failure mode of RL post-training in recommender systems and offers a simple, tested fix that practitioners can apply.

generative recommendationreinforcement learningrecommender systems
Research Editor selected

RecoGEM: A Serving System for Production-Scale Generative Recommendation

RecoGEM is a production serving system for generative recommendation, where item identifiers are generated autoregressively instead of scoring candidate items. It uses a runtime-engine decoupling architecture to handle GR-specific responsibilities such as candidate-state transitions and live artifact updates, while the execution engine optimizes self-attention, candidate selection, and execution planning for short decoding sequences and large candidate populations. The system has been deployed for over nine months across production lines serving hundreds of millions of users and more than ten GR model families. On OneModel workloads, it reports over 1,000 QPS per serving instance, P99 latency below 35 ms, and a 2.88x throughput improvement over representative baselines.

Why it matters: It offers concrete architecture and production performance details for serving generative recommendation models at industrial scale.

Generative recommendationModel servingRecommender systems