Tag
A discussion criticizing the addictive 'attractor basins' of short-form video and recommendation feeds, highlighting Andrej Karpathy's weekend project HackerTube, which used Claude to filter 43M Hacker News comments down to 6,454 highly-recommended YouTube videos for a much better viewing experience.
Andrej Karpathy used Claude Opus 5.5 to build HackerTube (hackertube.org), a YouTube-style site featuring 6,454 videos recommended across 40K Hacker News comments from 2007-2026, developed end-to-end in about two hours.
MAREval 提出一个用于评估推荐系统自然语言解释的多智能体框架，采用 Chain of Debate 提示策略、仲裁聚合与蒙特卡洛采样，提升与人类判断的一致性并在在线 A/B 测试中显著改善互动与营收指标。
该论文提出 BoLe 双重判定框架，分析 AI 一键生成视频带来的服务型内容涌入如何打破内容生态平衡，并通过简化数学模型论证推荐池机制在内容增长超过阈值后必然失效。
Xiaohongshu (RedNote) 研究团队发布 Vela 框架，将真实用户行为记录转化为训练任务与轨迹，通过监督微调和强化学习提升 LLM 在用户相关任务（+11.5 分）及通用基准（平均 +4 分）上的表现。
QD-Rec 提出一种质量-多样性优化方法，结合上下文老虎机与基础模型，通过档案机制演化生成多样化、有理由支撑的高质量推荐，适用于并购、客户定向等高风险决策场景。
ReMaRk 提出通过将生成列表与所有有效回复的标题联合评分来消除 GRPO 训练中的回复噪声，在保持或超越 Rank-GRPO 性能的同时显著降低 LLM 推荐模型的 RL 训练成本。
NVIDIA recsys-examples now offers an end-to-end HSTU generative recommender inference workflow on Dynamo-Triton, combining PyTorch AOTI compilation with FlexKV-backed KV caching to achieve up to 5.93x lower latency on an RTX PRO 6000 Blackwell GPU.
This paper introduces a synthetic data generation pipeline and self-improvement loop for bootstrapping conversational recommendation agents at Spotify, resulting in significant improvements in user engagement and performance.
Kent Dodds describes how his approach to working with AI agents has shifted from detailed low-level instructions to higher-level oversight, where he relies on agent recommendations for trade-off analysis.
IntBMoE is a novel Mixture-of-Experts method that decouples participation, execution, and materialization costs through block-level conditioning and sparse execution, demonstrating improvements in image classification, language modeling, and recommendation systems, with real-world deployment in AMap's generative recommendation system.
QueryFormer is a unified transformer architecture that won the KDD Cup 2026 Tencent UniRec Challenge for post-click conversion rate prediction, focusing on query generation and efficient scaling.
This paper introduces a hybrid candidate generation approach for vacation rental recommendations, combining collaborative filtering and graph neural networks to improve recall by 14.8% over baseline methods.
Lu Fang and Ilina Mitra from Meta will present a deep dive on building high-performance recommendation inference systems at PyTorch Conference 2026, detailing end-to-end production workflows and advanced optimizations for large-scale platforms.
PRQ-KMeans is a post-hoc tokenization method that improves residual quantization for semantic identifiers by removing global-mean components, refining centroids, and using projection residuals, achieving significant performance gains in industrial search and public recommendation benchmarks.
This paper proposes DASO, a tree-aware post-training method for generative recommendation that addresses difficulty mismatch in GRPO by profiling rollout groups and reallocating based on prefix-match depth, improving performance on public benchmarks.
ERASE introduces a novel training schedule that detaches subgraphs to overlap backward passes with forward work, improving throughput by up to 9.51% in large-scale recommendation systems while preserving model performance.
This paper presents a lifecycle framework for using LLMs as judges to evaluate recommendation explanations at Netflix, covering phases from development to deployment and monitoring, with positive A/B test results showing improved user engagement.
This paper proposes a Unified Sequential Composition Model (USCM) and Latent Expansion Monte Carlo Tree Search (LE-MCTS) for fashion outfit generation, achieving state-of-the-art performance on multiple datasets.
Elon Musk retweeted Keith Coleman's explanation of an algorithm using a simple analogy for a 5-year-old, comparing it to a smart robot selecting toys for a feed.