Dark | Light
[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]

[@_reachsumit](/creator/twitter/_reachsumit)
"Haystack Engineering: Context Engineering for Heterogeneous and Agentic Long-Context Evaluation @Mufei_Li et al. present a benchmark built on Wikipedia's hyperlink network"  
[X Link](https://x.com/_reachsumit/status/1976480398659158396) [@_reachsumit](/creator/x/_reachsumit) 2025-10-10T02:50Z 3371 followers, XXX engagements


"OneRec-Think: In-Text Reasoning for Generative Recommendation Kuaishou integrates dialogue reasoning and personalized recommendation"  
[X Link](https://x.com/_reachsumit/status/1977972340861579713) [@_reachsumit](/creator/x/_reachsumit) 2025-10-14T05:38Z 3371 followers, XXX engagements


"HoMer: Addressing Heterogeneities by Modeling Sequential and Set-wise Contexts for CTR Prediction Meituan presents a unified transformer that jointly models panoramic sequences and set-wise item interactions for click-through rate prediction. 📝"  
[X Link](https://x.com/_reachsumit/status/1977975410718847366) [@_reachsumit](/creator/x/_reachsumit) 2025-10-14T05:51Z 3370 followers, XXX engagements


"Differentiable Fast Top-K Selection for Large-Scale Recommendation Kuaishou introduces a differentiable Top-K operator with linear O(n) time complexity for cascade ranking systems. 📝 👨🏽💻"  
[X Link](https://x.com/_reachsumit/status/1977978132725456936) [@_reachsumit](/creator/x/_reachsumit) 2025-10-14T06:01Z 3371 followers, XXX engagements


"Mixture of Experts Approaches in Dense Retrieval Tasks Presents a modular framework that integrates a single Mixture-of-Experts block into dense retrieval models to enhance performance particularly for lightweight models. 📝 👨🏽💻"  
[X Link](https://x.com/_reachsumit/status/1980095582883733998) [@_reachsumit](/creator/x/_reachsumit) 2025-10-20T02:15Z 3368 followers, XXX engagements


"Enhance Large Language Models as Recommendation Systems with Collaborative Filtering Integrates collaborative filtering into LLMs for recommendations without fine-tuning using a separate Critic model to provide feedback and refine outputs. 📝"  
[X Link](https://x.com/_reachsumit/status/1980095836471349355) [@_reachsumit](/creator/x/_reachsumit) 2025-10-20T02:16Z 3367 followers, XXX engagements


"GRank: Towards Target-Aware and Streamlined Industrial Retrieval with a Generate-Rank Framework Kuaishou presents a structured-index-free retrieval framework combining target-aware generation with lightweight ranking in billion-scale industrial systems"  
[X Link](https://x.com/_reachsumit/status/1980096288462114972) [@_reachsumit](/creator/x/_reachsumit) 2025-10-20T02:18Z 3371 followers, XXX engagements


"MTmixAtt: Integrating Mixture-of-Experts with Multi-Mix Attention for Large-Scale Recommendation Meituan proposes a unified MoE architecture with Multi-Mix Attention for large-scale recommendation featuring automatic token clustering. 📝"  
[X Link](https://x.com/_reachsumit/status/1980096626988659131) [@_reachsumit](/creator/x/_reachsumit) 2025-10-20T02:19Z 3368 followers, XXX engagements


"Structure-R1: Dynamically Leveraging Structural Knowledge in LLM Reasoning through Reinforcement Learning @jlwu002 transforms retrieved content into structured representations optimized for reasoning using RL"  
[X Link](https://x.com/_reachsumit/status/1980097061187112973) [@_reachsumit](/creator/x/_reachsumit) 2025-10-20T02:21Z 3371 followers, 2735 engagements


"How role-play shapes relevance judgment in zero-shot LLM rankers Reveals that role-play prompts significantly impact zero-shot LLM ranking performance with role-play signals predominantly encoded in early layers. 📝 👨🏽💻"  
[X Link](https://x.com/_reachsumit/status/1980505563143500285) [@_reachsumit](/creator/x/_reachsumit) 2025-10-21T05:24Z 3369 followers, XXX engagements


"Towards Context-aware Reasoning-enhanced Generative Searching in E-commerce Kuaishou introduces a context-aware reasoning-enhanced generative search framework that unifies heterogeneous contexts and uses self-evolving post-training. 📝"  
[X Link](https://x.com/_reachsumit/status/1980506719517372811) [@_reachsumit](/creator/x/_reachsumit) 2025-10-21T05:29Z 3370 followers, XXX engagements


"Enterprise Deep Research: Steerable Multi-Agent Deep Research for Enterprise Analytics @aksh_555 et al. at Salesforce introduce a multi-agent deep research system to generate comprehensive research reports"  
[X Link](https://x.com/_reachsumit/status/1980508262110998804) [@_reachsumit](/creator/x/_reachsumit) 2025-10-21T05:35Z 3370 followers, XXX engagements


"Improving Visual Recommendation on E-commerce Platforms Using Vision-Language Models Mercari fine-tunes SigLIP on product image-title pairs to get XXX% offline improvement and XX% CTR increase in production for visual similarity-based recommendations. 📝"  
[X Link](https://x.com/_reachsumit/status/1978704533145477218) [@_reachsumit](/creator/x/_reachsumit) 2025-10-16T06:08Z 3371 followers, XXX engagements


"LIME: Link-based User-Item Interaction Modeling with Decoupled XOR Attention for Efficient Test Time Scaling Meta uses link embeddings and XOR attention to achieve cross-attention expressiveness with two-tower efficiency enabling 10x inference speedup"  
[X Link](https://x.com/_reachsumit/status/1980826713291714732) [@_reachsumit](/creator/x/_reachsumit) 2025-10-22T02:41Z 3371 followers, XXX engagements


"Search Self-play: Pushing the Frontier of Agent Capability without Supervision Alibaba introduces a self-play training method for deep search agents where the model acts as both question proposer and solver. 📝 👨🏽💻"  
[X Link](https://x.com/_reachsumit/status/1980827675167281361) [@_reachsumit](/creator/x/_reachsumit) 2025-10-22T02:44Z 3371 followers, XXX engagements


"Lost in the Maze: Overcoming Context Limitations in Long-Horizon Agentic Search Separates search and browse tools with periodic summarization to enable longer more efficient web searches while managing context constraints. 📝 👨🏽💻"  
[X Link](https://x.com/_reachsumit/status/1981196650900312210) [@_reachsumit](/creator/x/_reachsumit) 2025-10-23T03:11Z 3371 followers, XXX engagements

[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]

@_reachsumit "Haystack Engineering: Context Engineering for Heterogeneous and Agentic Long-Context Evaluation @Mufei_Li et al. present a benchmark built on Wikipedia's hyperlink network"
X Link @_reachsumit 2025-10-10T02:50Z 3371 followers, XXX engagements

"OneRec-Think: In-Text Reasoning for Generative Recommendation Kuaishou integrates dialogue reasoning and personalized recommendation"
X Link @_reachsumit 2025-10-14T05:38Z 3371 followers, XXX engagements

"HoMer: Addressing Heterogeneities by Modeling Sequential and Set-wise Contexts for CTR Prediction Meituan presents a unified transformer that jointly models panoramic sequences and set-wise item interactions for click-through rate prediction. 📝"
X Link @_reachsumit 2025-10-14T05:51Z 3370 followers, XXX engagements

"Differentiable Fast Top-K Selection for Large-Scale Recommendation Kuaishou introduces a differentiable Top-K operator with linear O(n) time complexity for cascade ranking systems. 📝 👨🏽💻"
X Link @_reachsumit 2025-10-14T06:01Z 3371 followers, XXX engagements

"Mixture of Experts Approaches in Dense Retrieval Tasks Presents a modular framework that integrates a single Mixture-of-Experts block into dense retrieval models to enhance performance particularly for lightweight models. 📝 👨🏽💻"
X Link @_reachsumit 2025-10-20T02:15Z 3368 followers, XXX engagements

"Enhance Large Language Models as Recommendation Systems with Collaborative Filtering Integrates collaborative filtering into LLMs for recommendations without fine-tuning using a separate Critic model to provide feedback and refine outputs. 📝"
X Link @_reachsumit 2025-10-20T02:16Z 3367 followers, XXX engagements

"GRank: Towards Target-Aware and Streamlined Industrial Retrieval with a Generate-Rank Framework Kuaishou presents a structured-index-free retrieval framework combining target-aware generation with lightweight ranking in billion-scale industrial systems"
X Link @_reachsumit 2025-10-20T02:18Z 3371 followers, XXX engagements

"MTmixAtt: Integrating Mixture-of-Experts with Multi-Mix Attention for Large-Scale Recommendation Meituan proposes a unified MoE architecture with Multi-Mix Attention for large-scale recommendation featuring automatic token clustering. 📝"
X Link @_reachsumit 2025-10-20T02:19Z 3368 followers, XXX engagements

"Structure-R1: Dynamically Leveraging Structural Knowledge in LLM Reasoning through Reinforcement Learning @jlwu002 transforms retrieved content into structured representations optimized for reasoning using RL"
X Link @_reachsumit 2025-10-20T02:21Z 3371 followers, 2735 engagements

"How role-play shapes relevance judgment in zero-shot LLM rankers Reveals that role-play prompts significantly impact zero-shot LLM ranking performance with role-play signals predominantly encoded in early layers. 📝 👨🏽💻"
X Link @_reachsumit 2025-10-21T05:24Z 3369 followers, XXX engagements

"Towards Context-aware Reasoning-enhanced Generative Searching in E-commerce Kuaishou introduces a context-aware reasoning-enhanced generative search framework that unifies heterogeneous contexts and uses self-evolving post-training. 📝"
X Link @_reachsumit 2025-10-21T05:29Z 3370 followers, XXX engagements

"Enterprise Deep Research: Steerable Multi-Agent Deep Research for Enterprise Analytics @aksh_555 et al. at Salesforce introduce a multi-agent deep research system to generate comprehensive research reports"
X Link @_reachsumit 2025-10-21T05:35Z 3370 followers, XXX engagements

"Improving Visual Recommendation on E-commerce Platforms Using Vision-Language Models Mercari fine-tunes SigLIP on product image-title pairs to get XXX% offline improvement and XX% CTR increase in production for visual similarity-based recommendations. 📝"
X Link @_reachsumit 2025-10-16T06:08Z 3371 followers, XXX engagements

"LIME: Link-based User-Item Interaction Modeling with Decoupled XOR Attention for Efficient Test Time Scaling Meta uses link embeddings and XOR attention to achieve cross-attention expressiveness with two-tower efficiency enabling 10x inference speedup"
X Link @_reachsumit 2025-10-22T02:41Z 3371 followers, XXX engagements

"Search Self-play: Pushing the Frontier of Agent Capability without Supervision Alibaba introduces a self-play training method for deep search agents where the model acts as both question proposer and solver. 📝 👨🏽💻"
X Link @_reachsumit 2025-10-22T02:44Z 3371 followers, XXX engagements

"Lost in the Maze: Overcoming Context Limitations in Long-Horizon Agentic Search Separates search and browse tools with periodic summarization to enable longer more efficient web searches while managing context constraints. 📝 👨🏽💻"
X Link @_reachsumit 2025-10-23T03:11Z 3371 followers, XXX engagements

creator/twitter::129969370/posts
/creator/twitter::129969370/posts