Junior_prompt_engineer

Fast and Faithful: Real-Time Verification for Long-Document Retrieval-Augmented Generation Systems @XunzhuoLiu et al. present a real-time hallucination verifier for RAG that extends encoder context to 32K tokens. 📝 arxiv.org/abs/2603.23508 🤗 huggingface.co/llm-semantic-r…

Reasoning over Semantic IDs Enhances Generative Recommendation Proposes a two-stage framework that enables LLMs to reason over discrete item tokens for generative recommendation, using enriched SID-language alignment and RL. 📝 arxiv.org/abs/2603.23183 👨🏽‍💻 github.com/HappyPointer/S…

OneSearch-V2: The Latent Reasoning Enhanced Self-distillation Generative Search Framework Kuaishou presents a generative search framework that enhances complex query understanding. 📝 arxiv.org/abs/2603.24422 👨🏽‍💻 github.com/benchen4395/on…

🚀 OpenResearcher paper is finally released! 🔥 We explore how to synthesize long-horizon research trajectories for deep-research agents — fully offline, scalable, and low-cost, without relying on live web APIs. 📄 huggingface.co/papers/2603.20… 🧩Two key ideas: Offline Corpus — One-time bootstrapping seeds 10K gold passages + 15M-doc FineWeb corpus. 📚 Explicit Browsing Primitives — Just 3 ops: search / open / find. The agent learns not just what to retrieve, but how to inspect docs and localize evidence at multiple scales. 🔎 📊 Results: 54.8% on BrowseComp-Plus with our 30B-A3B — #1 open-source under the same search engine setup. Beating much larger models like GPT-4.1, Claude-Opus-4, Gemini-2.5-Pro, and DeepSeek-R1. 💡 Insights: Beyond accuracy, we dissect deep research pipeline design—from data filtering and agent configuration to retrieval accuracy dynamics (RQ1-RQ5). Try it yourself: 🛠️ Code: github.com/TIGER-AI-Lab/O… 🤗 Models & data: huggingface.co/collections/TI… 🚀 Demo: huggingface.co/spaces/OpenRes… #llms #agentic #deepresearch #tooluse #opensource #retrieval #SFT

KARMA: Knowledge-Action Regularized Multimodal Alignment for Personalized Search at Taobao Alibaba identifies Semantic Collapse in LLM-based personalized search and proposes train-only decodability regularization to preserve semantic knowledge. 📝 arxiv.org/abs/2603.22779

Autonomous agents for data labeling github.com/HumanSignal/Ad…

awesome-mcp-servers is the only MCP resource list you need covers everything: → browser automation → cloud platforms (AWS, K8s, Cloudflare) → databases, dev tools, file systems → AI agents, search, monitoring & more github.com/punkpeye/aweso…

Zero-Shot Vulnerability Detection in Low-Resource Smart Contracts Through Solidity-Only Training arxiv.org/abs/2603.21058

[CL] Measuring Reasoning Trace Legibility: Can Those Who Understand Teach? D Roytburg, S Sridhar, D Ippolito [CMU] (2026) arxiv.org/abs/2603.20508

A Brief Comparison of Training-Free Multi-Vector Sequence Compression Methods @Robro612 et al. evaluate training-free token pruning vs. pooling for multi-vector retrieval, finding token merging strictly superior for reducing index size. 📝 arxiv.org/abs/2603.22434

Exclusive Self Attention (XSA). paper breakdown: k-a.in/XSA.html

SkillRouter: Retrieve-and-Rerank Skill Selection for LLM Agents at Scale Alibaba proposes a 1.2B retrieve-and-rerank pipeline for selecting skills from ~80K pools, showing that full skill body text is the decisive signal for selection. 📝 arxiv.org/abs/2603.22455

GEM: A Native Graph-based Index for Multi-Vector Retrieval Presents a native graph-based indexing framework for multi-vector retrieval that constructs a proximity graph directly over vector sets and achieves up to 16x speedup. 📝arxiv.org/abs/2603.20336 👨🏽‍💻github.com/sigmod26gem/si…

[AI] Demonstrations, CoT, and Prompting: A Theoretical Analysis of ICL X Tong, Y Zeng, J Zhang [Microsoft Research & University of Wisconsin-Madison] (2026) arxiv.org/abs/2603.19611

Interesting paper: Exclusive Self Attention (XSA). The key idea is simple: force attention to model information orthogonal to self-value. A nice example of how small changes to the attention mechanism can still matter. arxiv.org/abs/2603.09078

let me get you started in local AI and bring you to the edge. if you have a GPU or thinking about diving into the local LLM rabbit hole, first thing you do before any setup is join x/LocalLLaMA. this is the community that will help you at every step. post your issue and we will direct you, debug with you, and save you hours of work. once you're in, follow these three: @TheAhmadOsman the oracle. this is where you consume the latest edges in infrastructure and AI. if something dropped you hear it from him first. his content alone will keep you ahead of most. @0xsero one man army when it comes to model compression, novel quantization research, new tools and tricks that make your local setup better. you will learn, experiment, and discover things you didn't know existed. @Teknium maker of Hermes Agent, the agent i use every day from @NousResearch. from Teknium you don't just stay at the frontier, you get your hands on the tools before everyone else. this is where things are headed. if you follow me follow these three and join the community. you will be ahead of most people in this space. if you run into wrong configs, stuck debugging hardware, or can't get a model to load, post there so we can help. get started with local AI now. not only understand the stack but own your cognition. don't pay openai fees on top of giving them your prompts, your research, and your most valuable thinking to be monitored and metered. buy a GPU and build your own token factory.

So short-sided. At a time when the world is being held hostage by chokepoints in the supply of fossil fuels and the value of autonomous energy is even more obvious, this decision is Luddite.

- One bombed hospital might be a horrible mistake - Two bombed hospitals are undeniably crimes against humanity - More than a hundred bombed hospitals is the work of the most effective and barbaric terrorist organization the world has ever seen.

I am super excited to announce our new work on recursive LLMs for long context reasoning. To solve the problem, we use 1930s math (lambda calculus) and show Uuuuge 🤪 improvements in accuracy and latency. Check out the paper here: arxiv.org/pdf/2603.20105 Code can be found here: github.com/lambda-calculu… #ai #machinelearning

A Super Fast K-means for Indexing Vector Embeddings @LeonardoKuffo et al. introduce a k-means variant that prunes unnecessary dimensions during clustering, achieving faster indexing than FAISS on CPUs and cuVS on GPUs. 📝 arxiv.org/abs/2603.20009 👨🏽‍💻 github.com/cwida/SuperKMe…