Tag
The article explains how agent memory systems can leak data due to post-filter tenant scoping in vector search and recommends scoping at write time to prevent cross-tenant data exposure.
The article describes testing to reduce reading cost in an AI verification pipeline while preserving high recall and minority evidence, noting challenges with current approaches and inviting community input.
The article proposes a Markdown-based memory system for AI agents that uses simple file storage and editing rules to outperform complex memory runtimes, ensuring accurate and sourced fact retrieval.
OntologyAligner presents a three-stage framework combining ontology-aligned retrieval and LLM reranking for biomedical ontology normalization, achieving state-of-the-art performance on HPO tasks and introducing the PhenoNormBench benchmark.
ReGround is a large-scale dataset for grounding reviewer comments in multimodal evidence from scientific papers, revealing the challenges and importance of integrating text, tables, and figures in retrieval tasks.
This paper introduces a framework for multilingual multimodal entity linking that improves accuracy on rare entities by combining reasoning and retrieval, with significant gains on the MERLIN benchmark.
Nebius has launched the AI Builder Program, offering AI builders resources such as runnable examples, blueprints, courses, and over $400 in credits to facilitate building AI systems.
EdgeMem proposes an LLM-free method for agent memory construction and retrieval using a multi-anchor hypergraph, preserving evidence to improve performance in memory-grounded question answering.
The paper introduces PersonaLink, a training-free method that distills user history into a bounded persona, matching retrieval on classification tasks but not on regression, highlighting a task-type asymmetry.
This paper introduces Skill Following and the Retrieval-Invoked Actual-Use Effect to evaluate whether skill retrieval in LLM agents genuinely improves task performance, revealing that aggregate metrics can be misleading.
Garry Tan has released new evaluations for gbrain, an open-source retrieval layer for AI agents, showcasing state-of-the-art performance in memory reading and writing without LLM-in-loop retrieval.
Hi-Q is an evidence-conditioned framework that dynamically refines multi-hop queries into hierarchical trees based on corpus support signals, improving retrieval and answer accuracy in multi-hop question answering.
A tweet discusses the storage concerns of late interaction models and shares interesting results on hierarchical pooling and asymmetric quantization techniques.
A 30-question breakdown explaining how embeddings, vector search, and retrieval systems work, covering similarity metrics, training methods, indexing techniques, and evaluation.
This paper presents a case study on skill discovery and routing in a multimodal agent harness, showing that partial in-prompt exposure of skills can create lexical competition that hinders correct selection, linking small-scale in-context retrieval to large-scale approaches.
Napster has pivoted to an AI agent platform, illustrating how AI training data quickly becomes outdated, leading models to provide confident but incorrect information about company identities.
This article highlights that benchmark scores for Engrava, a memory layer tool, can vary due to reader or judge stage variance even when retrieval results are unchanged, using LongMemEval results to illustrate the issue.
KBMR is an MLLM-based retriever for knowledge-based visual question answering that uses semantic alignment to improve retrieval and VQA accuracy, showing significant gains over CLIP baselines.
Sentence Transformers v6.0 introduces MultiVectorEncoder for ColBERT-style late interaction retrieval, enhancing support for multi-vector embedding models with a familiar API.
This paper investigates the conditions under which skills enhance LLM agent performance, analyzing success and failure factors through controlled experiments and proposing a taxonomy of skill-use modes.