Live intelligence

Latest AI updates

3 updates
Research

StitchRAG Builds Compact RAG Contexts by Stitching Sentences with Text Diffusion

StitchRAG is a framework for constructing compact contexts in retrieval-augmented generation, aiming to cut the computational overhead of long retrieved passages while avoiding noise that can hurt performance. It sets a per-passage size reduction ratio through a utility maximization that trades retained information against computation cost, then extracts answer candidates and query-related keywords as anchors and collects passage sentences containing them. The collected sentences are linked together using text diffusion to form a coherent context. Experiments on five open-domain QA benchmarks report consistent gains over multi-step and compression baselines, with ablations supporting dynamic size allocation, anchor preservation and diffusion-based stitching.

Retrieval-augmented generationContext compressionText diffusion
Research Editor selected

Coca-Cola: Latent Context Compression for LLM Agents

Coca-Cola compresses tool observations into latent tokens embedded in the agent's input space, reducing token consumption by 70-89% and runtime by 24-56% while improving task performance on QA benchmarks. Action-Preserving Training aligns next-action distributions to retain decision-critical information.

Why it matters: Offers a concrete, novel method for agent context compression with significant efficiency gains and open-source code.

LLM agentscontext compressionlatent representation
Research Editor selected

In-Model Context Compression with Agentic Soft Tokens for Long-Horizon LLM Agents

The paper introduces Agentic Soft-Token Compression (ASTC), a training-free method that uses soft tokens within an LLM agent to compress accumulated context for long-horizon tasks. ASTC leverages the model's native recurrent dynamics and hybrid Gated Delta Network (GDN)-attention to carry forward absorbed information as agentic soft tokens while retaining exact values as hard tokens. Experiments show ASTC matches or exceeds uncompressed accuracy with up to 50.68% fewer attention pairs, and trained compressors require 15.4 to 103.7 times more attention pairs per successful task.

Why it matters: This research presents a novel, training-free approach to context compression that significantly reduces computational cost while maintaining accuracy, offering practical value for deploying LLM agents in long-horizon tasks.

LLM agentscontext compressionsoft tokens