Long Context Modeling
PulseAugur coverage of Long Context Modeling — every cluster mentioning Long Context Modeling across labs, papers, and developer communities, ranked by signal.
5 day(s) with sentiment data
-
Kyutai and Tencent release new AI memory and robotics models
Kyutai has released REKI 2.4B VLA, a robotic model featuring 32 memory registers and achieving 90.4% on the LIBERO-Plus benchmark with a latency of 21.8 ms on NVIDIA H100 hardware. Concurrently, Tencent has launched Hy-…
-
New T2Mem Framework Builds LLM Memory During Inference
Researchers have introduced T$^2$Mem, a novel framework for constructing memory in Large Language Models (LLMs) that defers the process to inference time. Unlike traditional methods that build memory offline, T$^2$Mem p…
-
New research questions long-term memory evaluation methods in LLMs
A new paper on arXiv details an audit of long-term memory evaluation methods for retrieval chains. The study found inconsistencies in scoring due to reader variation and the need for repeated judging, with scores fluctu…
-
Mnemon agent uses dual-system approach for LLM long-term memory
Researchers have introduced Mnemon, a novel memory agent designed for long-term context in large language models. Mnemon distinguishes between fast, judgment-based tasks (System 1) and slow, reasoning-based tasks (Syste…
-
AI research tackles long-term memory with faster retrieval and smarter storage
Multiple research papers explore advanced memory systems for AI agents, aiming to overcome the limitations of fixed context windows and improve long-term knowledge retention. One approach, galahad-kv, uses local NVMe di…
-
NylonME memory engine adds 'sleep module' for improved commonsense reasoning
A new memory engine called NylonME has incorporated an "Async Reflection" module, inspired by the human brain's memory consolidation process during sleep. This module allows the system to build connections between dispa…
-
NylonME Memory Engine boosts LoCoMo benchmark recall to 84.6%
The NylonME Memory Engine has significantly improved its performance on the LoCoMo benchmark, jumping from a 47.1% recall rate to 84.6% in just two weeks. This improvement was achieved through a multi-step process that …
-
Developer creates benchmark for AI companion apps to combat affiliate spam
A developer has created a benchmark called companion-bench to evaluate AI companion applications, addressing the lack of objective testing for these apps. Unlike benchmarks that test raw language models, companion-bench…
-
New method supervises LLM agent memory using audit trails
Researchers have introduced Hindsight Memory-PRM, a novel method for supervising memory management in long-horizon Large Language Model (LLM) agents. This approach leverages the audit trail of retrieval hits and answer-…
-
New SearchWiki framework learns to navigate knowledge wikis for active information seeking
Researchers have developed SearchWiki, a framework designed to synthesize a corpus into a structured, navigable knowledge wiki. This system trains an agent, WikiResearcher-9B, to actively seek information through multi-…
-
New benchmarks and retrieval methods advance AI conversational memory utilization · 2 sources tracked
Two new research papers explore advancements in conversational AI memory, focusing on how models utilize and retrieve information from extended dialogue histories. The first paper introduces UTILMEM, a benchmark designe…
-
New CABLE system enhances AI agent long-term memory retrieval
A new research paper introduces CABLE, a system designed to improve long-term memory retrieval for AI agents. CABLE constructs links between memories that are complementary to semantic similarity, aiming to surface evid…
-
ArborMem framework enhances LLM memory for complex conversations
Researchers have introduced ArborMem, a novel memory framework designed for large language models to manage complex conversational states. ArborMem represents conversations as a navigable forest of interaction states, a…
-
New memory framework enhances AI dialogue agents' long-term understanding
Researchers have developed FTA-Mem, a novel memory framework designed to enhance long-term dialogue understanding for emotional support agents. This system addresses the challenge of low-density dialogue by creating str…
-
New research probes context compression in AI agents, finding temporal data loss
A new research paper introduces Salience-Weighted Consolidation (SWC), a framework inspired by sleep-based memory consolidation, to analyze the effectiveness of gist-based context compression in long-horizon language mo…
-
Comprehension Memory slashes LLM context costs, boosting efficiency
A new paper introduces Comprehension Memory (CoMem), a technique designed to significantly reduce the memory and computational costs associated with long-context language models. CoMem operates by caching intermediate l…
-
Huawei open-sources MindMemOS for AI agent memory evolution
Huawei's Noah's Ark Lab has open-sourced MindMemOS, a memory operating layer for AI agents designed to address issues of memory transferability, evolution, and temporal understanding. MindMemOS decouples memory from ind…
-
TrajWiki framework enhances LLM agents with source-grounded memory trajectories
Researchers have introduced TrajWiki, a novel memory framework designed to enhance the long-horizon dialogue capabilities of large language model agents. Unlike existing methods that treat memory as static or overwritab…
-
New benchmark reveals AI abstention capability depends on question distance
A new benchmark, RE-call, has been developed to measure an AI agent's ability to abstain from answering when information is not present in its knowledge base. The benchmark introduces the concept of "excision distances"…
-
InferScale system optimizes LLM serving with reusable KV state
Researchers have developed InferScale, a new GPU-native system designed to enhance the serving of personalized large language models. InferScale addresses the issue of repeated prompt prefilling in memory systems by uti…