Search: “context”

Real Long-Term Memory for AI: A 50-Million-Token Window That Is Faster and Cheaper Than Recompute

arXiv:2610.10845v1 Announce Type: cross Abstract: A large language model can only use the text that fits in its context window, and it recomputes its internal key-value (KV) state for a prompt every t

9 Oct 2026

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Research

VFold: Symmetry-Aware Cross-Layer Value Cache Compression

DGX agent

arXiv:2610.12338v1 Announce Type: new Abstract: While caching key-value (KV) states accelerates Large Language Model (LLM) decoding, this cache can dominate memory usage at long context lengths. One s

researcharxiv-cs-cl

9 Oct 2026

CoDR: Training-Free Confidence-Drift Remasking for Diffusion Language Models

arXiv:2610.08833v1 Announce Type: new Abstract: Masked diffusion language models (MDLMs) decode by repeatedly committing tokens to masked positions, but these commitments are usually irreversible. A t

researcharxiv-cs-cl

Halogen + Qwen Flash Next keeps getting better

With latest Halogen version update (0.17.2), decode is consistently at ~45 tps even at high context with Qwen 3.8 Flash Next on a 128GB Strix Halo. This is some great work u/peonist-ai. Have been pump

KVFetch: Temporal Prefetching for the Missing Half of KV Cache Compression

arXiv:2610.08811v1 Announce Type: new Abstract: As context windows scale to tens or hundreds of thousands of tokens, KV cache compression has become essential for efficient LLM inference. Existing met

PHRBench: A Behavioral Evaluation of Post-Hallucination Reasoning in LLMs

arXiv:2610.10455v1 Announce Type: new Abstract: Hallucinated information can propagate through multi-stage LLM systems and become part of the context for subsequent reasoning. Existing studies of post

The Dichotomy Between Pattern Recognition and Step-by-Step Reasoning

arXiv:2610.09186v1 Announce Type: new Abstract: We argue that pattern recognition and step-by-step reasoning are two ends of a spectrum. A large language model (LLM) learns to reason step-by-step when

tutorialsarxiv-cs-lg

Agents

Agent Lightning v1.0: A 3,500-Line Lightweight Agentic RL Framework for Training Agents with Real Harnesses

Training AI agents with reinforcement learning can be challenging because their tools, context, and decision-making are managed by complex frameworks. Agent Lightning connects existing agents to RL tr

agentsmicrosoft-research

b11474

model-releasesllama-cpp-releases

feat: add GLM5Next MTP, optimize (#29928) llama : add GLM5-Next NextN (MTP) graph Build the GLM5-Next multi-token-prediction head as graph_mtp: the NextN block embeds enorm(tok)+hnorm(h) through eh_pr

Capacity, Responsiveness and Alignment: What Makes a Latent Structure Actionable

arXiv:2610.06897v1 Announce Type: new Abstract: Localizing latent structures in the activation space of language models (LMs) is central to understanding and controlling their behavior. Yet, localized

local-aiarxiv-cs-cl

Defense-in-Depth for LLMs: Evaluating Memory Gates Against Activation-Induced and Memory-Induced Sycophancy

arXiv:2610.07403v1 Announce Type: new Abstract: Long-term memory allows Large Language Models (LLMs) to maintain personalized context across interactions, but retrieved user history can induce memory-

Leveraging a four-quadrant approach for evaluating Redpine Science

arXiv:2610.07937v1 Announce Type: new Abstract: Redpine Science gives models and agents a single access point to a wide range of peer-reviewed literature, queried directly through the Model Context Pr

Agents

Persistent Memory in Multi-Agent LLM Inference: What It Costs, What It Buys, and When You Can Tell

arXiv:2610.07782v1 Announce Type: new Abstract: Decomposing long-context inference across cooperating agents bounds the active KV cache per call rather than total evidence, which matters when KV-cache

agentsarxiv-cs-ai

Valid for Free: Homophily-Gated Conformal Prediction for Training-Free Node Classification with Tabular Foundation Models

arXiv:2610.08564v1 Announce Type: new Abstract: Tabular foundation models (TFMs) can classify the nodes of a graph without training on it, by reading node and neighborhood features as table rows next

When Old Facts Return: Re-Reads, Reverts, and the Limits of Temporal Memory

arXiv:2610.07715v1 Announce Type: cross Abstract: A memory system can retire an obsolete value and later restore it merely because the same old statement appears again. A re-read of an old source and

researcharxiv-cs-cl

Why is ik_llama.cpp said to be faster than Mainline? On my hybrid multi-GPU rig, Mainline easily beats it

I constantly see recommendations saying that ik_llama.cpp (ikawrakow's fork) is the undisputed king of hybrid CPU/GPU offloading and MoE performance. However, every time I benchmark it against mainlin

let me introduce you to the langchain stack 1. langgraph gives you this level of control 2. traces give you the complete run lineage 3. vers…

model-releasesharrison-chase--x

let me introduce you to the langchain stack 1. langgraph gives you this level of control 2. traces give you the complete run lineage 3. version your skills and instructions in context hub 4 & 5. deepa

Agents

Managed Apache Iceberg at scale: How Spanner powers Lakehouse runtime catalog

As customers modernize to lakehouse architectures, they are standardizing on open formats such as Apache Iceberg to create a shared data estate across compatible engines. This enables you to build AI-

agentsgoogle-cloud-ai

Modeling Deletion Requests in Machine Unlearning

arXiv:2610.04310v1 Announce Type: new Abstract: Machine unlearning is seen as a promising approach to enable users to exercise the 'right to erasure' in the context of AI models. We ask how users migh

Multi-Task Partially Supervised Learning for Super-Resolution and Semantic Segmentation on Earth Observation data

arXiv:2610.06389v1 Announce Type: new Abstract: Super-resolution and semantic segmentation are known to benefit one another, especially in the Earth observation context. However, learning both tasks i

researcharxiv-cs-cv

Pythia: Toward Foundation World Models for Multimodal Time Series

arXiv:2610.05240v1 Announce Type: new Abstract: Time-series foundation models offer a unified approach to forecasting across heterogeneous domains. Textual context and auxiliary observations provide c

tutorialsarxiv-cs-lg

ReMAP: Restoring the Perceptual Cycle with Reasoning-Time Latent Visual Memory

model-releasesarxiv-cs-cv

arXiv:2610.05097v1 Announce Type: new Abstract: As multimodal large language models (MLLMs) reason for longer, attention to the initial visual input diminishes, weakening visual grounding. Visual memo

Time-series Foundation Models for Predictive Control: The Role of Excitation

arXiv:2610.06447v1 Announce Type: new Abstract: Deploying model predictive control (MPC) requires constructing or identifying a predictive model for each target system. Time-series foundation models (

Towards Looped Models Done Right, Part II: Rethinking at Fixed Points

arXiv:2610.06833v1 Announce Type: new Abstract: Every recurrence of a looped language model adds cost in training, decoding, prefill, and reinforcement learning (RL). The closer recurrent states get t

tutorialsarxiv-cs-lg

b11412

model-releasesllama-cpp-releases

llama : fix unexpected graph reallocation in the k-pool models (#29958) llama : fix unexpected graph reallocation in the k-pool models Both k-pool models built a graph shape that depends on state the

BitNest: Bit-Nested Speculative Decoding for Memory-Efficient LLM Inference Acceleration

arXiv:2610.02800v1 Announce Type: new Abstract: Speculative decoding accelerates autoregressive generation by using a lightweight draft to propose multiple tokens for parallel verification. However, e

CONTRA: Discovering and Qualifying Behavior-Changing Questions for Selective Clarification in LLM Code Generation

arXiv:2610.01769v1 Announce Type: cross Abstract: Coding agents can generate code that appears correct but implements behavior the user never intended. This mismatch can arise when an agent silently r

CORE: COverage CAlibration and Evicted-Mass REdistribution for KV Cache

arXiv:2610.02235v1 Announce Type: cross Abstract: Long-context decoding is increasingly constrained by key--value (KV) cache memory and bandwidth. Existing fixed-budget compression methods typically s

researcharxiv-cs-ai

Event-guided Neural Video Compression

arXiv:2610.02265v1 Announce Type: cross Abstract: Neural video codecs derive motion and temporal contexts mainly from RGB frames, leaving room for cross-modal guidance from complementary temporal obse

researcharxiv-cs-cv

Infermeld: a Linux kit for running one GGUF across AMD + NVIDIA GPUs with llama.cpp

Following on from club-5060ti and club-rdna16, I’ve put together Infermeld: a small, open-source Linux companion kit for running one GGUF across an AMD GPU and an NVIDIA GPU, powered by llama.cpp. I’m

Learning a Fact Is Not Learning How to Retrieve It

arXiv:2610.03251v1 Announce Type: new Abstract: A model trained on 'The capital of X is Y' may produce 'Y' after 'The capital of X is' but fail after 'The capital of X:'. We call these different ways

tutorialsarxiv-cs-ai

LiteEMG-FM: An Efficient and Deployable Foundation Model for Robust EMG Sensing

arXiv:2610.02497v1 Announce Type: new Abstract: Electromyography (EMG) signals vary substantially across individuals, body regions, recording sessions, and sensing hardware, limiting the generalizatio

local-aiarxiv-cs-lg

Hardware

Practical limit hit. Decoding so fast that tool calls (cpu) starting to become real limit not decode or prefill. Single RTX5090. Porting Kenshi to Godot project.

Hi folks, LIVE PROJECT PAGE TLDR: Moral of the story. You need better CPU to do actual agentic coding doing real work... I've been on a mission to make my RTX5090 go brrr for past 2 months so much so

hardwarer-localllama

Sentry: Learning to Recover from LLM Agent Failures at Test Time

arXiv:2610.02994v1 Announce Type: cross Abstract: LLM agents often fail mid-task due to invalid tool calls, repeated actions, or poorly grounded reasoning, and learning from these failures is a path t

Unmasking Propaganda: A Comparative Analysis of Masked and Causal Language Models

arXiv:2610.03077v1 Announce Type: new Abstract: Propaganda detection is an essential task in natural language processing (NLP), particularly in the context of manipulative political communications. Ho

Built a quick, sub-15ms Rust CLI/TUI to pack repos into prompts without burning 40k tokens on lockfiles and junk

Whenever I feed codebases into local models (Qwen, DeepSeek R1) or API models, the biggest annoyance is prompt pollution: - Lockfiles (`Cargo.lock`, `package-lock.json`) burning 30,000+ tokens for zer

4 Oct 2026

Anyworld, a self-hosted multiplayer text RPG where a local LLM is the Dungeon Master

Hey everyone, I’ve been working on a game called Anyworld. It’s a browser-based multiplayer (single player also supported) text adventure inspired by the early days of AI Dungeon, especially its brows

3 Oct 2026

LiquidAI/LFM2.5-Encoder 250M/350M

LFM2.5-Encoder-350M is a multilingual bidirectional encoder built on the LFM2 architecture — a larger encoder for maximum downstream quality. It is a masked language model with full bidirectional atte

3 Oct 2026

The cognitive cost of constantly adapting to ChatGPT is starting to add up : cognitive scientist/PhD student lived experience

I use ChatGPT a lot. Probably a lot more than is reasonable. I read David Robinson’s Atlantic piece today about leaving OpenAI and his criticism of its culture of trial-and-error / “iterative deployme

local-air-chatgpt

3 Oct 2026

Counterfactual Auditing of Bias in Open-Source Large Language Models for Clinical Triage

arXiv:2610.01963v1 Announce Type: new Abstract: Emergency department (ED) triage is a high-stakes prioritization task in which demographic, socioeconomic, and system-context information may improperly

local-aiarxiv-cs-ai

Do Your Own Research: Learning to Forecast by Learning to Search

model-releasesarxiv-cs-lg

arXiv:2610.01955v1 Announce Type: new Abstract: Outcome-based reinforcement learning can train language models to forecast real-world events, but prior forecasting work either freezes research context

DuplexSpeechBench-Document Grounding: Benchmarking Document Grounding and Hallucinations in Voice Agents

arXiv:2610.00316v1 Announce Type: cross Abstract: Voice agents enable low-latency, natural interaction, yet their ability to faithfully ground responses in external documents remains underexplored. We

I spent 3 weeks testing local Qwen3.8 on the new low-latency SGLang/vLLM recipes: DFlash2 2.8x. Builds: RadixArk + Inferact 27B NVFP4, 27B BF16, orcarouter 27B Uncensored, Flash-Next NVFP4

Hey guys, Last time I tested Qwen3.8-Flash-Next on its own. This time I put three Qwen3.8 checkpoints through the same 10 tests on the same RTX PRO 6000: RadixArk/Qwen3.8-27B-NVFP4 (dense) orcarouter/

OpenMTB-Audit: Exposing Over-Refusal and Clinical Expert Perspectives in LLM-Based Molecular Tumor Board Safety Evaluation

arXiv:2610.01497v1 Announce Type: new Abstract: Molecular tumor boards integrate genomic findings, clinical context, and therapeutic evidence to support precision oncology. As AI enters this workflow,

Safety

Argus: Academic Integrity in the Era of Generative AI

arXiv:2609.36073v1 Announce Type: cross Abstract: The rapid proliferation of large language models (LLMs) in the context of education has introduced significant challenges in enforcement of academic i

safetyarxiv-cs-ai

Coverage Before Control: Route-Instruction Grounding and Steering for Controllable Retrosynthesis

arXiv:2609.39955v1 Announce Type: new Abstract: Single-step retrosynthesis models are commonly evaluated by their ability to recover recorded reactions. In practice, chemists may need to choose among

tutorialsarxiv-cs-ai

Index-Translate: A Multilingual Translation Model Family -- Text, Speech, Controlled Dubbing, and Long-Document Translation

arXiv:2609.40181v1 Announce Type: new Abstract: We introduce Index-Translate, a multilingual translation model family that combines a shared multilingual foundation with specialized training for gener

Safety

Over-Personalization Is a Decision Failure: Generation-Induced Apply Bias in LLMs

arXiv:2609.34284v2 Announce Type: replace Abstract: Personalized LLMs must decide, for each stored preference, whether the current context calls for applying or suppressing it, which we call its appli

safetyarxiv-cs-cl