The paper proposes Randomized YaRN, a training method that improves length generalization in large language models by exposing them to out-of-distribution positional representations during training on…

The paper proposes Periodic RoPE (P-RoPE) combined with a dual-layer attention mechanism to overcome the positional encoding limitations of LLMs, enabling theoretically infinite context understanding.

LongAttnComp introduces a novel, two-stage fine-tuning framework for context compression that significantly improves long-context reasoning performance, matching or exceeding full-context accuracy on…

This paper introduces Test-Time-Training Robot Policies (RoboTTT), a robot model and training recipe that scales visuomotor context to 8K timesteps, enabling new capabilities like one-shot imitation a…

This paper proposes RECONTEXT, a training-free inference method for improving long-context reasoning in large language models using model-internal relevance signals and recursive evidence replay.

This paper proposes LBR, a framework to mitigate length bias in large language model-based recommendation systems.

This paper introduces BamiBERT, a new Vietnamese language model based on BERT that addresses limitations of PhoBERT and sets a new state-of-the-art among base-sized Vietnamese encoders.

This paper reports controlled experiments on prompt-design decisions for instruction-following and context length in AI models, finding significant degradation in performance beyond certain thresholds…

KACE introduces a novel knowledge-adaptive context engineering framework that separates knowledge storage from usage, significantly improving mathematical reasoning accuracy on challenging benchmarks…

This study re-evaluates LLM package hallucination rates on a new cohort of frontier models, finding a significant reduction in overall hallucination rates but identifying a persistent, model-agnostic…

This paper proposes a context engineering approach that scales memory with distinct information instead of tokens, using a novelty-gated cache and state-space summary.

The paper introduces RHELM, a new benchmark designed to test LLMs' long-term memory by simulating realistic, complex, and evolving dialogues that integrate multiple heterogeneous data sources.

This paper identifies and describes five recurring architectural patterns and four anti-patterns for MCP servers, a standardized interface for connecting large language models to external tools and se…

This paper demonstrates that retrieval-augmented in-context learning systems for document QA are vulnerable to membership inference attacks, proposing novel black-box methods that exploit query prefix…

The paper introduces the Sovereign Context Protocol (SCP), an open-source, attribution-aware data access layer designed to standardize how Large Language Models (LLMs) connect to and track usage of hu…

The paper introduces a Contextual Integrity (CI) framework and a new benchmark (DelegateCI-Bench) to rewrite user queries sent to cloud LLMs, ensuring only task-essential information is retained while…

This paper presents an efficient algorithm for right-to-left sequence prediction based on a new complexity measure called arithmetic repetition complexity, and demonstrates its application to predicti…

This paper compares the preservation of linguistic content in different text encodings (tokens, bytes, pixels) using a shared bottleneck, revealing their distinct strengths in surface form preservatio…

This paper proposes SelfCompact, a scaffold that allows models to decide when and how to compact long agent traces, improving summarization at a lower token cost.