All papers

The paper tests whether role-playing language models leak private or unobserved character information across simulated interactive scenes.

The paper introduces Memory Forcing, partitioning cache into archive and working banks to prevent mid-horizon history loss in streaming video generation.

Introduces a benchmark evaluating how well vision-language models predict unseen physical outcomes following scene interventions.

Introduces a framework allowing agents to rewind long-horizon environments to earlier states while retaining learned exploratory knowledge.

Investigates whether frozen recurrent networks can provide sufficient memory for partially observable reinforcement learning without end-to-end training.

The paper analyzes gradient dynamics in top-k distillation, showing that renormalization prevents optimization along directions controlling tail token probabilities.

This paper presents a case study of an interpreter co-creating a local computer-assisted interpreting tool using large language models.

The paper examines machine grading on undergraduate essays across three universities, finding models compress score distributions and vary across institutions.

Introduces a differentiable token merging operator running in linear time with an exact sequence output budget.

Introduces an instruction-following benchmark and training recipe to improve goal adherence in full-duplex conversational models.

Proposes an evaluation framework to measure how LLM agents handle uncertainty and resolve conflicting knowledge during execution.

The paper analyzes how pruning and subsequent recovery stages interact, finding instruction-following collapses earlier than domain knowledge during model compression.

The paper introduces BRIDGE, a runtime intervention framework that detects and repairs broken information handoffs across dependency graphs in multi-agent workflows.

Proposes a graph-based token pruning method inspired by sheaf geometry to reduce prefill computation in multimodal large language models.

The paper meta-analyzes 25 years of automated verbal deception detection models and identifies a persistent accuracy ceiling around 70 to 75 percent.

Presents an execution-based evaluation pipeline using browser agents to verify language model solutions for no-code software bug reports.

Presents a multi-agent LLM framework that automatically tunes specific internal algorithm components while preserving overall neural network structures.

The tutorial examines machine learning engineering agents from a data management view, highlighting challenges in pipeline execution, data discovery, and debugging.

The paper tracks automatic speech recognition performance from 2021 to 2025, finding persistent accuracy disparities between standard American English and minority accents.

The paper presents an elastic vision architecture that dynamically reconfigures width and supports partitionable inference across heterogeneous edge devices.

The paper introduces a 7,000-task benchmark to evaluate how multimodal models selectively use or ignore conversational context.

Proposes a reinforcement learning framework that tracks external versus internal knowledge sources to reduce search agent fabrication.

Introduces a key-value cache compression method representing attention activations as sums of compact Kronecker products.

This paper presents a three-stage generation and repair framework to resolve API constraints and tensor shape errors in deep learning programs.

This paper audits gender and stereotypical bias in multilingual job-matching embedding models using robust statistical tests.

The paper evaluates how language models correct user misconceptions across five languages, showing indirect pushback often fails to fix factual errors.

The authors propose decoupled positional encodings enabling full-duplex speech models to run internal thinking streams at hardware-adaptive speeds.

Studies reflective prompt optimization in research agents, demonstrating that gains on held-out tasks can stem from prompt-level memorization.

Proposes a training method to adaptively skip turn-by-turn agent reasoning when previous thoughts remain sufficient for future actions.

The paper proposes a code summarization method that encodes abstract syntax tree paths using self-supervised learning to reduce training overhead.

Evaluates AI text detectors under attacks applied to both human and machine text, revealing high false-positive rates on edited human writing.

Presents a randomized experiment framework to test and curate self-written agent playbooks, preventing task failure caused by unverified lessons.

The paper introduces ViMoD, maintaining a compact visual context while dynamically routing back to fine-grained tokens during multi-step reasoning.

Introduces a post-training framework that trains text-to-SQL agents directly inside interactive database execution environments.

The paper proposes compression-layer modifications to reduce accuracy loss when using dynamic LZW vocabularies during language model inference.

The authors combine zero-shot test suite generation with agentic refinement to improve test coverage and mutation scores.

Shows linear prompt-injection monitors overfit to conversation features and reduces their false alarms by adding matched harmless negative examples.

The paper presents a simulated benchmark across eleven unconstrained environments to evaluate vision-language models on documenting real-world human physical work.

This paper introduces Choice2Seq, a benchmark measuring how often language models retain correct multiple-choice answers when generating them step by step.

Evaluates whether language models can act as world models to predict empirical experimental outcomes across AI research workflows.

Presents a symmetry-aware value cache merging method to compress key-value memory during long-context decoding without architectural changes.

This case study evaluates a non-programmer using an AI development tool to build a database application in a non-English language.

The paper repairs multimodal chain-of-thought final answers using original image features without altering the generated reasoning trace.

Demonstrates an attack method that alters embedded multi-bit text watermarks without knowing the secret generation key.

The paper examines interpretive lock-in where planning agents generate code passing self-generated tests while missing the actual specification requirements.

Introduces an instruction-following benchmark and training recipe to improve goal adherence in full-duplex conversational models.

Designs an inference serving system optimized for dynamic, token-level routing across multiple language models.

The paper evaluates pre-trained code models under a standardized experimental setup to compare the impacts of their pre-training tasks fairly.

This paper introduces a system to ensure tool-using language model agents comply with natural language company procedural manuals.

The paper presents Diffuplex, a full-duplex spoken language model that separates speech timing from content generation using discrete diffusion.

The paper evaluates long-term memory gating models to test whether they distinguish established factual updates from mere suggestions.

Presents a rolling masked diffusion method that accelerates full-duplex spoken dialog models by forecasting multiple future interaction frames.

Proposes a co-evolution framework that coordinates agent harness prompt modifications with model weight updates using self-generated experience.

The paper examines how prompting conditions and automated agent reviews affect code complexity placement across two software architectures.

The paper presents a theoretical framework using nesting properties to calculate distance bounds and compare language model truncation sampling algorithms efficiently.

This paper analyzes label collapse failure modes across diffusion and autoregressive language models during in-context learning classification.

This paper develops an uncertainty estimation framework for omnimodal models using external embeddings to score reasoning grounding and coherence.

The paper introduces an executable contract and monitor to verify that delegated authority revocations actually take effect across distributed agent gateways.

The paper describes training a competitive decoder-only search reranker from scratch using modest computational resources and common datasets.

This paper uses activation steering and probing to identify distinct internal representations for rule-based versus consequence-based moral reasoning in language models.