All papers

The paper presents a replay-driven decoupled reinforcement learning framework to improve system efficiency during language model post-training.

The paper introduces a hierarchical benchmark for evaluating visual generative models on multi-step world state prediction tasks.

The paper analyzes why search-augmented language models produce outdated answers even when relevant evidence is retrieved.

The paper trains language model auditors to detect multi-turn policy violations during tool use by autonomous agents.

Introduces ProtoArena, a black-box dynamic benchmark evaluating the ability of AI agents to discover and exploit cross-component network protocol vulnerabilities.

Proposes a visual verifier that decomposes complex prompts into sub-questions and assigns continuous scores to improve multimodal evaluation.

The authors present a benchmark to evaluate conversational agents that follow correct factuality but violate procedural task steps.

The paper provides a formal algebraic framework and normal forms to structure and audit enterprise AI agent workflows.

The paper introduces decision models that balance fast single-pass predictions with adaptive reasoning for uncertain cases using reinforcement learning.

The authors introduce a framework to track safety model judgments through output parsing to prevent malformed responses from causing bad decisions.

The paper uses human interventions to localize and steer computations behind agent failure to decide between acting and clarifying.

Presents a benchmark based on real criminal cases showing how correction framings and cultural adaptations bypass safety guardrails.

This paper presents a distillation method using frozen language models to guide molecular flow generators toward natural language discovery goals.

Evaluates enterprise policy enforcement across structured JSON versus single-token outputs, measuring the impact of explicit reasoning on decision stability.

This paper introduces an agentic post-training framework that coordinates multiple agents to evaluate and adaptively train text-to-image models.

Analyzes how misconceptions in early planning stages cause cascading failures during automated code repair by language model agents.

The authors present a benchmark to evaluate conversational agents that follow correct factuality but violate procedural task steps.

The paper proposes adding an explicit copy-paste prediction head to language models to improve in-context retrieval and copying efficiency.

The paper argues that software agency should be treated as a core organizing abstraction in computer science rather than merely an AI application.

Proposes a hybrid image tokenizer combining compact 1D global tokens with adaptive sparse 2D tokens for local detail refinement.

Introduces a dataset benchmarking human-agent interaction and collaboration across long-horizon professional workflows.

Examines whether source code quality metrics predict the token costs of multi-agent pull request review pipelines.

The paper personalizes mixture-of-experts models and analyzes which specialized experts encode distinct user-specific features.

The paper presents a four-gate verification protocol to evaluate whether audio-conditioned visual prediction models genuinely rely on input audio signals.

Analyzes how multi-agent workflow architecture affects the propagation and attenuation of coordinated stance manipulation across sequential agent chains.

Introduces a benchmark evaluating whether language agents can make correct scientific judgment calls derived from human interventions.

Analyzes how misconceptions in early planning stages cause cascading failures during automated code repair by language model agents.

Introduces LatentRiskBench to evaluate safety risks in enterprise agents where action safety depends on latent workflow context.

The paper proposes data-space iteration, a few-step generative method removing flow discretization by directly refining noise predictions in data space.

Proposes a multimodal drag-based image editing framework that generates visual reasoning traces to guide point-based transformations.

Introduces a dataset benchmarking human-agent interaction and collaboration across long-horizon professional workflows.

Introduces a benchmark evaluating single-forward-pass discrete decision models versus generative reasoning models across diverse closed-loop tasks.

CameraVLM couples motion states and on-screen flow dynamics to improve camera-motion understanding in vision-language models.

The paper examines how providing additional context, such as codebases and benchmark transcripts, affects automated LLM safety auditors.

Introduces an information-theoretic framework to reweight supervised fine-tuning data based on conditional information gain.

Evaluates test-time search in computer-use agents, finding critics perform better when invoked only at predicted final states.

Investigates whether continuous latent reasoning tokens genuinely compute answers or merely settle early on predetermined conclusions.

The paper introduces an architecture combining private prediction paths and graph collaboration to maintain multimodal inference accuracy when modalities are missing.

The paper examines how shared KV caches cause language model agents to leak goals across competing clients, offering internal intervention techniques.

Investigates on-policy distillation for multi-turn agents by decoupling objective interaction states from internal student reasoning trajectories.

Introduces a benchmark evaluating whether language agents can make correct scientific judgment calls derived from human interventions.

Proposes an ensemble framework that weights judge confidence and masks disagreements when training models with rubrics as rewards.

Shows agent failure detection models often rely on trivial cues like run length, exposing benchmark flaws when attributing critical failure steps.

Evaluates static Transformer block skipping in time-series foundation models to reduce inference latency with minimal forecasting degradation.

Presents a teacher-student framework to evaluate whether remote sensing visual queries are genuinely supported before predicting bounding boxes.

The paper investigates performance vulnerabilities in self-improving agents when their co-evolved skill banks undergo post-deployment drift.

Proposes a trajectory-aware retrieval method that indexes tool sequences in a trie to cut token costs and improve selection accuracy.

Tests a hosted decision model to show that adding irrelevant options systematically shifts output probability distributions between top candidates.

Presents deployment experiences and design strategies for ByteExpress, an RDMA communication layer for distributed training and inference workloads.

Benchmarks multimodal decision models across multiple prompts for classifying breast ultrasound imaging datasets.

Tests whether transcripts of correctly solved tasks from saturated benchmarks can predict model performance on unseen harder evaluations.

Introduces a benchmark evaluating single-forward-pass discrete decision models versus generative reasoning models across diverse closed-loop tasks.

The paper introduces a generative pipeline to reconstruct multi-room 3D environments from sparse panorama images using geometric optimization.

Introduces a diagnostic evaluation framework to systematically test generalization and perturbation robustness in vision-language-action robotic models.

Reports that initialization-time gradient masks underperform random masks on catastrophic forgetting when confined to the same LoRA parameter pool.

Introduces a benchmark evaluating language model agents in real-time environments and a two-timescale training method to improve execution speed.

Studies whether conversational agents correctly enforce authorization rules when users attempt to cancel prior agent commitments to others.

The paper reveals a numerical side channel in floating-point bias addition that enables recovering full logit vectors from API queries.

This paper presents a method for upcycling pretrained language models into sparse architectures using token-indexed embeddings in residual pathways.

Presents a dual-stream memory architecture combining recurrent compression and frame retrieval for vision-language-action robotic policies.