All papers

Studies the transferability of internal activation interventions to reduce unsafe tool execution across model precisions and languages.

The paper presents a method using an external encoder to induce latent soft-token reasoning on frozen language model backbones.

The paper evaluates whether language models genuinely benefit from compressed memory by comparing them against memory-free baseline models.

The paper proposes an interface framework that tracks and controls application state changes made by graphical user interface agents.

The paper detects backdoor triggers in multimodal language models by tracking output evolution trajectories across internal hidden layers.

The paper proposes isolating internal model judgments from instruction formatting variations using a label-free hidden state subspace.

This paper develops a self-supervised representation learning approach based on pointwise mutual information factorisation for scalable generative latent tokenization.

The paper introduces a test showing that steering vectors designed for competing ethical frameworks behave interchangeably in open-weight models.

The paper introduces a distillation method that trains agent policies on explicit runtime control semantics recovered from replayed execution traces.

Introduces a decomposed rubric synthesis method to create training data for open multimodal models that generate visual code artifacts.

Demonstrates that machine unlearning often fails to prevent targeted information recovery across alternative multilingual access paths.

This paper introduces a two-stage priming framework combining reconstruction pretraining with distributional post-training for one-step generative modeling.

The paper benchmarks whether AI agents can recognize execution blockers and appropriately defer action on unsolvable tasks.

This paper presents a training-free contrastive decoding framework using attention-guided perturbations to reduce hallucinations in video language models.

Shows that periodically clearing agent search history and restarting from the top candidate prevents automated research agents from plateauing.

Proposes an availability framework to identify and resolve within-event timing mismatches in earnings-call financial NLP benchmarks.

Presents a serving system that dynamically manages GPU memory and prefetches weights for multi-model agent workflows.

This paper proposes SCOUT, a cascaded two-track framework combining visual candidate filtering, offline semantic summaries, and query refinement for document retrieval.

Introduces a prompt optimization framework that monitors item-level regressions and recoveries rather than relying solely on aggregate accuracy.

Studies multi-developer agentic systems where independently built specialist agents collaborate to outperform larger systems created by single developers.

Identifies a specific vision transformer attention head that spuriously attends to medical image corner text instead of lung pathology.

The authors propose Subspace Tuning, a fine-tuning method restricting weight updates to bottom singular subspaces to mitigate catastrophic forgetting in language models.

This paper introduces Anuvani LID, a dedicated 115M parameter model for spoken language identification across 22 Indian languages.

Argues that inferential accuracy fails to evaluate agent coordination safely because correlated errors can compound into costly team actions.

The paper analyzes model swapping in multi-session agents and proposes a history summarization method to preserve strong reasoning behaviors.

Introduces a data augmentation method that breaks spurious contextual correlations causing emergent misalignment during domain finetuning.

The paper presents a closed-form, training-free approach to merge specialized fine-tuned models into a single student using forward passes.

This paper introduces DarKnight, a framework that uses locally hosted vision-language models to solve dark web CAPTCHAs during threat intelligence gathering.

Presents a benchmark evaluating security agents through live, symmetric attack-and-defense competitions in isolated sandbox environments.

The paper presents a red-teaming method that injects stealthy attack operations across interdependent agent skills using code rewriting.

Proposes a lightweight protocol for AI agents to report runtime status and liveness to central servers and monitoring dashboards.

The paper compares the Muon optimizer against AdamW across controlled data, wall-clock time, and tuning budgets.

Analyzes why feature-based test-time adaptation fails to recover accuracy on corrupted inputs in generative multimodal language models.

Analyzes discrepancies between static tool specification errors and actual runtime failure rates in language model agent execution.

The paper introduces an evaluation framework to diagnose step-level execution stability and error recovery in graphical user interface agents.

Introduces an episode-level healthcare benchmark to audit how reliably language model agents escalate actions when facing workflow disruptions.

The paper introduces a system that retrieves document information into a fixed set of prompt embeddings for frozen language models.

This paper develops a self-supervised representation learning approach based on pointwise mutual information factorisation for scalable generative latent tokenization.

Documents safety alignment methods, negative preference tuning outcomes, and automated judge limitations for a Portuguese language model.

This paper introduces an architecture that stores and reuses specialized language agents as temporally extended options for complex coding tasks.

The paper demonstrates how correctness feedback undermines formal specification auditing and proposes a phase-separated alternative for Verus.

The paper presents an agent orchestration framework that trains smaller language models to navigate, inspect, and reason over long multimodal documents.

The paper proposes a reinforcement learning framework using quality-aware rewards to guide multimodal knowledge graph construction for retrieval-augmented generation.

The paper introduces a benchmark to evaluate whether frontier language model agents can execute open scientific research and produce follow-up papers.

The paper proposes a self-supervised method that uses repetitive tool calls as negative signals to reduce looping behavior in long-horizon agents.

Tests how well watermarks survive automated model self-evolution and introduces a mark designed to persist through rewriting attacks.

Presents a framework that decouples speech generation and synthesis token spaces to speed up spoken dialogue model inference.

The paper presents an agent orchestration framework that trains smaller language models to navigate, inspect, and reason over long multimodal documents.

The paper presents a multilingual benchmark to evaluate how language models resolve linter alerts while preserving program functionality.

The paper introduces a self-training method that uses runtime policy verification feedback as rewards to teach agents safety compliance.

Investigates cost-aware policies for adaptively switching between language models of varying fidelity in stochastic optimization pipelines.

Presents a framework that decouples speech generation and synthesis token spaces to speed up spoken dialogue model inference.

The paper studies the durability of inference-time value steering in multi-turn conversations when user and model values conflict.

The paper presents a decision model to evaluate language model efficiency based on energy and cost per completed agentic task.

The paper introduces a benchmark using fuzzing and transaction replays to evaluate the behavioral correctness of generated smart contracts.

The paper proposes a self-play framework to train multimodal foundation models using text proxies instead of paired cross-modal datasets.

The paper trains a small language model to generate and refine GPU kernel optimization scripts using compiler traces.

The paper introduces a benchmark to evaluate whether language models can accurately detect appropriate and inappropriate uses of persona information in dialogue.

The paper presents a taxonomy of failure modes in multi-agent systems where agents interact across organizational boundaries with private states.

The paper introduces a falsification-guided agent protocol requiring systematic hypothesis elimination to improve decision-making consistency on complex diagnostic tasks.