All papers

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

Presents a test-time method that extracts structured schemas from corpora to guide generative models toward more diverse outputs.

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

Examines how extended reasoning steps lead language models to interpret everyday logical terms strictly literally instead of inferring pragmatic meaning.

Analyzes how prompt variations redistribute internal feature activations in sparse autoencoders and traces their impact on downstream model decisions.

Introduces a metric to measure how non-linearly features learned by sparse autoencoders interact during downstream model behaviors.

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

The paper proposes a test-time aggregation method using bias control and injection to resolve disagreements among reasoning language model agents.

The paper presents a Korean dataset extracted from corporate sustainability reports to support information extraction on environmental, social, and governance topics.

The paper presents a self-guided reasoning framework where language models generate code to detect and diagnose errors in vehicle routing problem instances.

The paper analyzes multi-reward reinforcement learning failures where auxiliary objectives degrade primary performance and introduces a trust-anchored gating method.

Creates a structured schema that converts research papers into connected scientific entities to improve retrieval-augmented generation for autonomous agents.

Provides theoretical convergence guarantees and finite-sample bounds for advantage-based value sharing in reinforcement learning algorithms.

Introduces a benchmark and evaluation to test whether language models can successfully generate coherent nonlinear narratives.

Decomposes the alignment between language model activations and human brain scans using neurocognitively grounded semantic property representations.

Introduces a framework using language models to construct context-sensitive semantic representations that align with human brain activity during language comprehension.

The paper investigates the mathematical conditions under which the Muon optimizer's polar update acts as an optimal step.

The authors formulate classification and uncertainty quantification methods for pairwise difference learning frameworks.

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

The authors analyze tool graph memory in multi-turn agents and find that stored transitions provide little benefit over baselines.

The paper proposes a search framework to recover executable editing programs between before-and-after sewing pattern pairs.

The authors introduce a benchmark to audit mathematical claims and repair errors using an automated witness verification checker.

The paper evaluates feature encoders and pair selection strategies for pairwise difference learning in low-data classification regimes.

The paper formalizes the theoretical limits of automated gates evaluating self-improvement updates in autonomous agent code.

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

The authors introduce a Fisher-informed adapter merging method to maintain retained task performance during federated language model unlearning.

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

Proposes a structured knowledge admission and verification system to prevent continual learning agents from storing harmful past experience entries.

Models repeated games with endogenous delegation to analyze trade-offs between economies of scale and institutional accountability.

Investigates how partial interventions in concept-based models can unexpectedly reduce downstream task performance due to information leakage.

Develops a domain harness and policy learning pipeline to improve language model agent execution in computer-aided design environments.

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

Learns instruction-conditioned multimodal representations to dynamically cluster objects based on user-specified partition criteria without fixed cluster counts.

Develops pessimistic value-iteration algorithms with theoretical suboptimality bounds for offline risk-sensitive reinforcement learning using optimized certainty equivalents.

Develops a solvable topic model theory explaining when sequence models adopt in-context learning rather than specialised memorisation during training.

Analyzes strategic identifiability in multi-agent games, showing when inverse reinforcement learning fails to recover true underlying agent rewards.

Presents a theoretical orthogonal polynomial framework explaining why shallow neural networks learn low-frequency sinusoidal components before high frequencies.

Analyzes how continual learning methods affect demographic fairness over time across eleven deepfake generation techniques.

The paper pairs language models with symbolic verifiers to test and update reusable hypotheses for environment diagnostics and repair.

The paper aggregates irregular electronic health records into temporal graphs to predict risks of gastrointestinal cancers.

The paper outlines a long-context midtraining stage using software engineering data to prepare base language models for agentic tasks.

The paper optimizes batched task and motion planning to generate high-quality autonomous demonstrations for robotic manipulation training.

The paper separates task generation from model update inheritance to avoid compounding errors in self-evolving language model loops.

The paper tests how language models weigh price momentum, commentary tone, and earnings surprises when forecasting financial outcomes.

The paper aligns functional brain scans with multimodal representations to decode and retrieve perceived natural speech segments.

The paper introduces a commercially permissive multilingual instruction dataset covering five European languages to train open language models.

The paper extends integrated gradients to trace feature attributions along causal paths for high-dimensional models.

The paper demonstrates that a single adversarial agent can cause benign models in multi-agent discussions to quietly adopt false beliefs.

The paper demonstrates specification gaming in a simulated cleaning task and tests a constrained reinforcement learning mitigation.

The paper introduces a hierarchical reinforcement learning method designed for agentic tasks that span multiple interaction sessions.

The paper introduces a reinforcement learning pipeline enabling humanoid robots to manipulate bulky objects using their arms and torso.

The paper evaluates how choices in rating scale design systematically alter text evaluation scores produced by large language models.

The paper introduces a multi-agent system to automate data curation and dataset editing for supervised fine-tuning.

The paper introduces a hierarchical task network framework that separates multi-agent planning from execution scheduling.

The paper presents a benchmark that isolates agent memory evolution from downstream retrieval and generation failures.

The paper presents an end-to-end trained multimodal agent that combines visual graphical interface actions and executable code execution.

The paper introduces a framework to detect spurious shortcuts in safety benchmarks used to evaluate tool-using AI agents.

The paper evaluates 19 safety guards across multiple non-English languages to measure how translation alters step-level safety decisions for agents.

The paper trains language model agents to reorganize unstructured document collections into structured workspaces to speed up future multi-document queries.

The paper presents a method for robots to generate targeted preference queries to learn underspecified rewards from humans.