All papers

The paper proposes a method combining offline execution-verified prompt updates and test-time refinement to improve program generation in language models.

The paper investigates base language model benchmarking setups to identify evaluation configurations that produce consistent, non-random, and monotonic performance signals.

The paper presents a double-annotated sentiment dataset for Nepali and evaluates several open-source and commercial language models on it.

The paper introduces a training-free sampling framework that uses internal model activations to efficiently estimate rare failure probabilities in language models.

The paper proposes a parameter-efficient method leveraging frozen model representations for multi-output text regression tasks across diverse benchmarks.

The paper presents a framework to formalize compute, labor, and workflow costs as core measurement design criteria for automated essay scoring.

The paper evaluates the ability of language model agents to diagnose appliance faults using a simulated appliance benchmark based on sensor data.

The paper presents an approach using character-level tokens, larger models, and synthetic data for the multilingual BabyLM data-efficiency challenge.

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

The paper formulates task routing across language model providers as an auction mechanism to encourage truthful cost reporting while minimizing context switching.

The paper introduces a multi-stage topic modeling pipeline that merges fine-grained document clusters using a language model to balance efficiency and quality.

The paper proposes an experimental design that separates example selection from mass redistribution to analyze the effects of training-data quality filtering.

The paper shows that differences in predicted target tokens, such as whitespace, can falsely look like layer specialization across different corpora.

The paper analyzes prefix replay curricula and handoff rules under reinforcement learning to improve long-horizon task completion for language model agents.

The paper presents a vision-language policy combining pre-aligned representations and multi-stage learning to improve robotic person tracking among distractors.

The paper introduces a framework that decomposes user feedback into atomic rules to retain only reusable preferences for personalizing language models.

The paper proposes a multiple-summary metric and dataset to evaluate graded proposition salience across diverse text genres.

The paper introduces a reinforcement learning framework using intermediate-state branching and segment-level credit assignment for diffusion language models.

The paper studies how visual input signals affect scaling behavior and non-visual reasoning during native multimodal pretraining.

The paper analyzes post-training in audio-language models to separate genuine audio reasoning gains from question-format memorization.

The paper presents adapted small language models designed to directly score agricultural candidate choices in a single forward pass.

The paper develops a streaming end-to-end inverse text normalization architecture using a pretrained text-to-text model with a read-tag-write policy.

The paper proposes a decoding algorithm that denoises consecutive blocks concurrently by promoting sampler states in diffusion language models.

The paper analyzes reward hacking in verifiable instruction following and introduces a likelihood bonus method to maintain answer quality.

The paper presents an item response theory framework to jointly evaluate student performance on standard questions and tutoring dialogue subtasks.

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

The paper introduces a multi-agent framework that reframes harmful requests using Moral Foundations Theory to test model jailbreak vulnerabilities.

The paper proposes a prefix-augmented distillation method providing one-step student branch feedback during language model pretraining.

The paper proposes a pair-norm guided mixed-precision quantization method for key caches during long-output transformer inference.

The paper reports a pre-registered test measuring the stability of history-based forecasts for model brand mentions over twenty days.

The paper identifies evaluation pitfalls in multimodal machine unlearning and introduces a benchmark grounded in real-world knowledge.

The paper introduces a localized Japanese benchmark for evaluating tool-using conversational agents in customer-support scenarios.

The paper proposes an interface for typed decision models that selectively invokes and interrupts reasoning to balance accuracy and latency.

The paper introduces a soft margin preservation target for supervised fine-tuning to retain non-target token rankings.

The paper proposes a training-free interface for heterogeneous language models to communicate facts using shared symbolic word indices.

The paper introduces a 1,000-problem benchmark for constraint-based planning and decision-making in the Bangla language.

The paper introduces a unified routing framework that learns both execution mode selection and model assignment across diverse tasks.

The paper analyzes retrieval failures in Mamba architectures under query-first settings and explores token representations and selection gates.

The paper presents a formal spatiotemporal framework to specify and verify road vehicle safety rules across continuous trajectories.

The paper presents a four-stage training pipeline combining preference optimization and distillation for instruction-following machine translation.

The paper reveals vulnerabilities in prompt-predictability metrics used to estimate human contribution in AI-assisted writing.

The paper benchmarks ten self-distillation methods in agentic reinforcement learning across varied model sizes, tasks, and computational costs.

The paper introduces a speculative decoding system combining self-refinement and pipelined drafting to reduce wall-clock generation time.

The paper investigates how team composition and organizational interdependence affect performance in self-evolving multi-agent systems.

The paper introduces a CPU front-end tokenizer that encodes only session deltas to speed up multi-turn 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 diagnostic evaluation framework to measure how inference-time techniques affect parametric knowledge extraction in models.

The paper introduces a two-part benchmark to evaluate language models on general and examination knowledge concerning Taiwan.

The paper presents an audio evaluation benchmark assessing accuracy drops between speech-input and text-input pipelines in Southeast Asian languages.

The paper analyzes question-level variations and prompt-execution mismatches when transferring reasoning prompts across different language models.

The paper proposes a diagnostic test hiding target identity to evaluate structure-property chemical reasoning in language models.

The paper introduces a method that weights rubric criteria through pairwise model comparisons to improve evaluation accuracy.

The paper presents an agentic reinforcement learning framework that navigates organized chemical spaces for cross-scaffold molecular discovery.

The paper introduces a dual-process memory framework coordinating small and large models for efficient long-context reasoning.

The paper shows pruned reasoning models fail by looping, proposing an activation steering method to restore proper termination.

The paper tests a heuristic to set the number of transformer attention heads using the intrinsic dimension of token representations.

The paper systematizes grading requirements for open-text legal exams and compares eight evaluation paradigms using human and model judges.

The paper introduces a lightweight gray-box hallucination detector combining token probabilities with vector representations of token identities.

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

The paper introduces a gated semantic alignment framework with multi-source rewards to improve multimodal emotional support conversations.