All papers

Introduces a compression-based patching strategy for byte-level language models that allocates compute where upcoming bytes are unlikely to repeat.

The paper identifies an internal activation direction in spoken language models to control and transfer conversational turn-taking styles.

The paper shows that the Bradley-Terry model fails on Chatbot Arena for similar models, and introduces a parameter-efficient correction for ties.

The paper profiles the security attack surface and execution privileges of third-party agent skills across public marketplaces.

The authors evaluate language models on finance peer review tasks, finding predictive ability for publication outcomes through correctness rather than originality assessments.

The paper presents PocketTSLM, deploying an on-device soft-prompt time-series language model on smartphones for personal health data.

The paper introduces a confidence-calibrated controller to optimize inference compute and test-time expansion decisions in reasoning models.

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

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

The authors demonstrate that encoding malicious instructions in agent skills bypasses scanners and causes models to execute harmful actions upon decoding.

Introduces a benchmark spanning ten tasks to evaluate multimodal model verification and rejection explanations on generated 3D content.

The paper introduces a hybrid generative-tagging architecture to improve efficiency and coverage in product attribute and value extraction.

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

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

Runoff eliminates option-order bias in multiple-choice model predictions by scoring items independently before running balanced candidate comparisons.

The paper investigates learning physical damping properties from impact audio and distills this knowledge into vision models.

The paper presents PocketTSLM, deploying an on-device soft-prompt time-series language model on smartphones for personal health data.

This paper proposes a semantic reconstruction game to accurately evaluate jailbreak success by filtering out harmless repetitions and refusals.

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

This paper studies how different framings of user claims induce sycophancy in large language models on factual questions.

The paper reduces vision-language model parameters by pruning layers and applying depth recurrence over remaining language backbone blocks.

The paper proposes RoPERelay to reorganize positional geometry and rotary embeddings after sparse KV cache token eviction.

The study evaluates coding agents generating images via executable code, analyzing their strengths in geometric constraints over photorealistic drawing.

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

This paper introduces a timestamp theory and persistent ledger to route updated information to pending LLM decisions before action.

Introduces a framework that enables vision-language models to iteratively adapt frozen visual encoder computation based on the input question.

Proposes a compressed representation using Gaussian trellises to quantize conditional memory lookup tables down to two bits.

Introduces a benchmark to evaluate how language model agents learn and self-evolve reusable skills when exposed to non-adversarial noise.

The paper presents an agent that combines language models with operational policies to generate verified operational reports for high-performance computing systems.

This paper introduces a benchmark to measure how language model agents handle ambiguous or missing goals, context, and constraints.

Trains vision-language models via reinforcement learning to actively navigate and inspect 3D environments using single views before answering questions.

Compares abstract-only versus full-text inputs and few-shot versus fine-tuning approaches for generating biomedical lay summaries.

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

The paper studies using an agentic verification architecture to enable small on-device models to generate factual clinical documentation.

This paper conducts an adversarial evaluation comparing a probability-based safety judge against large language model judges and safety classifiers across multiple attack types.

Proposes a decoupled inference-time guidance algorithm to filter unsafe outputs in text-to-image diffusion models without altering weights.

Shows across 200 models that proposed intrinsic tokenizer metrics do not reliably predict downstream language model performance.

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

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

This paper examines how single-agent long-term memory accumulates repeated historical claims and introduces a benchmark and framework to mitigate unverified persistence.

Identifies early-stage issues with classifier-free guidance in flow matching models and proposes a correction for controllable image editing.

Introduces a hybrid speculative decoding framework splitting draft tasks between mobile neural processing units and cloud servers.

The paper introduces a benchmark to measure language model agent performance under real-time latency and streaming temporal constraints.

The paper compiles teacher model reasoning traces into verified programs that guide small models through complex procedural tool-use tasks.

This paper provides four auditing tests to determine whether failure-attribution judges actually perform multi-step checks or merely rely on spurious surface cues.

Decomposes agentic spatial reasoning by using a metric 3D perception tool to resolve visual inconsistencies before spatial reasoning.

Analyzes how calibration language choices affect block-influence depth pruning rankings in multilingual language models.

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

This paper introduces plug-in decision heads that read action choices directly from visual agent hidden states rather than generating text tokens.

Analyzes answerability signals in frozen visual question answering models, finding standard confidence and alignment metrics fail to reliably detect unanswerable inputs.

Proposes a framework that increases image diffusion throughput by fusing and processing multiple generation samples simultaneously in manifold space.

The paper evaluates single-pass decision models against rational choice axioms and identifies ordering and rival-option sensitivity failures.

The paper proposes a multi-view hierarchical memory framework to improve information retrieval from enterprise group chats.

Finds that framing malicious coding prompts as engineering specifications systematically reduces language model refusal rates across multiple models.

The authors present an active spatial search framework enabling vision-language agents to ground 3D objects in city-scale environments via interactive tool use.

This paper explores internal-state sharing in multi-agent language models through a central pool rather than pairwise transfer bridges.

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

The paper introduces a benchmark to measure language model agent performance under real-time latency and streaming temporal constraints.

The paper presents a benchmark to test how well AI agents can reconstruct dynamic, editable 3D scenes from dynamic reference videos.

Presents ECMR, a dynamic evidence-aware controller that guides multi-agent deliberation steps to prevent false consensus and reduce redundant reasoning.