Model releases, architecture work and prompting research on large language models — from frontier-lab announcements to the arXiv papers behind them.
The paper evaluates five vision‑language models on autonomous driving tasks under various visual input conditions, finding that visual corruption affects accuracy and confidence differently across models and datasets. It then tests Visual Evidence Augmentation (VEA) as an inference‑time technique to enhance reliability, observing mixed improvements depending on the model and setting.
The paper proposes a new framework for collaboration between a small language model (SLM) and a large language model (LLM) that treats the interaction as an information acquisition problem under an API budget constraint. Instead of delegating reasoning tasks, the SLM remains the primary reasoner and selectively queries the LLM advisor with targeted questions, using a three-stage RLVR approach to decide when to call the advisor, how to phrase queries, and how to integrate the responses. Experiments on mathematical reasoning and coding tasks show that this strategy improves the performance–cost tradeoff compared to existing baselines and can transfer to other advisor model families without additional training.
LineupRL introduces a reinforcement learning framework with verifiable rewards for time series captioning, using a frozen large language model to identify the correct time series from a set of distractors based on a generated caption. This approach bypasses the limitations of supervised fine‑tuning and traditional RL rewards that poorly transfer to open‑ended time series generation. Experiments on two captioning benchmarks, as well as forecasting and reconstruction tasks, show that LineupRL outperforms both SFT and RL baselines across all metrics, and its trained 3B vision‑language model surpasses a 72B model distilled from SFT captions. The method also demonstrates resistance to reward hacking and produces captions that accurately trace trends and name key values.
The paper introduces a continuous evaluation framework that assesses both outcome-level and process-level aspects of evolving enterprise AI agent skills. It applies this framework to two variants of a Business Value Determination skill, running 240 trials across multiple models, harnesses, and specifications. The results show that while most trials pass final numerical checks, a large majority still exhibit process-level deviations, and dependency attribution reduces the number of failed checks per run. The framework also provides reusable regression tests and highlights specification sensitivity across configurations.
AVSD-Scenes is a new dataset of 12,291 audio‑visual scene descriptions for urban environments, built from the TAU Urban Audio‑Visual Scenes dataset. The descriptions are generated by first creating modality‑specific text with Qwen2‑Audio‑7B and Qwen2.5‑VL‑7B, then merging them with large language models (Qwen3‑14B, Mistral‑Small‑3.2‑24B‑Instruct‑2506, Gemma‑3‑27B‑it) to produce multimodal narratives that combine auditory and visual cues. Benchmarks show that these multimodal descriptions improve semantic alignment, cross‑modal retrieval, and scene classification accuracy (up to 95.4%) while remaining discriminative even without explicit scene labels.
The paper surveys recent advances in Retrieval Augmented Generation (RAG), a technique that integrates external retrieval into language model generation to reduce hallucinations and keep knowledge current. It introduces a four‑axis taxonomy—efficiency, defense, interactivity, and reasoning—to organize contemporary RAG research, covering retrieval methods, fusion strategies, embedding optimizations, and reinforcement learning policies. The survey also reviews evaluation practices, domain‑specific applications, and architectural variants, while highlighting ongoing challenges such as retrieval quality, reliability, domain adaptation, scalability, and explainability.
The study evaluates counterfactual bias in ten open‑source large language models (LLMs) for pediatric Emergency Severity Index (ESI) prediction. By creating paired clinical vignettes that differ only in demographic or socioeconomic variables, the authors measure shifts in acuity assignment, finding that counterfactual sensitivity varies widely across model families and sizes. A fine‑tuned Qwen2.5‑7B model exhibited the lowest sensitivity, while larger or medical‑domain models sometimes showed greater shifts, highlighting the need for fairness assessment before clinical deployment.
The paper introduces Causal Memory Policy (CMP), a framework that identifies the utility of memories in memory‑augmented language models by intervening on retrieval rather than on storage. CMP reserves fixed context slots for memories sampled with known propensities and estimates utility using self‑normalized inverse propensity weighting, providing unbiased estimates and exact variance. Experiments show that CMP improves discrimination between required and non‑required memories and reveals that identified utility alone is insufficient for retention decisions across unseen queries.
The paper introduces Bayesian Dialectical Argumentation (BDA), a method for aggregating answers from multiple large language models (LLMs) in a council setting. BDA treats each LLM’s typed moves—proposals, challenges, and concessions—as evidence in a classical annotator model, estimating per-agent reliability even when some agents are persistently unreliable. By weighting evidence according to these inferred reliabilities, BDA produces calibrated posterior probabilities for candidate answers and can invert unreliable agents instead of merely outvoting them, achieving superior calibration and robustness on both binary and multi-class benchmarks without extra LLM calls.
SPHERE is an adaptive VR indoor scene generation framework that turns isolated 3D synthesis into continuous human‑AI co‑creation. It learns persistent spatial preferences from multimodal user interactions, abstracts these into hierarchical constraints for geometric resilience, and employs a human‑in‑the‑loop reinforcement learning loop to refine retrieval policies. A mixed‑design study with 42 participants and offline ablation show that SPHERE reduces corrective edits and physical effort while avoiding bias toward shallow object‑level traits, producing geometrically resilient, profile‑aligned layouts.
Mimir is a physics‑grounded large language model agent designed for long‑horizon irrigation control. It operates on two timescales: a fast scale that uses a structured physical interface and deterministic simulator to validate and refine LLM proposals before execution, and a slow scale that consolidates recurrent failure patterns into persistent contextual principles. Across multiple sites, crops, and years, Mimir achieves the lowest aggregate control cost and reduces irrigation usage by about 51% compared to historical schedules, while ablation studies confirm the importance of forward simulation, verified revision, and persistent context.
The paper proposes a hidden‑state probing method for detecting hallucinations at the span level in large language model outputs, moving beyond token‑wise binary classification. By examining layer‑wise activation patterns, the approach identifies the exact onset and continuation tokens of hallucinations, achieving higher precision‑recall AUC than random baselines despite class imbalance. Additionally, the authors introduce a cross‑model detection framework where one model observes another’s internal representations, showing that an external observer can match or surpass the generator’s own self‑detection of hallucination onsets, even when the observer is smaller.
The paper introduces VISTA, a visual harness that equips a general-purpose multimodal model with long‑horizon vision and a lossless visual memory. VISTA enables the model to directly perceive and actively retrieve past observations, allowing it to reorganize visual input during reasoning. On the ARC‑AGI‑3 benchmark, VISTA boosts Claude Opus 5.0’s Relative Human Action Efficiency from 40.68 to a perfect 100.00, completing all 25 public games with 57.4% fewer actions than first‑time human participants, and it also outperforms baselines on three additional visual game and puzzle benchmarks.
The paper demonstrates that reading a large language model (LLM) judge’s verdict from the logits of its first generated token—an approach used in constrained decoding and likelihood‑scoring evaluation—introduces a significant distortion in position bias. Because judges do not always start with a verdict token (12–49% of cases for Qwen3 judges and <3% for Llama‑3.1‑8B and Phi‑3.5‑mini), this readout often returns the first response rather than a true judgment, inflating position bias by up to 42 points while barely affecting judge accuracy. The authors recommend reporting the frequency with which a judge leads with a verdict token to provide a more accurate assessment of position bias.
BudgetSchemaBench is a diagnostic tool for evaluating how different schema‑context budgets affect text‑to‑SQL systems. It automatically derives relevance labels from gold SQL, tests four budgets across 80 databases, and compares three schema representations while keeping table rankings fixed. The study shows that increasing the budget improves execution accuracy, especially for lexical retrieval, and that dense retrieval already captures most needed tables at low budgets.
The article surveys safety concerns for self‑evolving agents that continually update their internal state, such as model parameters and memories, from new interactions. It introduces the SAVER framework, which tracks reusable influence, adaptation, violations, exposure, and response to assess whether safety properties persist as agents evolve. The survey finds that legitimate state can become unsafe when its persistence, authority, or scope expands beyond its original conditions, and highlights gaps in current research on descendant repair and longitudinal evaluation.
The study investigates how large language model (LLM) advisors affect complex problem‑solving in a simulated clothing‑factory setting. Two preregistered experiments (N=200 and N=198) found that participants with AI support reported higher confidence and understanding, expended less effort, and in some cases achieved better performance or avoided bankruptcy. Within the AI‑supported group, more frequent changes to the AI’s recommendations were linked to improved unaided performance and knowledge.
The study reports that white men lacking an undergraduate degree receive the lowest average ratings from large language models across credit, hiring, and rental application scenarios. Using 18 models and 32 demographic profiles, the researchers found that these individuals scored 75.87 in credit, 92.71 in hiring, and 86.62 in rental housing, while black women with degrees scored highest. The findings show that white men without degrees consistently rank lowest or second‑lowest in most model‑scenario combinations, underscoring a distinct disadvantage not captured by broader racial or gender categories.
The paper introduces GRDisaster, a multi-task geospatial reasoning framework that leverages vision‑language models to interpret, geolocalize, and assess damage in crowdsourced disaster imagery. It builds on a new benchmark dataset of 26,340 images from PhotoMappers, linking volunteer geographic information, street‑view imagery, and remote sensing data across multiple disaster events from 2018 to 2024. GRDisaster combines deterministic and probabilistic cross‑view geolocalization with multi‑view fusion, and introduces spatial reasoning indicators to validate cross‑view matches and quantify disaster severity using expert‑verified annotations.
Proof‑Gated Signing (PGS) is a method for securing on‑chain AI agents that control wallets by simulating proposed transactions, extracting their effects, and using an SMT solver to verify a declarative value‑and‑permission policy across a range of oracle‑uncertainty prices. PGS then compiles on‑chain post‑conditions—such as wallet balance bounds, payee receipts, allowance caps, and ownership checks—ensuring that any execution satisfying these conditions also satisfies the policy, even when the chain state drifts due to front‑running, contract upgrades, or token‑parameter changes. In a testbed of 260 scenarios, PGS prevented 93.6 % of harmful cases and passed 97.5 % of benign ones, outperforming simulation‑only checking and static allowlists, while incurring about 41 k gas and 0.1–0.2 s per check.