InFlowOp attacks the costly cycle of building and fixing multi-agent LLM workflows by pricing every decision in a single label-free cost. The framework first constructs a workflow at the right granularity via bidirectional cost-driven coalescing, then detects and repairs faults locally during execution without reference answers. The authors also introduce Braid, a benchmark designed to demand multi-agent coordination, and report accuracy gains up to +11.97% over a single agent across eight domains and six backbones.

ReHoPER is a zero-shot, inference-only reasoning method that improves LLM compositional reasoning by formulating it as receding-horizon planning. At each step, it generates a short horizon of candidate intermediate questions, answers one, updates its history, and replans across multiple stochastic paths before aggregating by majority vote. The paper reports consistent gains over self-consistency on MuSR, MoreHopQA, and a new controlled benchmark iLLC, with the largest improvements on the hardest compositional tasks. While the method is task-agnostic, its accuracy gains come at a significant inference-time cost.

The paper asks how language models retrieve relational facts from a context list. Its core claim, the ordinal addressing hypothesis, is that models navigate to a fact not by the entity name in the question, but by the fact's serial position — its order of mention. The authors formalize this as additive fact addresses $a_p$ in hidden states and test it via activation steering across fourteen models, reporting transfer across contexts, low-rank geometry, and a pretraining trajectory reminiscent of human serial-position effects.

ActiveSaddler addresses the overlooked problem of which training scenarios to use during automated LLM-agent harness optimization. It treats curriculum selection as a non-stationary bandit whose arms are dynamically instantiated failure patterns, allowing the scenario distribution to co-evolve with the harness rather than being fixed upfront. On GAIA2 and Terminal-Bench 2.0, this yields test Pass@1 gains of 4.4 and 7.5 percentage points over the same optimizer with a fixed scenario order, arguing that adaptive curricula are a critical new dimension for harness optimization.

Earnings call transcripts are long, jargon-dense documents that are hard to analyze at scale, and citation-grounded LLM outputs are costly to evaluate because they rely on expert annotators. This paper proposes numeric evidence evaluation (NEE), an automatic metric that checks alignment of salient financial figures across claims, citations, and source documents to bypass human judgment. The authors also introduce ECTs-100, an automated benchmark of long-context Q&A pairs built from S&P 500 transcripts, and test a failure mode they call conscious incompetence—whether LLMs abstain when evidence is missing. Their findings matter because they show that strong groundedness scores mask real correctness gaps and a troubling tendency to hallucinate (~48%) under informational insufficiency.

Robotic foundation models currently rely on multi-step flow matching to generate actions, which creates a severe inference bottleneck for real-time control. This paper proposes Kinematic MeanFlow (K-MF), a one-step action policy that reformulates MeanFlow by decoupling the time derivative across two sub-intervals to accommodate the unique late-stage surge and widening spread of local acceleration in robotic velocity fields. By introducing an intermediate anchor point and a learnable partition strategy, K-MF cuts action-head latency by up to 74% while matching or exceeding the success rates of multi-step baselines across manipulation and navigation tasks.

This paper addresses the static nature of Large Language Models that prevents dynamic adaptation to streaming contexts. The authors introduce In-Place Test-Time Training, which repurposes existing MLP down-projection matrices as “fast weights” that update during inference via a Next-Token Prediction (NTP)-aligned objective. Unlike prior TTT methods that require architectural changes, this approach enables “drop-in” enhancement of pretrained models without retraining from scratch.

GISTBench evaluates whether LLMs can accurately extract user interests from behavioral interaction histories in recommendation systems. Unlike traditional benchmarks that optimize for item prediction accuracy, it verifies if predicted interests are actually grounded in engagement signals using two novel metrics: Interest Groundedness ($IG$) and Interest Specificity ($IS$). The authors find that current LLMs struggle primarily with recall—discovering all verifiable interests—rather than hallucination, revealing critical bottlenecks in evidence counting across heterogeneous signal types.

Recursive Language Models (RLMs) tackle the long-context problem by treating prompts as external environment variables that an LLM can programmatically manipulate through a REPL. Instead of feeding long prompts directly into the neural network, RLMs use symbolic code execution to decompose, filter, and recursively invoke sub-models over prompt snippets. This allows processing inputs up to 10M+ tokens—two orders of magnitude beyond typical context windows—while maintaining strong performance on complex aggregation tasks.

This paper tackles the challenge of scaling reinforcement learning for long-horizon tool-using agents, where LLMs must orchestrate dozens of tool calls to satisfy multifaceted constraints. The authors propose STAR, a post-training pipeline that decomposes the RL design space across five axes—reward shaping, model scaling, data composition, algorithm selection, and environmental stability—to derive a practical, scale-aware recipe for training.

Diffusion Language Models (DLMs) train with a static single-step masked prediction objective but infer via multi-step progressive denoising, creating a train-inference mismatch that compounds errors. MemDLM bridges this gap through Bi-level Optimization: an inner loop updates fast weights (Parametric Memory) to capture local trajectory experience, while an outer loop conditions the base model on this memory. The approach yields faster convergence, lower exposure bias, and substantial gains on long-context needle-in-a-haystack tasks, with an optional inference-time adaptation that acts as an emergent in-weight retrieval mechanism.

Selective prediction systems in LLMs abstain from answering uncertain questions to mitigate hallucination harms in high-stakes domains. This paper identifies a critical failure mode of entropy-based uncertainty quantification: the 'confidently wrong' regime where models produce low-entropy hallucinations. The authors propose combining entropy signals with correctness probes using logistic regression, and advocate for deployment-facing metrics—E-AURC and TCE—over AUROC to ensure systems can reliably operate at strict safety thresholds.

Ara-BEST-RQ introduces dedicated self-supervised speech models for Arabic dialects. The authors curate 5,640 hours of Creative Commons Arabic speech covering 20 dialects and train Conformer-based BEST-RQ models up to 600M parameters. Their 300M model achieves state-of-the-art dialect identification performance using fewer parameters than competing Whisper-based systems. This work helps close the gap for underrepresented Arabic dialects in speech technology.

Cross-lingual dysarthria detection in Parkinson's disease is hampered by language-dependent structure in self-supervised speech representations that confounds pathology classification. This paper proposes a centroid-based 'language shift' (LS) that aligns source-language embeddings toward target-language distributions using only healthy control speech, enabling zero-shot transfer without model retraining. The approach addresses the critical data scarcity in clinical speech applications while aiming to disentangle linguistic variation from motor impairment markers.

This paper tackles the problem of speaker traits entangling with synthesis source information in speech deepfake source verification. The authors propose a Speaker-Disentangled Metric Learning (SDML) framework that combines Chebyshev polynomial approximations for gradient stability with Riemannian geometry (hyperbolic space) to separate speaker identity from source generator artifacts. Evaluated on four new cross-protocols using the MLAAD benchmark, the method aims to prevent models from relying on speaker shortcuts when verifying synthetic speech origins.

Parallel decoding promises faster text generation than autoregressive models but historically sacrifices quality due to simplified conditional independence assumptions. This paper introduces Gumbel Distillation, which leverages the Gumbel-Max trick to create a deterministic mapping from latent noise to teacher outputs, effectively providing the parallel student a blueprint for joint token distributions. By conditioning on Gumbel noise rather than relying on naive factorization, the method narrows the quality-efficiency gap, delivering substantial improvements across masked diffusion and multi-token prediction architectures.

The paper tackles the 'semantic parsing burden'—the effort required to translate natural language into structured RDF/OWL representations for knowledge graphs. It proposes the Semantic Ladder, a five-level framework ($L_1$ to $L_5$) enabling progressive formalization from raw text snippets to higher-order logic. By introducing Rosetta Statements as semantic anchors and emphasizing modular semantic units, the work aims to lower barriers to knowledge graph construction while maintaining semantic continuity.

As consumer-grade EEG headphones enter the market, a critical question emerges: can language models adapt to your specific neural signature? This paper demonstrates that frozen LLMs already contain person-specific linear directions in their activation spaces that predict individual brain activity during reading, achieving a ninefold improvement over population averages. The findings suggest that deep neural networks encode stable, individual cognitive fingerprints that could enable future brain-computer interfaces to personalize AI to the user wearing the headset.

This paper analyzes temporal dynamics in Swiss digital news across French, German, and Italian language regions using a triangulated methodology that combines quantitative NLP with qualitative interpretation. The authors process 1.7 million articles to study how different event types—Brexit, Swiss Wolf, Christmas, and the British Royal Family—are covered across linguistic boundaries, introducing domestication profiles and proximity salience ratios to quantify cultural proximity effects.

This paper tackles the challenge of automating BT-RADS (Brain Tumor Reporting and Data System) classification for post-treatment glioma MRI surveillance. BT-RADS requires integrating complex information: volumetric tumor changes, medication effects (steroids, bevacizumab), and radiation timing. The authors propose an end-to-end pipeline combining CNN-based tumor segmentation with a multi-agent LLM system to extract clinical variables from unstructured notes and apply algorithmic scoring logic. This matters because manual BT-RADS scoring is error-prone, with prior studies showing substantial inter-reader variability and inconsistent application of clinical context.