Selective state space models (SSMs) use a recurrence to mix token information, a process analogous to attention in transformers. By modeling token evolution as an ordinary differential equation and applying input‑to‑state stability, the study proves that SSMs exhibit local exponential stability of consensus equilibria and delineates their domain of attraction for time‑varying weight matrices. Experiments on a pretrained Mamba‑2 model reveal that the output gate controls the degree of consensus, preventing tokens from fully converging.
By Jo\~ao Pedro Silvestre, \'Alvaro Rodr\'iguez Abella, Paulo Tabuada
The paper evaluates three cost‑efficient large language models—Gemini Flash 3, GPT‑5.4 mini, and Claude Haiku 4.5—on 992 algorithmic programming tasks written as Java Spring Boot service methods. Across eight configurations (model, coding tool, and prompt variants) the authors generated 7,593 methods, classified them into eight outcome categories, and executed 7,936 requests. While structural compliance was high, only 12.9% of returned answers were correct, and 38.4% of methods failed to compute the value they returned, with the most reliable methods being the least accurate.
By Chandimal Adikari, Nandika Herath
MoRE: Mixture of Reused Experts is a hybrid architecture that combines Mixture-of-Experts (MoE) with weight‑sharing techniques. It shares expert pools across adjacent layers while each layer keeps its own router, and introduces lightweight depth embeddings to help shared experts differentiate layer contexts. Experiments on models ranging from 114 M to 1.15 B parameters show MoRE achieves lower perplexity and better downstream performance than standard MoEs and other weight‑sharing models, with only minimal changes to existing MoE implementations.
By Eric S. Qiu, Utku Umur Acikalin, Justin Lovelace, Christian Belardi, Arjun B. Mulchandani, Carla P. Gomes, Kilian Q. Weinberger
The paper compares Knowledge-Graph Based Augmentation (Graph-RAG) with Retrieval-Augmented Generation (RAG) for answering culturally specific questions. Using the LatamQA dataset, Graph-RAG, built automatically from Wikipedia via KGGen, matches RAG performance and reduces the base LLM’s error by 72% with a standard KG and 78% with a benchmark-aware variant. The approach also transfers zero‑shot to Portuguese, showing multilingual applicability.
By Pablo Poulenard, Yannis Karmim, Valentin Barri\`ere
The paper introduces DualViewEval, a benchmark compression technique for agent evaluation that jointly uses outcome and process signals to learn an exact-size miniset and predict full-benchmark scores. By analyzing large-scale trajectories, the authors identify six complementary process signals linked to final agent performance. Across five agent benchmarks and five baselines, DualViewEval achieves superior compression (24×–40×) and lower error rates, while also revealing capability differences among agents for efficient development.
By Xinshuai Guo, Junjie Wu, Dolly Deng, Yinghui Li, Hai-Tao Zheng, Suncong Zheng, Maxm Pan
MiST (Mid-trained Security Transformer) is a suite of 8B and 32B language models tailored for cybersecurity, achieving strong performance on public benchmarks. The approach uses a mid-training stage that adapts general pre-trained models to the domain by curating a compact, expert-vetted seed corpus and generating high-quality synthetic training data, rather than continual pre-training on large raw text. MiST checkpoints improve mean cybersecurity accuracy by +13.1 and +8.6 absolute percentage points over Qwen baselines for 8B and 32B models, respectively, and provide a stronger initialization for downstream task-specific fine-tuning and reinforcement learning.
By Oded Ovadia, Elad Ben Zaken, Elad Guttman, Orly Moreno Kadosh
The study evaluates deep learning surrogates for wildfire spread prediction, training four architectures on 10,584 high‑resolution simulations from Catalonia. Results show that only surface fuel load significantly predicts burn probability, and convolutional models mainly use distance to the fire front while a transformer model emphasizes fuel and terrain. When applied to a new region without retraining, the models still perform reasonably, with only a modest accuracy drop.
By Marcin Lawenda, Aleksandra Krasicka, David Caballero, Luis Torres, {\L}ukasz Szustak
arXiv:2602. 02888v2 Announce Type: replace-cross Abstract: Hallucinations remain a major obstacle for large language models (LLMs), especially in safety-critical domains.
By Ahmad Shapiro, Karan Taneja, Ashok Goel
The paper investigates how multi‑agent large language models (LLMs) can correct each other’s mistakes, but also how peer pressure can overturn correct answers. It argues that a safeguard— a ‘brake’ that blocks harmful revisions while allowing beneficial ones— is essentially a correctness probe, and that models’ self‑knowledge (measured by AUROC 0.64–0.89) limits the effectiveness of such a brake. The authors find that even white‑box steering cannot break this ceiling, and that adding information before revision, rather than filtering after, is the more promising approach.
By Yibo Hu
The paper demonstrates that emergent capabilities in machine learning models can be forecasted with lead time, calibrated uncertainty, and controlled false‑alarm rates. Using per‑seed analysis on transformers, the authors show that the formation time of a previous‑token head predicts the emergence of an induction head with Spearman ρ = 0.977 and a median lead of 975 training steps. Conformal intervals, blind pre‑registered tests, and a multiplicative rule relating anchor and event times further validate the predictive framework across multiple model families and configurations.
By Gunner Levi Howe
The paper demonstrates that architectural changes—specifically looped transformers and boundary operators—can alter scaling exponents in pre‑training, yielding exponential performance gains for a given computational budget. Looping, or recursive depth, enables model growth that matches larger models (e.g., a 7.4B looped architecture matching GPT‑3 13B) with significantly less compute, while boundary operators provide additional, though smaller, efficiency improvements. In data‑constrained, multi‑epoch scenarios, increasing loops with scale serves as a useful regularizer, suggesting that deeper computational depth drives compute‑efficiency gains that grow with model size.
By Zixi Chen, Akshay Vegesna, Samip Dahal, Andrew Gordon Wilson
The paper introduces a novel generalization of the Adam optimizer to manifold settings, specifically targeting homogeneous spaces such as the Stiefel, symplectic Stiefel, and Grassmann manifolds. By exploiting a global tangent space representation (the Lie subspace), the authors eliminate the need for projection steps and enable all Adam operations to be performed directly on these manifolds. The new optimizer is applied to train transformers and a symplectic autoencoder, achieving orthogonality constraints to machine precision and outperforming existing methods.
By Benedikt Brantner
The paper introduces DantinoX, an open‑source JAX/Flax library that unifies autoregressive decoding, discrete masked diffusion, and continuous flow‑matching language modeling under a single modular Transformer backbone. By keeping the backbone, tokenizer, initialization, and training infrastructure consistent, users can switch between generation paradigms, attention mechanisms, or hardware topologies with only a configuration change. This design enables controlled cross‑paradigm comparisons within one API for training, streaming inference, and benchmarking.
By Marco Simoni, Aleksandar Fontana, Giulio Rossolini, Andrea Saracino
The paper re‑examines the impact of removing the encoder from Action Chunking Transformers (ACT), a model used for robot manipulation learning. Contrary to the original claim that encoder removal drops success rates from 35% to 2%, the authors find no such dramatic decline in their re‑runs, though minor variations remain uncertain. They attribute the discrepancy to factors like training length and checkpoint selection, and note that the encoder’s latent variable offers little reconstruction benefit on the tested benchmark, while its removal speeds up training.
By Bo Kang
The paper examines how large language model (LLM) services route queries to models of varying size based on a cheap complexity estimate. It finds that this routing is not register neutral: queries written in non‑standard English registers (e.g., African American English or second‑language English) are systematically assigned to lower‑capacity models because they appear shorter due to omitted function words. Experiments on 37,704 learner sentence pairs and a controlled corpus show that this bias leads to significantly lower accuracy across all model tiers, including the highest‑capacity cloud models, while the routing decision itself adds little marginal cost.
By Simran Koul
The article examines how Byte‑Pair Encoding (BPE) tokenization handles Polish, an inflectional language, and finds that BPE tends to stabilize frequent surface fragments of grammatical exponents rather than true grammatical categories. It introduces the concept of grammatical form anchoring, showing that certain Polish verb forms can signal the speaking subject without an explicit pronoun, and highlights that language models may lack a stable grammatical "I" and can shift gender or mirror user forms. The study proposes Roclawski’s segmentation‑flexional forms as a diagnostic framework and suggests that more stable Polish modeling would require sublexical stabilization, anchoring grammatical form in the inflectional system, and maintaining the grammatical "I" in dialogue.
By Elzbieta Dawidek (University of Lower Silesia DSW Ideis)
The paper introduces FlowGen, a system that automates the construction of use case flows using large language models (LLMs). FlowGen extracts semantic elements via an LLM-based Semantic Information Processing module, builds a Semantic Relational Graph encoded by an enhanced R-GAT for basic flow generation (BFGen), and adds branch point prediction (BPP) and branch-conditioned alternative flow generation (AFGen). Experiments on 13 public and 7 industrial datasets show FlowGen outperforms baselines across precision, recall, F1, and AUC metrics for all three components.
By Guangyu Wang, Bangqi Li, Ji Wu, Zhijun Shao
The paper introduces a structured framework for claim-level discourse analysis in dense health narratives, addressing the limitations of existing topic- or sentiment-based representations. It identifies an average of 13.22 atomic claims per minute in social media health videos and proposes tuples that link each claim to thematic aspects, stance, and multidimensional pragmatic attributes. A benchmark of 1,191 manually annotated claims from 60 videos across four health domains is created, and experiments show that large language models perform well on thematic categorization and stance prediction but struggle with high-dimensional pragmatic profiling, indicating a need for task-specific inference strategies.
By Farnoushsadat Nilizadeh, Elham Pourabbas Vafa, Shirin Nilizadeh, Eduard Dragut
arXiv:2609. 18960v1 Announce Type: new Abstract: Quality-aware synthetic-data selection rests on a proxy: examples that an LLM judge rates as good should also help a downstream model learn.
By Son Ha Xuan, Phat T. Tran-Truong, Xuan-Bach Le
The paper introduces "question archaeology," an evaluation task that asks models to infer the single, authentic question that motivated a text. It presents a new dataset of commissioned texts paired with their original research questions and distractors, and evaluates both proprietary and open‑source LLMs. Results show newer models outperform older ones, with BERT-based models lagging, and current LLMs even surpassing human performance on this task.
By Claudiu Creanga, Liviu P. Dinu