HyperGuide introduces a hyperbolic geometric signal to guide multi-step reasoning in large language models, addressing the trade-off between efficient single-pass generation and computationally heavy tree-search methods. By projecting LLM hidden states into hyperbolic space, the approach leverages the space’s asymmetry—compact near the origin and exponentially expanding toward the boundary—to encode solution proximity and branch differentiation. A lightweight head and a fine-tuned low-rank adapter use this signal to improve reasoning accuracy, especially on deeper reasoning chains, with consistent gains across multiple benchmarks.
By Yuyu Liu, Haotian Xu, Yanan He, Sarang Rajendra Patil, Mengjia Xu, Tengfei Ma
The paper presents multicalibration as a method for unbiased prevalence estimation when measurement devices (diagnostic tests, classifiers, or large language models) exhibit covariate shift. It shows that standard calibration fails under such shift, while multicalibration—calibrating conditional on input features—maintains near-zero bias in simulations and real-world applications. The authors demonstrate the approach on employment data across U.S. states and political text classification across four countries, emphasizing the need for calibration data that covers key feature dimensions.
By Fridolin Linder, Thomas Leeper, Daniel Haimovich, Niek Tax, Lorenzo Perini, Milan Vojnovic
The paper introduces IMAVB, a 500‑clip benchmark that tests whether omnimodal large language models can detect when a textual premise contradicts their visual or audio input. Experiments on eight open‑source models and Gemini 3.1 Pro reveal a Representation‑Action Gap: internal states encode mismatches, yet the models rarely reject false premises, exhibiting under‑rejection or over‑rejection. A probe‑guided logit adjustment improves rejection behavior, suggesting the main bottleneck is in translating perception to action rather than in perception itself.
By Trung Nguyen Quang, Yiming Gao, Fanyi Pu, Kaichen Zhang, Shuo Sun, Ziwei Liu
The study investigates how different observation fidelities affect the problem‑solving performance of embodied large language model (LLM) agents. Using the Lockbox mechanical puzzle, agents were tested with raw RGB, RGB‑D, and perfect ground‑truth symbolic observations in both physical and simulated environments. Surprisingly, agents performed best with raw RGB input and worst with perfect observations; introducing moderate perceptual noise in simulation further improved success rates, suggesting that some errors can mitigate repetitive action loops.
By Oussama Zenkri, Oliver Brock
The paper introduces CFGzip, an offline token space compression technique that reduces the search space for context-free grammar (CFG) decoding engines. By compressing the token vocabulary, CFGzip achieves up to 75‑fold latency reductions during batched inference, bringing overhead on the hardest grammars down to roughly 1.2–2×. This enables scalable constrained decoding for complex CFGs that were previously impractical due to high computational costs.
By Michael Sullivan, Alexander Koller
The paper introduces masked self‑distillation, a post‑training framework that trains a language model to internalize portions of its own intermediate reasoning traces. By varying the fraction of trace internalized, the authors demonstrate that models can achieve higher inference efficiency and improved task performance on math and graph‑coloring problems. Experiments on Qwen3‑4B and Qwen3‑8B show that the method generalizes well to in‑domain out‑of‑distribution cases without catastrophic forgetting, and that supervised fine‑tuning alone can reduce trace length at the expense of generalization.
By Durgesh Kalwar, Vardhan Palod, Jaya Adithya Pavuluri, Subbarao Kambhampati
The paper presents a reinforcement learning framework for training autonomous large language model agents to act as sellers in multi-product bargaining markets. It models the problem as a Partially Observable Markov Decision Process and introduces a structured message protocol to map natural language into a regulated decision space. Using Reinforcement Learning from Verifiable Rewards, the trained agents match limited inventory to buyers more effectively, achieving seller surplus extraction and allocation quality comparable to or better than trillion‑parameter models, and generalize to unseen market conditions.
By Shuze Daniel Liu, Claire Chen, Jiuqi Wang, David Simchi-Levi, Thorsten Joachims
CANTO is a transformer neural operator that directly maps continuous CAD geometry—represented as sequences of NURBS patches—to physical fields without requiring meshing. It tokenizes control points, knot vectors, and weights to predict continuous surface and volume fields at arbitrary query locations. Evaluated on automotive and aircraft aerodynamics benchmarks, CANTO achieves state‑of‑the‑art accuracy, notably a 19.8% reduction in surface‑pressure error on HiLiftAeroML, and enables gradient‑based inverse design, finding lower‑drag designs on AhmedML.
By Daniel Leibovici, Nikola Borislavov Kovachki, Dawon Ahn, Ruben Ohana, Ira J. S. Shokar, Abouzar Ghasemi, Semih Akkurt, Rishikesh Ranade, Neil Ashton, Jan Kautz, Jean Kossaifi
The paper introduces WaLLM, a general‑purpose large language model chatbot deployed on WhatsApp to investigate how users interact with open‑ended AI in everyday settings. The study finds that health and well‑being queries dominate user interactions, and that proactive communication and communal lists—adapted to WhatsApp’s affordances—enhance engagement and content discovery. These insights inform the design of general‑purpose LLM services on messaging platforms.
By Hiba Eltigani, Rukhshan Haroon, Asli Kocak, Abdullah Bin Faisal, Noah Martin, Fahad Dogar
The study investigates whether vision–language models (VLMs) can infer human engagement from gameplay videos using the GameVibe Few‑Shot dataset across nine first‑person shooter games. Three VLMs were tested under six prompting strategies—including zero‑shot, theory‑guided prompts based on Flow, GameFlow, Self‑Determination Theory, and MDA, and retrieval‑augmented prompting—evaluating both pointwise engagement prediction and pairwise prediction of engagement change. Results show that zero‑shot predictions are weak and often do not beat simple majority‑class baselines; retrieval‑augmented prompting improves pointwise prediction in some cases, while pairwise prediction remains difficult, and theory‑guided prompts do not reliably help and may reinforce superficial shortcuts.
"whyItMatters":"The findings highlight a perception–understanding gap in current VLMs, indicating that while they can recognize visible gameplay cues, they still struggle to robustly infer human engagement across games."
By Ziyi Wang, Qizan Guo, Rishitosh Singh, Xiyang Hu
Rewind-IL is a training‑free online safeguard for generative action‑chunked imitation learning policies. It uses a zero‑shot failure detector based on Temporal Inter‑chunk Discrepancy Estimate (TIDE) and a state‑respawning mechanism that returns the robot to a verified safe intermediate state. The system builds a checkpoint library offline with a vision‑language model and monitors self‑consistency online, rewinding execution to the latest safe checkpoint when a failure is detected, thereby improving reliability in long‑horizon manipulation tasks.
By Gehan Zheng, Sanjay Seenivasan, Matthew Johnson-Roberson, Weiming Zhi
The paper investigates the edge‑of‑stability (EoS) phenomenon observed in full‑batch Adam optimization. It identifies Adam’s second‑moment adaptation as a negative‑feedback mechanism that drives the dynamics toward the stability boundary, characterized by the *active curvature*—the preconditioned curvature along the preconditioned gradient direction. The study provides rigorous characterizations across various settings, predicts *gradient reversal* near the edge (consecutive gradients pointing in nearly opposite directions), and shows that averaging iterates suppresses these oscillations, yielding smoother and lower loss curves.
By Yiman Fong, Heng Yang
The paper addresses the problem of large language models (LLMs) over-refusing to comply with benign safety-related instructions. It identifies that a small number of hypersensitive safety heads in transformer attention misfire on hard-safe prompts, causing abnormal attention entanglement and high-entropy routing conflicts that lead to refusals. The authors propose Semantic Routing Calibration (SRC), a lightweight, training‑free inference framework that dynamically suppresses these problematic heads and fuses logits to restore trustworthy reasoning while preserving intrinsic safety performance.
By Zixuan Wang, Bingjie Zhang, He Zhao, Dandan Guo
The paper introduces the Function‑Space Transformer (FST), a neural framework that learns from functions by using a spatially adaptive continuous latent representation. FST places features at anchors whose positions are predicted from input samples and refines these anchors through recursive function‑space interactions, allowing the representation to adapt to each input rather than relying on a fixed grid. Experiments on PDEBench (Burgers and Darcy flow) and ImageNet‑1K show that FST outperforms or matches state‑of‑the‑art baselines such as Perceiver IO and the Vision Transformer while using fewer parameters.
By Guorui Sang, Pedram Rooshenas
The paper introduces Uncertainty‑Normalized Margin DPO (UNM‑DPO), a method that augments direct preference optimization with strength‑dependent margins and a learned prompt scale derived from a heteroskedastic Bradley‑Terry model. Two training objectives—advantage‑only (AO) and whole‑residual (WR)—compare implicit rewards from response log‑probability ratios, with WR enabling identification of the prompt scale and a practical learning procedure. Building on WR, ULNM‑DPO‑WR further normalizes rewards by response length, and experiments on HelpSteer2/3 and AlpacaEval show higher win rates, mean rewards, and shorter responses compared to baseline DPO and SimPO.
By Sadegh Khorasani, Petrus Mikkola, Matthias Grossglauser
LEARN-TS is a method for multivariate time‑series anomaly detection that uses a frozen language model to generate two semantic representations: one that captures window‑specific observation context without exact numerical values, and another that provides a fixed, dataset‑agnostic normality prompt. These representations guide a patch‑masked reconstruction process, allowing the model to estimate normality discrepancy and produce timestamp‑level anomaly evidence. Experiments on four benchmarks show that LEARN‑TS achieves the best mean performance in 13 of 16 dataset‑metric comparisons, and ablation studies confirm the benefits of observation conditioning, joint normality alignment, and semantic references.
By Jahyeob Koo, Kio Yun, Byoungmo Koo, Jun-Geol Baek
The paper introduces SparLeak, a side‑channel attack that exploits a new GPU micro‑architectural leakage called Sparsity‑Induced Memory Access (SIMA) caused by sparse attention in large language models. By capturing SIMA traces during LLM inference, SparLeak can infer user‑query attributes from prefill‑phase traces and reconstruct autoregressive responses from decoding‑phase traces. Experiments on three LLM architectures, three sparse attention mechanisms, and three privacy‑sensitive datasets show high success rates (90.9% for attribute inference and 87.3% for response reconstruction), underscoring the need to address SIMA leakage in sparse‑attention deployments.
By Fahao Chen, Linkang Du, Jinhao Zhou, Peng Li, Zhou Su
The paper introduces ID Balancing, an Integral‑Derivative controller that improves expert load balance in extremely sparse Mixture‑of‑Experts (MoE) models. By scaling its integral term with load error and activating the derivative term only when imbalance worsens, ID Balancing achieves over 50% reduction in worst‑case backbone MaxVio and 12% reduction in training‑average backbone MinVio compared to leading baselines, while preserving language‑modeling performance across various routing settings. The method’s benefits grow with increased sparsity, making it a promising approach for scaling larger MoE models.
By Peng Jin, Zihan Qiu, Zekun Wang, Bo Zheng, Yang Xu, Tian Xie, Xiao Li, Huaqing Zhang, Haoran Lian, Rui Men, Dayiheng Liu
The paper investigates rounding strategies for low‑precision recurrent state caches in Mamba‑style and hybrid language models. It shows that a deterministic golden‑ratio Weyl dither consistently yields quantized models closer to full precision than stochastic rounding, across various models, storage formats, and long decoding horizons, without extra cost. In contrast, round‑to‑nearest can appear effective in short tests but degrades over long generations due to error accumulation.
By Snigdha Chandan Khilar
The paper investigates test‑time scaling for language models, noting that sequential methods often stall due to entering attractors—sets of answers that block further exploration. Across 27 experiments, over half of sequential trajectories hit an attractor within four iterations. Introducing a simple model‑mixing intervention reduces attractor incidence by 21.2 percentage points, expands solution coverage beyond a compute‑matched parallel baseline, and boosts recursive self‑aggregation accuracy by at least 2.2 percentage points.
By Joseph Rance, Fabio Pizzati, Juil Sock, Woody Bayliss, Marc G\'orriz Blanch, Philip Torr, Adel Bibi