The paper proposes a hybrid narrative framework that blends pre‑authored content with real‑time LLM generation, using semantic similarity to balance authorial control and player agency. It also introduces a self‑evolving, feedback‑driven optimization system that refines prompts through in‑context learning, removing the need for manual prompt engineering. A case study and participant evaluation demonstrate that this approach improves narrative coherence while maintaining player agency and aligning with authorial intent.
By Ruisheng Sun, Jiaqi Shao, Weijia Han, Linxi Ju, Yunzhen Zhang, Binyan Sun, Rong Chen, Benjamin Blas Maristany, Ye Lu, Rui Hu, Ernesto Escobar, Bing Luo
GenDB‑V is a system that uses AI agents to automatically synthesize database engines tailored to specific hardware and workloads, while also providing formal verification of correctness. It introduces SQLSpec to formalize SQL semantics, enabling joint synthesis of efficient code and modular proofs that bridge low‑level execution to SQL semantics. Evaluations on TPC‑H and SEC‑EDGAR show significant speedups over existing engines and reveal correctness issues in alternative LLM‑based synthesis methods.
By Jiale Lao, Immanuel Trummer
The paper introduces Library of Layers (LoL), an architecture that dynamically orders transformer layers during inference, allowing a single model to adapt its computation to the task at hand. LoL’s single hyperparameter—the size of its global library of transformer blocks—enables a smooth transition between parameter reuse and sparsity, unifying dense, depth‑recurrent, and mixture‑of‑experts trade‑offs. Experiments show LoL matches or exceeds these baselines on text, synthetic Sudoku and maze tasks, and demonstrates task‑specific routing in multilingual and safety‑focused settings.
By Abdellah Rahmani, Sean McLeish, Ang Li, Sanae Lotfi, Pascal Frossard, Micah Goldblum
The paper introduces Anthropogenic Regional Adaptation, a new framework for aligning vision‑language models with specific regional contexts while preserving global performance. It proposes the GG‑EZ method, which filters regional data and merges models to achieve this alignment. Experiments on three VL architectures and a Southeast Asia case study show 5‑15% improvements in cultural relevance metrics with over 98% retention of global performance.
By Samuel Cahyawijaya, Peerat Limkonchotiwat, Tack Hwa Wong, Hitesh Laxmichand Patel, Amit Agarwal, Manuel Antonio Rufino, Carlos Rafael Catalan, Muhammad Reza Qorib, Vicky Feliren, Holy Lovenia, Aye Hninn Khine, Frederikus Hudi, David Anugraha, Alham Fikri Aji, Romrawin Chumpu, Viet-Thanh Pham, Minghan Wang, Mohamed Fazli Imam, Ruochen Zhang, Joseph Marvin Imperial, Khumaisa Nur'aini, Do Xuan Long, Musa Izzanardi Wijanarko, Joel Ruben Antony Moniz, Patrick Amadeus Irawan, Hanif Muhammad Zhafran, Isaiah Flores, Salsabila Zahirah Pranida, Jun Kevin, Jostin Jerico Rosal, Patricia Nicole Monderin, Kun Kerdthaisong, Ahmad Mustafid, My Chiffon Nguyen, Natchapon Jongwiriyanurak, Siva Worajitwannakul, Haochen Li, Adrian Xuan Wei Lim, Bin Wang, Muhammad Ravi Shulthan Habibi, Lynnette Hui Xian Ng, Mithil Bangera, Yeshil Bangera, Priyaranjan Pattnayak, Dun Li Chan, Sherissa Caren Djuniwar, Cho Chan Myei Oo, Hee Ming Shan
Circuit Oracle is an off‑the‑shelf large language model that, using a multi‑agent harness and task‑specific skills, can answer natural‑language questions about a target model by reading its attribution graph via tool calls. The system was evaluated on three safety‑relevant proxy tasks—detecting spurious features in probe circuits, eliciting hidden knowledge from taboo‑finetuned models, and suppression jailbreaking via feature‑level interventions—achieving 78.2 % accuracy on spurious‑feature detection and mixed results on the other tasks. In addition to task scores, Circuit Oracle produces pruned subgraphs and context‑specific feature explanations, which the authors release for further study.
By Shariar Kabir, Hong Kiat Tan, Swastik Agrawal, Sai V R Chereddy, Sriram Balasubramanian
S1CogBench is a new benchmark that evaluates System One decision models—typed decision models like Jev and Laya—on 24 cognitive tasks. It tests whether these models can use evidence, respond to feedback, handle memory load, and anticipate future opportunities, while keeping conditions and scoring separate from any disclosed adaptations. The study finds wide variability in reliability among decision models, shows that good performance does not guarantee human-like predictions, and reveals that speed largely stems from generating decisions without text output.
By Dengzhe Hou, Kazunori D Yamada
Depth pruning removes entire Transformer blocks to shrink and speed up language models, but the resulting damage is uneven across capabilities. In a study of four 1.2–1.7 B instruction‑tuned models, function calling and mathematical reasoning collapsed in every case, while other skills varied in order of degradation. Fine‑tuning a short ‘healing’ run can restore some abilities, with the choice of which blocks to retrain and the allocation of a limited token budget influencing recovery outcomes.
By Piyush Dhoka, Vedant Pandhare, Varun Inamdar, Aadarsh Pathre, Pavitha Nooji
The paper investigates gated knowledge distillation (KD) for low‑resource Bangla generation, testing two gating strategies—EWAD and CHAD—to selectively trust teacher signals. Across six Bangla tasks and multiple student models, neither gate improves over plain KD, and KD generally underperforms cross‑entropy fine‑tuning on gold outputs. The primary advantage of KD is regularisation, with teacher‑free label smoothing narrowing the performance gap, though gating only modestly alters this effect.
By Dipto Sumit, Zobeida Zaman Chowdhury, Syeda Fabiayee Jahan, Farig Sadeque
The article presents a detailed, step‑by‑step derivation of the backpropagation algorithm for a fully connected feedforward neural network with sigmoid activations and squared‑error loss. It starts with a single‑weight example, extends to general error terms and matrix equations, and then scales the derivation to multiple training examples and mini‑batch gradient descent. Each equation is annotated with dimensions and linked to a NumPy implementation trained on handwritten Arabic digits, with full code provided.
By Faris Alasmary
The paper investigates how large language model (LLM) agents can collude in multi‑agent systems and introduces the J‑space as a tool to detect and steer such collusion. Through interpretability experiments on DeepSeek‑V4‑Flash and a steganographic card‑counting task, the authors show that the J‑space captures collusion behaviors, such as misrepresentation and hidden information. They also propose J‑PACT, an unsupervised method that successfully identifies colluding pairs across several open‑weight models, achieving high AUROC and accurate pair identification.
By Marco Molinari, Gabriele Bettineschi
The paper presents a bilingual (English/Arabic) benchmark for evaluating how well open-weight, locally deployable language models can accurately recall specific Hadith texts. Five models—Mistral‑7B, Gemma3‑4B, Jais‑7B, ALLaM‑7B, and Phi4‑mini‑3.8B—were tested on 3,000 Hadiths per language across six canonical Sunni collections, with each model given only a bibliographic reference and asked to reproduce the exact text. None of the 30,000 responses achieved both high lexical and semantic fidelity, and Arabic outputs performed uniformly worse than English, indicating that these models, without retrieval augmentation, cannot reliably provide verifiable religious scripture.
By Dayam Nadeem, Safdar Mustafa, Ameen Ur Rehman, Shahab Saquib Sohail, Mohammad Nadeem, Usman Naseem
The paper investigates how a language model’s first answer to a yes/no question influences its subsequent answer to a different question about the same entity, a phenomenon termed self‑commitment. By fixing the initial answer and measuring changes in the second across 3,600 question pairs for 22 open‑weight models (0.5B–72B parameters), the authors find that each answer writes a commitment direction into the residual stream, and the similarity of these directions predicts answer coupling. They show that ablating a shared direction reduces coupling significantly, while the reading circuit for this direction varies with model size, instruction tuning, and prompts, and that sparing the attention sink during ablation preserves low perplexity cost in most models.
By Marcus Armstrong
MultiInpaintCOCO extends the English InpaintCOCO dataset to 13 typologically diverse languages, creating a multilingual benchmark of 953 minimal-pair examples focused on fine-grained visual details. The study evaluates state-of-the-art multilingual multimodal retrieval models, revealing that performance drops significantly for lower-resource languages and that many models exhibit a systematic bias toward English captions. The benchmark and code are publicly available at https://github.com/ufal/multi-inpaint-coco.
By Andrei-Alexandru Manea, Jind\v{r}ich Libovick\'y, Philipp J. R\"osch
VeriHarm-Bench is a new benchmark that measures harmful success in large language model coding agents by automatically generating verifiable tests that detect unintended harm while preserving the original task objective. The framework minimally perturbs coding tasks to elicit harmful behavior and synthesizes executable harm tests alongside task‑correctness tests. In evaluations, frontier agents such as GPT‑5.6‑Sol and Claude‑Opus‑5 achieve high task accuracy (≈95%) but only a 37% safe success rate, showing that task success can coexist with significant unintended harm.
By Wei-Lin Chen, Zhepei Wei, Xinyu Zhu, Yu-Min Tseng, Tzu-Han Lin, Yu Meng
The paper introduces SHIFT, a method for unlearning in large language models that avoids explicit forgetting targets by shrinking a fine‑tuned model toward its pre‑fine‑tuning weights and briefly relearning the retain set. In the linearized regime, this shrink‑and‑heal lookahead moves the model a fixed fraction toward the retrained model, and SHIFT extrapolates these per‑token changes into distillation targets for iterative refinement. Evaluated against eleven baselines under the OpenUnlearning protocol, SHIFT achieves the highest aggregate score across all TOFU settings, aligns 3.7× closer to the oracle than the next best method, and preserves the most retain knowledge on MUSE, maintaining proximity to the oracle even under larger optimization budgets and relearning attacks.
By Hoang-Chau Luong, Lingwei Chen
The paper introduces TAILOR, a framework that uses Conditional Value at Risk (CVaR) to guide large language model (LLM) synthesis of heuristics for computer systems. By optimizing the paired CVaR of cost differences relative to a baseline, TAILOR ensures improvements in both average performance and tail optimality gaps without needing to compute the unknown optimum. Experiments on Azure GPU batch admission, Azure VM packing, and server placement demonstrate that TAILOR achieves the best mean–tail trade‑off and outperforms robust, human‑designed alternatives in the tail while maintaining comparable average gains.
By Pantea Karimi, Roozbeh Bostandoost, Dany Rouhana, Siva Kesava Reddy Kakarla, Pooria Namyar, Mohammad Alizadeh, Venkat Arun, Behnaz Arzani
The paper introduces Orthogonal Drift Correction (ODC), an inference-time guidance method that mitigates semantic drift in text-to-image generation without retraining or extra user input. ODC operates in two stages: first it detects drift by comparing the generated image to the prompt in a shared vision-language embedding space, isolates the orthogonal component of the drift, and converts it back into text; second it refines the text conditioning for a second generation pass using an adaptive rank-reduced concept removal module. Experiments show ODC improves prompt-image alignment, raising CLIP scores by up to 5.4% on COALBench-300 and PartiPrompts, and boosting the 3-in-1 score by 4.5% on T2I-CompBench’s complex subset.
By Kasra Kamali, Basim Azam, Tuan Duong Trinh, Jean Honorio, Naveed Akhtar
The paper presents a rubric‑based assessment pipeline that uses lightweight language models and a vision‑language model to evaluate student reports in cybersecurity education. It demonstrates that small models can achieve up to 97% accuracy on text‑only assessments and 94% on rich‑text assessments across various low‑resource hardware setups. The solution is open‑source, locally deployable, and designed to reduce instructor workload while addressing privacy and cost concerns.
By Sandesh Kumar Kafle, Jan Vykopal, Valdemar \v{S}v\'abensk\'y
The paper examines how different preprocessing steps in fine‑tuning data influence the performance and fairness of a generative large language model recommender system. Using a controlled pipeline that includes supervised fine‑tuning, catalog‑constrained decoding, and prompt augmentation, the authors evaluate three preprocessing choices—item‑title representation, demographic balancing, and auxiliary item‑knowledge enrichment—on the MovieLens‑1M dataset. Results show that catalog constraints eliminate out‑of‑catalog items, prompt augmentation improves consistency and utility, and that demographic balancing and knowledge enrichment can reduce utility gaps for certain groups while preserving overall performance, though gender balancing may lower utility without improving gender indicators.
By Furkan Yasir Goksu, Yaren Yilmaz, Sule Gunduz Oguducu
The paper investigates how monitoring the internal representations of Large Reasoning Models (LRMs) during their Chain-of-Thought (CoT) reasoning can predict future behavior. By evaluating a probe at each generated token, the authors construct probe trajectories that capture the continuous evolution of a concept’s probability, and extract signal‑processing features to characterize volatility, trend, and steady‑state behavior. These trajectory features improve predictive performance over static probes, adding an average of 21.7 AUROC points, and the study demonstrates that template‑based training data and pooling operations are essential for effective monitoring across multiple reasoning models and datasets.
By Maciej Chrabaszcz, Aleksander Szymczyk, Marcin Sendera, Tomasz Trzcinski, Sebastian Cygert