The paper demonstrates that a natural-language interface can be trained offline to control a xenobot—a synthetic multicellular construct—by using a vision‑language model to judge whether archived intervention outcomes match language descriptions. By treating existing intervention–outcome data as a fixed dataset, the authors train a language‑to‑intervention mapping without new experiments, achieving 80% accuracy on held‑out data compared to a 66.7% baseline. This approach shows that language‑driven control of living systems can be learned purely from archival data and automated visual assessment.
By Nam H. Le, Douglas Blackiston, Michael Levin, Josh Bongard
The paper introduces a method for looped transformers that share a key‑value cache across recursions, reducing memory usage without sacrificing quality. Experiments show that the Looped Prediction Transformer (LPT) and its hybrid variant achieve lower perplexity on FineWeb‑Edu while using 76‑79% less context memory compared to standard transformers. Analysis reveals that shared memory develops distinct representations and serves as a gradient highway, enabling later recursions to focus on shared information.
By Giovanni Monea, Keshav Ramji, Yousef El-Kurdi, Luis A. Lastras, Yoav Artzi, Nathan Godey, Ram\'on Fernandez Astudillo
Inherit-MAS introduces explicit inheritance at both workflow and execution levels for multi‑agent systems built from large language models. A meta‑model generates a workflow of specialized agents, while a judge evaluates each candidate and diagnoses deficiencies. During refinement, workflow inheritance discards unhelpful nodes and applies validated edits, and execution inheritance reuses stored results only when the request and context match, reducing redundant computation.
By Songtao Wei, Yi Li, Zhichun Guo, Bingzhe Li
DAGS introduces a lightweight, attention‑free conditioning scheme that disentangles appearance and geometry for a frozen image diffusion transformer (DiT), enabling high‑fidelity, temporally stable renders with independent control. Two small convolutional encoders generate per‑frame conditioning features, which are injected as learned residuals into the image tokens, avoiding the quadratic cost of attention. Coupled with a recurrent lighting stabilizer and a training‑free temporal guidance term, DAGS transforms a per‑frame image model into a streaming renderer that outperforms real‑time denoisers and diffusion renderers in PSNR and temporal stability while requiring far less compute than path tracing.
By Karthik Mohan Kumar, Damian Andrysiak, Pedro Antonio Pena, Kunal Tyagi, Rama Harihara
Pincer is a defense mechanism for coding agents that operates at the resource layer, complementing existing tool‑call defenses. It uses a digital twin—an isolated‑context model that learns and enforces dynamic, user‑specific least‑privilege policies—to act as a proxy for the agent’s permission requests. The authors introduce a user‑centric dataset to emulate the learning phase and demonstrate that Pincer outperforms several baselines in both security and utility, especially against adversarial attack types.
By Mayank Rathee, Alexander Stepanov, Shalin Madabhavi, Jinhao Zhu, Raluca Ada Popa, Ion Stoica
The paper introduces CausalBridge, a framework that uses causal structure to assign semantics to unnamed variables in numerical measurements. By discovering a causal graph from the data and solving for variable embeddings constrained by this graph, the method aligns variable meanings with a language model, outperforming association‑based approaches. Experiments on questionnaires and robotics scenarios show high accuracy even when most variable names are masked, enabling rapid and cost‑effective system naming.
By Shuhao Zhang, Xuran Zhou, Han Guo, Pengtao Xie, Yujia Zheng
TPBench is a benchmark for dialogue compression that focuses on turning-point eviction, where a user’s change in intent or constraints is lost during compression. It evaluates three information targets—initial user goal, current slot value after revision, and both together—using existing dialogue-state annotations from MultiWOZ and SGD, without requiring new crowdsourcing. The benchmark shows that different compression methods rank differently across probes, and that deleting the turn containing a user update significantly harms current-value accuracy, while removing irrelevant turns does not.
By Minji Park, Seunghyun Yoon, Hyuk Lim
Dynamic Expert Pruning for Multi-Agent Systems introduces DEP, a method that generates a per-request mask of experts based on an agent’s system and task prompts. Unlike static pruning, DEP uses a lightweight predictor trained on workflow transcripts to activate only the experts needed for each specific request, requiring no per-configuration calibration. Experiments across various tasks, models, and MoE architectures show DEP outperforms static pruning and merging baselines, especially when few experts are retained, and it generalizes to unseen workflows.
By Jabin Koo, Soheil Abbasloo, Sungjae Lee, Jungseul Ok
Sentry is a failure‑management layer for large language model agents that learns from failures at test time. It retrieves relevant lessons from an external playbook when a failure occurs, verifies recovery without task rewards, and stores new lessons only if recovery succeeds, keeping the playbook out of the agent’s context. Across multiple benchmarks, Sentry outperforms both runtime‑intervention and context‑evolution baselines, and its lessons transfer to unseen tasks.
By Changxiu Ji, Amy Lu, Qizheng Zhang, Kunle Olukotun
FrugalEvo introduces a cost‑aware evolutionary framework that pairs a high‑cost, powerful LLM with a cheaper LLM to explore and refine program solutions, respectively. It incorporates a cache‑efficient evolution process that maximizes prefix reuse across steps and defines a Budget‑Aware Area Under the Curve (BA‑AUC) metric to evaluate solution quality within a fixed cost budget. Across ten mathematical and systems optimization tasks, FrugalEvo matches or surpasses state‑of‑the‑art baselines in final solution quality and achieves higher BA‑AUC on nine tasks, while delivering new state‑of‑the‑art performance on circle packing for costs as low as $0.55 to $1.68.
By Hui Chen, Xuan Qi, James Xu Zhao, Zhaopeng Feng, Shilong Liu, Kuang Xu, Pang Wei Koh, Bryan Hooi
The paper introduces a counterfactual evidence audit to assess how large language model agents respond to different sets of documents. By exposing agents to two mirrored sets of five documents and measuring changes in six downstream decisions, the authors predict how the agents will behave when presented with a full 45‑document context. Across multiple open‑weight model families and tasks, the five‑document audit accurately forecasts full‑context effects, reducing prediction error by 62% and correctly identifying the direction of most material effects.
By Rana Muhammad Usman
arXiv:2602. 13298v4 Announce Type: replace-cross Abstract: This paper presents a controlled comparative study of convolutional neural network (CNN) topology and image classification performance across the architectural families VGG, ResNet, and GoogLeNet, evaluated on CIFAR-10 under a unified training protocol.
By Manfred M. Fischer, Joshua Pitts
The paper introduces ASER, a pipeline that fine‑tunes a large language model to extract sensory attribute‑value pairs (e.g., color: matte black, scent: vanilla) from product reviews and distills them into a compact sensory bank for each catalog item. During recommendation, a frozen backbone uses a lightweight relational metric over this bank, achieving consistent improvements in HR@10 and NDCG@10 across five Amazon domains and four backbones, with evidence‑grounded extractions validated in 94.8% of cases. The study demonstrates that structured, evidence‑based sensory features enhance ranking quality beyond generic non‑sensory controls.
By Yeo Chan Yoon, Chanjun Park, Kyuhan Koh
The paper presents a systematic study of token consumption in AI agents performing coding tasks. It finds that agentic tasks are far more expensive—about 1000 times more tokens than code reasoning or chat—primarily due to input tokens, and that token usage varies wildly, with accuracy peaking at moderate costs. Models differ significantly in efficiency, and current frontier models cannot reliably predict their own token usage, often underestimating it.
By Longju Bai, Zhemin Huang, Xingyao Wang, Jiao Sun, Rada Mihalcea, Erik Brynjolfsson, Alex Pentland, Jiaxin Pei
The study investigates why large language models (LLMs) show different triage performance when answering clinician‑authored vignettes in multiple‑choice versus free‑text formats. Using sparse‑autoencoder features on Gemma 3 and Qwen3 models, the authors find that medical information is encoded similarly in both formats, but at the decision token the multiple‑choice scaffold dominates, with over 91% of attribution coming from scaffold‑peaking features. The effect varies by model, and shuffling option order eliminates simple positional bias, suggesting the format influence is tied to answer selection rather than earlier case processing.
By David Fraile Navarro, Berardino Como, Jialei Sheng, Soundariya Ananthan, Shlomo Berkovsky
The paper proposes a training‑free detector that uses sentence‑level context sensitivity to identify unsupported content in retrieval‑augmented generation (RAG) answers. By re‑scoring each sentence with full context, no context, and each chunk removed, the method flags the chunk whose removal most reduces a sentence’s likelihood as the likely source. Evaluated on RAGTruth, TofuEval, and RAGBench, the detector outperforms answer‑level faithfulness scores, achieving AUCs up to 0.745 and matching per‑chunk fact‑checkers while using only a fraction of the compute required by large‑language‑model judges.
By Mohamed Aly Bouke
Latent-MOPD is a new on‑policy distillation method that allows a single large language model student to learn from multiple specialist teachers by using both the teachers’ output distributions and their hidden state representations. The approach selects late‑layer targets based on teacher‑student relationships, bridges hidden width differences with a shared projection, and groups updates by domain, enabling gradual shift from representation to token supervision. Experiments show that Latent‑MOPD outperforms token‑only, representation‑only, and uniform‑averaging baselines across nine benchmarks in math, code, and logic, and even surpasses the best individual teacher on most tasks.
By Zhengyu Fang, Seoyeon Hong, Jie Yang, Muyang Li, Koyoshi Shindo, Brandon Joseph Lwowski, Jing Li
The paper introduces the Radio Alternating Operator Ansatz (RAOA), a recurrent computing architecture that uses programmable radio propagation as a computational depth. RAOA alternates energy‑derived problem updates with mixing updates over a persistent latent state, enabling repeated compositional passes even with the same learned controls. Experiments on discrete optimization, simulation of programmable propagation, and adaptation of pretrained language models show that RAOA can improve solution quality, handle higher‑order interactions, and adapt models with performance close to shallow MLPs, all within a single recurrent framework.
By Toshiaki Koike-Akino
The paper investigates how Mixture-of-Experts (MoE) Particle Transformers perform on the 188-class JetClass-II jet classification task. By varying expert count, routing capacity, top‑K, and auxiliary loss, the authors find that top‑1 MoE models can surpass dense baselines with similar nominal compute, but adding more experts yields diminishing accuracy gains. Activating multiple experts per token improves predictions at higher computational cost, and routing analyses reveal that expert assignments correlate with particle identity and kinematics, though this correlation does not consistently predict performance.
By Kaushik Pendiyala, Haris Zia, Trevin Lee, Timothy Legge, Alejandro J. De Leon, Zihan Zhao, Aaron Wang, Abhijith Gandrakota, Jennifer Ngadiuba, Richard Cavanaugh, Javier Duarte
The paper investigates how well a simple spatially shared linear map can predict changes in a model’s internal feature space caused by various image manipulations, including geometric, photometric, occlusion, and diffusion-generated semantic edits. Experiments across ConvNeXt, SwinV2, and DINOv3 show that this linear operator often performs nearly as well as more complex, higher‑capacity probes, especially in deeper layers of supervised backbones. The study concludes that a shared linear map is frequently sufficient to capture diverse image edits, with its leading singular components encoding semantic content and higher‑rank components refining details.
By Elias Krey, Nils Neukirch, Nils Strodthoff