arXiv_CL
-
Where Should a Stuck Agent Spend Its Next Tokens? Causal Attribution of Inference-Time Resources
2026-10-20 18:59:01Saiyue Lyu, Jerrold Huang, Yunbei Zhang, Janet Wang, Zhenting Wang, Yingqiang Ge, Haibo ZhangAbstract
Inference-time scaling is often treated as a token-budget problem, but a failed coding agent can seek qualitatively different resources: computation, information, or feedback. We introduce Banzhaf Resource Attribution (BRA), a controlled factorial framework that measures each resource's marginal contribution to repair and its interactions while holding the failed state and revision protocol fixed. We evaluate six models on 194 function-level failures from MBPP+ and HumanEval+ and 126 repository-level failures from SWE-bench Verified. Task-relevant specifications consistently improve repair, whereas public-test feedback provides small gains and retrieved code little to none. The value of additional computation declines with model capability, while information remains useful; resource channels largely exhibit diminishing returns, although complementary information can reinforce itself. Yet three of four frontier models select feedback in 95-100% of cases and obtain lower gains than uniform random resource selection. BRA thus exposes a gap between the resources agents prefer and those that measurably improve repair.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28834
PDF
-
A Parallel Multi-Dialect Arabic Dependency Treebank
2026-10-20 18:45:40Noor Abo Mokh, Mohamed Adel, Jamila El-Gizuli, Go Inoue, Ahmed Elshabrawy, Nizar HabashAbstract
Dependency parsing resource development for Arabic has largely focused on Modern Standard Arabic, leaving dialectal Arabic comparatively under-resourced. We introduce MADARTB, the first large-scale parallel dependency treebank spanning five city dialects: Beirut, Cairo, Doha, Rabat and Tunis, and extending an existing parallel MSA version. In total MADARTB consists of 5,000 sentences annotated using the CATiB dependency scheme. We analyze syntactic similarity across all city dialects and MSA. Although global construction profiles exhibit high similarity across varieties, local syntactic structures frequently diverge, revealing substantial structural variation across dialects. We further establish a cross-dialect dependency parsing comparison using a supervised parser and four LLMs. The supervised parser achieves 87.5% LAS, while the best LLM setup trails at 85.7% LAS across all six varieties on the test set. We release our annotations and models publicly.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28823
PDF
-
LIFT: Scoped Constraints for Evaluating and Training Instruction Following
2026-10-20 18:40:22Eran Goldstein, Yuval Belfer, Yonatan Belinkov, Daniel GissinAbstract
Common instruction following (IF) benchmarks, such as IFEval, evaluate the ability of language models (LMs) to follow simple constraints that can easily be verified programmatically. These same constraints, in turn, are used as reward signals for reinforcement learning. This design may lead to reward hacking of constraints, pushing models away from the main task. Recent attempts to address this resulted in benchmarks with richer constraints, but required expert labor, remained closed-access, and cannot be easily extended. We present Lift-Bench, an IF benchmark that asks a model to draft a document given a topic and a structured outline. Constraints are scoped to certain document sections, providing finer control over constraint density. Lift-Bench includes both verifiable and judged constraints, as well as a training counterpart, Lift-Train. Importantly, Lift-Bench provides both main task and instruction-following constraints, mitigating shallow reward hacking. On experiments with multiple models, we find that frontier models struggle with the harder subsets of our benchmark and analyze model performance along multiple axes. We also find that training on Lift-Train is much less susceptible to reward hacking than training on the common IFTrain. Lift-Bench and Lift-Train are publicly available, as well as the data generating pipeline.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28819
PDF
-
Combining Long-Term Memory and LLMs Comes at a High Cost: From Prompt Injection to Persistent Compromise
2026-10-20 18:33:05Pedro H. Barcha Correia, Matheus Azevedo de S\'a, Paulo H. Gomes de Senna, Giovanni Mandel Martignago, Victor Takashi Hayashi, Charles Christian Miers, Marcos A. Simplicio JrAbstract
The widespread adoption of generative artificial intelligence in recent years has brought a parallel rise in prompt injection and jailbreak attacks. By embedding malicious instructions, these exploits manipulate language models to bypass intended behavior and operational boundaries. In traditional chatbots, such attacks are confined to a single session because state memory is not preserved across conversations. However, persistent memory features in modern conversational interfaces enable cross-session attack propagation, establishing a threat vector known as memory injection. This work explores this emerging vulnerability in leading LLM-as-a-service platforms, highlighting its real-world severity and the need for countermeasures. While existing literature focuses on violations tailored to specific platform features, we present a general threat model. We demonstrate that all NIST attack goals are collectively achievable across ChatGPT Instant 5.3, Claude Sonnet 4.6, and Gemini 3 Flash under this model. We also show that memory injections can exploit multiple attack vectors, including personal memories, projects, skills, organization instructions, and Model Context Protocol (MCP)-connected resources. Our findings reveal that all evaluated services can be manipulated to persistently spread disinformation, generate harmful content, or simulate offline unresponsiveness, even in fresh sessions. In the particular case of Claude, shared-memory features allow an attacker to compromise multiple users within an organization and also exfiltrate user prompts. To address these vulnerabilities, we derive defense mechanisms from our proposed threat model and foundational cybersecurity principles. Our analysis indicates that nearly half of these mitigations remain unimplemented across the target services. Datasets and attack artifacts are available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28810
PDF
-
Where Do Legal Factors Live in LLMs?
2026-10-20 18:24:37Li Zhang, Kevin AshleyAbstract
In case-based legal reasoning, legal professionals identify legally significant fact patterns in a new dispute and compare them with those of decided cases. In AI & Law, HYPO dimensions and CATO factors represent these patterns symbolically, organizing expert knowledge distilled from case narratives as explicit, structured, interpretable, and auditable legal concepts and relations that support case comparison, legal argument, and precedent-based reasoning. Large language models (LLMs) are increasingly used for legal tasks, yet their internal organization of legal knowledge remains difficult to trace. We investigate whether LLMs form internal representations corresponding to all 26 CATO factors. Two research questions examine linear decodability and sparse localization with sparse autoencoders. Experiments cover three open models, Qwen3-1.7B, Qwen3-8B, and Gemma-2 2B, supported by cross-model comparison and human audit of controlled natural-language passages. Results show that all 26 factors are linearly decodable in all three models. Single-feature sparse readouts cover the complete factor catalog, with best-layer test AUCs ranging from 0.86 to 1.00. Five selected features approach full-vector readout performance. We report 78 model-specific layer-feature coordinates, one for each factor in each model. Together, these results show that information corresponding to expert-defined legal factors can be read from the internal activations of LLMs processing legal narratives. By locating these readouts, our study connects legal language and the symbolic abstractions of HYPO and CATO to learned internal representations.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28799
PDF
-
CALLIOPE: Private Insights into AI Use, Without Embeddings
2026-10-20 18:23:08Edith Cohen, Vadym Doroshenko, Badih Ghazi, Charlie Harrison, Pritish Kamath, Alexander Knop, Ravi Kumar, Ethan Leeman, Pasin Manurangsi, Mikhail Pravilov, Adam Sealfon, Yurii Sushko, Marika Swanberg, Chiyuan ZhangAbstract
We introduce CALLIOPE, a top-down framework for discovering topic hierarchies in large corpora of unstructured data (e.g., LLM conversations), while supporting differential privacy (DP). Unlike traditional and even recent LLM-augmented approaches that ultimately rely on clustering vector embeddings, our framework leverages LLM capabilities directly for the entire pipeline. Experiments demonstrate that this approach provides better utility over existing (embedding-based) methods in the DP setting, especially in the high privacy regime. We evaluate the quality of CALLIOPE outputs using a variety of traditionally studied clustering metrics, as well as some novel utility metrics that we introduce.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28797
PDF
-
Is Anyone Enforcing the Cartel? What Reinforcement Learning Teaches LLM Agents in Markets
2026-10-20 18:18:10Hieu Tran, Hong YuAbstract
Evaluations of large language model (LLM) agents report that pricing agents collude at others' expense. Yet collusion is defined by enforcement: a cartel, in which competitors share monopoly profits, holds only if a member who cheats is punished. We introduce an audit that tests enforcement by forcing a member to cheat in counterfactual copies of a market, and use it to study what reinforcement learning (RL) teaches agents in pricing and trading. Small LLMs trained by RL learn what their training rivals reward, but only what they readily try. In self-play, they learn fixed rules ignoring their rivals; against a punishing rival, to be deterred; against a cheating rival, to match its cuts, never the costly price war that would deter it, which they rarely try. A war forced into training is learned, but mostly only where it was forced. Against their own copies, deterred and matching agents can earn the same cartel profits, but only matching ones can enforce them, and then thinly: as a by-product and only against large cuts. Q-learners enforce collusion only in pricing. Prompted LLMs reach cartel-like outcomes mostly through rules that ignore the market. Outcome measures cannot tell these cases apart; the audit can.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28787
PDF
-
Measure First, Then Allocate: Calibrated Rollout Budgets for MCP Agents in Reinforcement Learning
2026-10-20 18:15:52Hieu Tran, Zonghai Yao, Sunjae Kwon, Hong YuAbstract
Training tool-using language models requires costly interaction, yet identical rewards within a rollout group provide no group-relative learning signal. We use reward disagreement to guide harness evaluation and the allocation of rollouts for policy training in Model Context Protocol (MCP) environments. We first introduce Discord, which tests harness edits using exact paired inference and decision-preserving early stopping: only disagreeing paired outcomes favor one harness over another. This insight motivates prospective allocation in BRACE (Bank-calibrated Rollout Allocation under Correlated Episodes). BRACE combines historical task evidence, uncertainty in task success rates, and a correlated reward model to forecast mixed-reward groups before allocating generation budget. Controlled harness experiments favor Discord, while live MCP experiments expose a retention-cost trade-off. At three Qwen3-VL model scales, BRACE exceeds the strongest plotted baseline peaks with fewer new training rollouts; historical-bank collection is counted separately. It also leads the compared methods on BFCL and NESTFUL without further training. Ablations probe the bank, estimator, and allocation score; prospective evaluation checks forecast calibration. These results support forecasting reward contrast before generation.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28785
PDF
-
Scaling Therapy Supervision: Assessing Psychotherapy Fidelity Using Large Language Models
2026-10-20 18:09:44Arvind Pillai, Alexis Roach, Jason Holden, Varsha D Badal, Subigya Nepal, Yu Yvonne Wu, Yuliang Chen, Sudarshan Regmi, Sarah Masud Preum, Nicholas C Jacobson, Nikhil Singh, Eric Granholm, Colin Depp, Andrew CampbellAbstract
Psychotherapy fidelity assessment measures how faithfully a therapist adheres to a treatment protocol, a critical determinant of positive patient outcomes. However, fidelity assessment is bottlenecked by manual expert review, rendering the process labor-intensive and unscalable. To scale therapy supervision and support clinician self-learning, we present an integrated application of large language models (LLMs) that automates psychotherapy fidelity workflows and generates actionable supervisory feedback from authentic, long-form clinical transcripts. We evaluate three open-weight models across multiple task decomposition configurations and reasoning strategies, demonstrating that evidence-first reasoning improves fidelity estimation. Furthermore, we establish the framework's external construct validity on an independent randomized controlled trial (RCT), where it reliably distinguishes Cognitive Behavioral Therapy (CBT) from non-CBT sessions across core therapeutic skills. Through a user study that evaluates the generated feedback, mental health experts indicate that the pipeline synthesizes actionable clinical feedback. Ultimately, our work offers a practical, scalable solution toward bridging the community mental health supervision gap.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28772
PDF
-
SkillEval: Learning Interpretable Ability Profiles of LLMs via Cognitive Diagnosis Models
2026-10-20 18:06:47Berke Arda, Peng Cui, Qiaoyuan Zheng, Rudolf Debelak, Mubashara Akhtar, Mrinmaya SachanAbstract
Current evaluation of LLMs on benchmarks typically aggregates performance over individual examples into a single scalar score. This approach is overly coarse and ignores the diverse scope of skills covered by benchmark datasets. To address this limitation and better exploit benchmark data, we propose SkillEval, a new evaluation framework that performs cross-model, skill-level analysis. The framework integrates advances from two areas: (1) it leverages the meta-cognitive capabilities of LLMs to automatically discover interpretable skills from questions, and (2) it applies cognitive diagnosis models (CDMs) from psychometrics to estimate per-skill abilities of LLMs. We conduct experiments on 3,811 LLMs evaluated on 9,523 items across five benchmarks spanning diverse domains (MATH, BBH, GPQA, MuSR, IFEval). Compared to the traditional average-based evaluation paradigm, SkillEval produces reliable, more calibrated and informative yet concise model profiles that more accurately predict model behavior on unseen items. We also demonstrate the utility of SkillEval in two downstream tasks: (1) LLM routing, which dynamically assigns each query to a capable but efficient model, reducing inference cost by over 60% while achieving improved performance, and (2) Computerized Adaptive Testing (CAT), where an unseen model can be efficiently evaluated using a trained CDM. Furthermore, we show that these skill profiles provide useful insights, for example, how benchmarks share skills and how instruction-tuned models differ from their base models.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28767
PDF
-
Automating ML for Science: Can Frontier Agents Climb Scientific Hills in the Wild?
2026-10-20 18:03:48Ming Zhong, Stacy Li, Nicholas Carlini, Matthew JagielskiAbstract
Frontier coding agents are increasingly framed as end-to-end automated researchers, and ML for science offers a demanding test of that promise: improving a scientific model often requires domain insight beyond routine ML engineering. We introduce Agent4Sci, a long-horizon testbed for measuring how far today's agents can improve the published best models on scientific ML benchmarks, spanning 20 benchmarks across 11 domains. Rather than ranking agents on a leaderboard, Agent4Sci asks what kinds of progress agents can make on these tasks under in-the-wild conditions, and where they fall short. Empirically, today's frontier agents advance the published best on every benchmark, with Claude Opus reaching 10.4% mean relative improvement and no run flagged for academic misconduct. These gains, however, come almost entirely from a bag of ML-engineering tricks rather than research-level insight. Trajectory analysis surfaces three recurring limitations in this setting: novelty miscalibration, resource under-utilization, and rapid collapse into local engineering sweeps. Lightweight plug-in skills for research ideation, experiment management, and scheduled checklists raise Claude Opus from 10.4% to 20.1% and Codex from 4.4% to 8.1%, while yielding more diverse, domain-relevant interventions. Agent4Sci gives the community a way to track and push frontier agents from ML engineering toward domain insight on real-world ML-for-science problems. Code and data are available on GitHub (this https URL) and Hugging Face (this https URL).
Abstract (translated)
URL
https://arxiv.org/abs/2610.28763
PDF
-
A Reverse-Gold Theorem: Positive-Data Identification by Substrate-Bounded Contrastive Learners
Abstract
Gold's theorem shows that no learner can identify an arbitrary language in a superfinite class from positive presentations alone. We prove a positive result of a different shape. Fix an innate \emph{recursive contrast operator} $R$ mapping any string to a finite set of structurally bounded perturbations of it, and a discriminative scorer of bounded capacity trained by InfoNCE-style contrastive estimation \citep{vandenoord2018,macollins2018} on self-generated perturbations. Under a faithfulness condition relating operator, scorer, and target, the trained scorer achieves \emph{PAC conditional identification} from positive data alone: with high probability it ranks all but an $\varepsilon$-fraction of positives above every out-of-language perturbation in their own orbit, with sample complexity polynomial in the capacity, the orbit size, the score bound, and $1/\varepsilon$. The negatives are constructed by the learner, not supplied by an oracle, and no global normalizing constant over strings is ever computed. The misranking bound follows from the geometry of the contrastive loss---per-example loss floors plus a definite loss increment under misranking---yielding an $\varepsilon^{-2}$ rate, together with an explicit approximation gap that quantifies the cost of in-language (``colliding'') perturbations and vanishes for orbit-pure operators. The ranking guarantee lifts to a global membership classifier under a margin condition, and the same operator composed with the learned scorer is a normalization-free Metropolis--Hastings generator. We instantiate the hypotheses for substitutable context-free languages \citep{clarkeyraud2007} and the subregular classes \citep{heinz2010,hayashikamigaito2025}, and state an open conjecture locating the construction in the mildly context-sensitive hierarchy.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28747
PDF
-
NarraSpace: Multimodal Scene-Grounded Evidence for Picture-Description Dementia Classification
2026-10-20 17:55:44Qingming Li, Ruiyan Sun, Satoshi NakamuraAbstract
Picture-description dementia classification uses a shared visual scene, but a transcript-level prediction alone does not identify the scene evidence underlying its inputs. NarraSpace links words and speech times to named picture elements and retrieves utterances for predefined scene events. Narrative-spatial descriptors summarize coverage and temporal organization. A frozen language model contrasts each event's text under examples from Alzheimer's disease and healthy-control training participants, alongside transcript surprisal and coverage measurements. Retained source records link event scores to their source utterances, speech intervals, and lexical matches. On the official 48-participant test set of the Alzheimer's Dementia Recognition through Spontaneous Speech benchmark, a fixed support vector classifier achieves $80.88\%\pm2.56\%$ mean accuracy over fifty seeds; averaged probabilities yield $81.25\%$ accuracy and $91.32\%$ area under the receiver operating characteristic curve. NarraSpace makes the construction of scene-linked classification inputs explicit and inspectable by connecting event-indexed language-model measurements to their source utterances, speech timing, and extraction records. Such source-linked measurements could support clinician review of picture descriptions by presenting quantitative summaries alongside the original speech evidence.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28616
PDF
-
On the Monitorability of Natural Language Autoencoders: Identification vs Discrimination
Abstract
Natural Language Autoencoders (NLAs) are proposed as oversight tools for language models: an Activation Verbaliser (AV) writes an English explanation of an activation vector, and an Activation Reconstructor (AR) maps the explanation back to a vector. Their value for oversight depends on a human monitor being able to tell a true account of an activation from a false one, given the AV's explanation. We formalise NLA training and NLA oversight deployment as two different referential games from Emergent Communication: training is a generative referential game, while oversight is a discriminative, multi-modal, partially-observable referential game played over a many-to-many relation between the monitored language model's activations and the accounts that are true of them. The gap between the two games predicts that AV's explanations will identify which activation they describe without discriminating its true account from closely matched false ones, since training never asks explanations to rule anything out. Using the NLA's own AR as a maximally-sympathetic surrogate monitor model, we experimentally confirm this theoretical insight on one released checkpoint: true accounts beat accounts of other activations in all 320 comparisons, but beat fluent rivals differing in a single claim only 60.2% of the time (standard error 6.7%, chance at 50%, $n = 35$). A meaning-preserving rewrite by an uninvolved model costs only 2.1% of the distance to another activation's account, so the failure is not due to the NLA pair having developed private conventions in the AV's explanations. Our rival accounts test only transcript-level truths, which the monitored language model's conversation settles, so these results bound the easiest case and leave claims about the monitored model's own processing untested. We state conjectures that would make both results consequences of the framework.
Abstract (translated)
URL
https://arxiv.org/abs/2610.18727
PDF
-
Cache Engineering for Agents
2026-10-20 17:32:24Tianyu Fu, Ruoxi Wang, Zihan Min, Zhongqi Ren, Zhihao Li, Tengxuan Liu, Shengxu Mu, Hanling Zhang, Yu WangAbstract
Large Language Model (LLM) agents can handle increasingly complex tasks, navigating long trajectories of observations and actions. As these trajectories grow, key-value (KV) cache I/O and storage become the dominant cost. Trajectories can also exceed the LLM's maximum context length, requiring control over active cache growth. Prior work uses context engineering to reduce cache size by trimming or summarizing history. However, modifying the trajectory prefix invalidates subsequent cached states and triggers costly re-prefilling of retained context. Task accuracy can also suffer when too little context is retained to limit re-prefill costs. We propose cache engineering as a new scheme. It allows agents to directly control their active KV cache without modifying the trajectory or re-prefilling retained context. Specifically, agents can proactively use drop to exclude selected cache from future attention and repos to adjust retained keys' position encodings. We formalize these cache operations and identify history dependence as a serving challenge: identical token sequences can yield different cache states under different cache operation sequences, making prefix matching insufficient for correct cache reuse. Thus, we co-design the AKV serving engine. It provides a stateless interface compatible with existing serving APIs, while internally managing stateful cache reuse, eviction, and restoration. Experiments show that a tool-response rolling-drop cache policy improves average benchmark scores by 1.7 points and reduces average cost by 27.7%, compared with conventional threshold-triggered summarization. With AKV, it achieves up to 1.2x the output throughput and a 1.5x speedup in P95 request latency. Recursive self-improvement of cache policies further reduces cost by 7.1% over expert design. Our code is at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28577
PDF
-
Qwen-Audio-3.1-ASR Technical Report
2026-10-20 17:08:12Chuanmeng Bian, Daren Chen, Peixin Chen, Zhigao Chen, Chong Deng, Zhiyun Fan, Zhifu Gao, Bo Gong, Qing Gu, Jiajun He, Yawei Hu, Yunjie Ji, Wei Ju, Jingbei Li, Xiangang Li, Xu Li, Zengxi Li, Zheng Li, Chengdong Liang, Baiji Liu, Yaming Liu, Ying Liu, Bin Ma, Yiping Peng, Yuezhang Peng, Zhendong Peng, Yu Pu, Yang Shi, Xin Shu, Jian Tang, Biao Tian, Peiyao Wang, Tianzi Wang, Wen Wang, Wupeng Wang, Cheng Wen, Yuzhong Wu, Zijian Xia, Yunchong Xiao, Nan Yang, Jianwei Yu, Jixing Yu, Binbin Zhang, Lei Zhang, Sitong Zhao, Guangdong Zhou, Yuan Zhou, Jianheng ZhuoAbstract
We present Qwen-Audio-3.1-ASR, a family of audio-language models that builds upon its predecesor Qwen-Audio-3.0-ASR with improved speech recognition and expanded capabilities in speaker-diarized transcription and general audio understanding. The family retains support for 30 languages and 16 Chinese dialects, together with instruction-controlled capabilities including historical-context conditioning, hierarchical hotword customization, and native transcription polishing. Speaker-diarized ASR produces structured outputs containing speaker labels, timestamps, and recognized text, maintaining consistent speaker labels across turns and the representation of overlapping speech. The general audio understanding capabilities cover audio-, speech-, and music-captioning, temporal grounding and sound event detection, as well as audio question answering and multiple-choice reasoning. These tasks broaden the application scope from recognizing spoken words to interpreting speakers, acoustic events, and their temporal relations. To accommodate different deployment requirements, the family includes variants for asynchronous file transcription, low-latency streaming, interactive voice input, and non-streaming recognition, with speaker-diarized transcription and general audio understanding available in certain variants. Evaluations span conventional ASR, multilingual and dialectal speech, speaker-diarized transcription, and general audio understanding, demonstrating the capabilities of the Qwen-Audio-3.1-ASR family across transcription and broader audio analysis tasks.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28711
PDF
-
Translation and data ethics
Abstract
This chapter introduces data ethics for translation and discusses a number of ethical issues regarding data. These include issues of copyright, ownership and bias, which apply to all types of data, whereas there are further issues particular to minoritised language communities and those on either side of the digital divide, as data is implicated in power imbalances and dynamics. We close with a discussion of responsibility and regulation, and reflect on the growing attention on data ethics from translation studies researchers.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28710
PDF
-
Present but Hard to Read: Probing Latent Reasoning with Minimal Pairs
Abstract
Probing studies of latent reasoning often read a model's state where the answer begins and take a failed probe as evidence that a value is absent. We test this with a digit copied from the question, in the looped model Ouro-1.4B and the standard transformer Qwen2.5-1.5B. Comparing prompts that differ only in that digit shows it is present at the last prompt token, in a code shared across questions (readout about 0.9). A probe that sees one prompt at a time, the usual setting, barely reads it: 0.36 in Ouro and 0.19 in Qwen even after 5,249 training questions. The digit's change is small rather than crowded: random noise lands on equally busy directions, while a random word in its place moves the state far more and is readable. Since every varied token leaves a trace, a successful probe is ambiguous too. We give a protocol, including a synthetic null for generic mixing, for testing computed values against copies and mixtures of the inputs. In pre-registered studies of two-step problems it finds a small signal for the unwritten intermediate a+b just before Ouro writes the answer's last digit, and its sign follows the operation, though mixing alone produces much of that sign-following part. Holding that value at another problem's from the first loop on does not move the written digit toward the other problem's, while the same patch of the answer does: the value is readable but not used.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28498
PDF
-
OptionWorld: A Benchmark for Diagnosing Failures of LLMs as Long-Term Personal Assistants
2026-10-20 16:38:43Ruoyu Zhang, Zhaocen Liu, Weiyi Tian, Melissa Feng, Shilong LiuAbstract
Long-term assistants must follow unfinished tasks across days as information and circumstances change. Finding information or starting an action, however, does not ensure that the task gets completed. We introduce OptionWorld, a benchmark that traces information acquisition, judgments, actions, and outcomes in simulated daily-life tasks to identify where current LLMs fall short as long-term personal assistants. In runs with archived decision inputs, the key facts remain in the model's last input before the deadline in 35.5% of failed instances. Failure analysis identifies the three most frequent problems: needed information is not obtained or combined, available opportunities to act go unused, and agents impose restrictions that the task does not require. Three follow-up experiments show that supplying missing facts restores task success in some cases; action reminders can restart work while subsequent steps still miss the deadline; and correcting a supplied note that wrongly forbids contact restores permitted inquiry. These findings locate shortcomings in maintaining task state: connecting facts to unfinished work, following through before deadlines, and separating actual restrictions from the agent's assumptions.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28491
PDF
-
Asymmetric Looped Transformers: Fixed Global Representations for an Iterative Decoder
2026-10-20 16:36:33Cheng Liang, Tengfei Zhang, Wenjie DU, Jianyu Wu, Changlian Ma, Shuchen Pu, Ya Zhang, Yanfeng Wang, Zhouhan LinAbstract
Looped Transformers increase effective depth by repeatedly applying shared layers, offering a way to scale computation without increasing parameter count. However, full-stack looping places context construction and iterative state refinement within the same recurrent computation, without explicitly distinguishing their roles. We introduce the Asymmetric Looped Transformer (AsymLoop), which separates single-pass context encoding from recurrent state refinement. A single-pass causal encoder produces layer-specific Global key-value (KV) caches that remain fixed across iterations, while a weight-tied recurrent decoder generates Local KV from its evolving state. Joint attention over both sources under a single softmax allows each update to integrate persistent contextual information with the current intermediate state. For training, we apply a heavy-tail-guided layerwise learning-rate method to the encoder and decoder. Experiments on FineWeb-Edu across four model scales, from 60M to 1B, show consistently lower validation perplexity and higher average downstream accuracy than Full Loop under matched effective depth and training token budgets. At 1B, AsymLoop improves average zero-shot accuracy across seven commonsense reasoning benchmarks from 49.58% to 50.85%. Further experiments extend these benefits to longer training budgets, an OLMo-style backbone, and sparse mixture-of-experts models. These results support asymmetric context encoding and state refinement as an effective design for recurrent language models.
Abstract (translated)
URL
https://arxiv.org/abs/2610.28489
PDF
