Tema

LLMs

2972documentos

225últimos 7 días

Rome Wasn't Built in One Turn: Replay-Grounded Reward Signals for Long-Horizon Freeciv Agents

Training language-model agents to play long-horizon strategy games is hard because deterministic feedback is sparse and often arrives only at the end of a game. In-game scores can offer dense signals, but they do not reliably predict future outcomes: in 8-68% of position pairs across eight strategy games, the higher-scoring position leads to the worse outcome. We term this phenomenon scoreboard traps. In Freeciv, an open-source turn-based empire-building game where several players manage units, cities, and diplomacy over hundreds of turns, we look into better reward signals for gameplay agents. To facilitate this, we build CivHarness, an open system that saves current game states, branches on different decisions, and replays the game to future turns to provide measurable long-horizon feedback from saved positions. We use CivHarness to train CivTelescope, an LLM Judge that predicts which one of two game positions will yield the better outcome. On scoreboard-trap pairs, CivTelescope is 1.3-2.0 times more accurate than off-the-shelf language models, and it generalizes to new game starts, later turns, and unseen rules. We then train a Freeciv gameplay agent with CivTelescope as a dense reward, and call this recipe CivMarshmallow. When used in the middle stage of the game where scoreboard traps are common, CivMarshmallow reaches an average training phase gain of 8.94 points, 77% more than the best baseline, the end-of-game reward. Over the full 120-turn game, CivMarshmallow uses a hybrid reward that starts from the scoreboard or the end-of-game reward and shifts to CivTelescope as the game advances; this hybrid reward scores highest among the rewards we compare.

arXiv AI Papershace 14 días

MEMIG: Self-evolving Memory Construction for Antibody Optimization

Memory has become central to self-evolving LLM agents, which store and retrieve experience from past rollouts to guide future decisions. Existing systems represent this experience as free-form natural language and maintain it by rewriting, which fits tasks whose experience is narratable. However, in scientific discovery and optimization, experience is often a quantity measured by an oracle under a specific context. Rewriting it as text leaves no index for where a record should be, replaces a measurement with a description, and cannot express how interventions interact. To address this, we argue that agent memory for biomolecule optimization should instead be indexable, quantitative, and structured, and we introduce MEMIG. This memory construction framework realizes these three properties. In detail, MEMIG addresses each insight with a domain-grounded identifier and accumulates the quantitative effect measured by a property oracle, so that updates remain local and retrieval becomes an exact lookup. Next, MEMIG connects insights whose interventions are applied together and measures their epistasis by decomposing each evaluated set. Our results show that MEMIG improves performance in most settings across in silico antibody optimization tasks.

arXiv AI Papershace 14 días

Experts Exhibit Structured Redundancy: Structure-Preserving Expert Pruning for Mixture-of-Experts Large Language Models

Sparse MoE mechanism significantly improves the trade-off between model capacity and inference efficiency. However, practical deployment still requires storing all experts, resulting in substantial memory overhead. We show that expert outputs in MoE layers exhibit structured redundancy. Across different models and task categories, the stable rank and entropy effective rank of expert outputs are consistently far below the nominal number of experts, while expert output distributions also exhibit fewer effective modes. Motivated by these observations, we reformulate expert pruning as a structure-preserving expert subset selection problem and efficiently optimize it via a greedy strategy, rather than only independently scoring and ranking individual experts or exhaustively searching over expert combinations. We propose two expert pruning methods under the unified structure-preserving framework: Projection Energy Expert Pruning (PEEP), which preserves the functional subspace by maximizing weighted projection energy, and Distributional Coverage Expert Pruning (DCEP), which preserves the distributional geometry by minimizing weighted coverage error. Both methods incorporate router-derived weights as priors, enabling them to jointly preserve the global structural relationships among experts while accounting for the local importance of individual experts. Experiments on DeepSeek-V2-Lite and Qwen3-30B-A3B demonstrate that, at the expert retention ratios of 75% and 50%, our methods outperform the state-of-the-art baselines on the majority of benchmarks and consistently achieve the best average performance across both general tasks and mathematical reasoning tasks.

arXiv AI Papershace 14 días

TRIDENT and SHAKTI: A Framework and Benchmark for Hindi AI-Text Detection

As Large Language Models (LLMs) become increasingly multilingual, distinguishing human-written from AI-generated text is essential for maintaining information integrity. However, research remains English-centric, with limited benchmarks for low-resource languages such as Hindi. To address this gap, we introduce SHAKTI, a large-scale Hindi benchmark covering four genres, eight LLM families, multiple prompting strategies, and adversarial transformations. We evaluate its quality through human annotation and analyze linguistic, grammatical, and sentiment-based differences between human- and AI-generated texts. Building on these findings, we propose TRIDENT, a round-trip translation-based detection framework with three variants: TRIDENT-S (static threshold), TRIDENT-D (dynamic threshold), and TRIDENT-C (logistic regression using multi-distance RTT features). TRIDENT leverages semantic consistency between original and round-trip translated texts to identify AI-generated content. Extensive experiments show that TRIDENT outperforms conventional and supervised baselines, with TRIDENT-S achieving the best overall performance. Further evaluations across seven languages, adversarial settings, and unseen LLMs demonstrate its multilingual applicability and robustness, while highlighting sensitivity to the translation model.

arXiv AI Papershace 14 días

RTL-Interact: Training EDA-Interactive Models for Multi-Task RTL Design

Large language models offer a promising route to register-transfer-level (RTL) design, but reliable implementations require checking and revision based on electronic design automation (EDA) tools' feedback. Existing RTL-specialized models often lack learned interaction with such EDA tools or rely on separate generation and repair components. We introduce RTL-Interact, a multi-task post-training framework that trains a single RTL-specialized model to generate RTL, construct task-appropriate EDA checks, and revise implementations using tool feedback. The framework combines automatically constructed multi-task demonstrations with staged supervision from chain-of-thought to tool-integrated reasoning. We further introduce Pattern-Aware Policy Optimization (PAPO) for multi-task reinforcement learning. PAPO augments Group Relative Policy Optimization with correctness-gated rewards that favor concise successful checking while rewarding effective revision, and uses task grouping to preserve coverage across objectives. On CVDP, RTL-Interact-32B achieves 48.7% weighted Pass@1 and 68.9% Pass@5, surpassing its teacher model, DeepSeek-V3.2, while RTL-Interact-8B achieves 44.9% weighted Pass@1. Additionally, we show that PAPO can reduce average tool calls by 63.8% while improving Pass@1. Both models also outperform all RTL-specialized baselines on VerilogEval v2 and RTLLM v2, demonstrating the effectiveness of our method.

arXiv AI Papershace 14 días

Beyond Response Scoring: An AI-aware Cognitive Supervision Framework for Automated Mathematical Short Answer Grading

Transformer-based language models have achieved remarkable success in automated short answer grading (ASAG). However, current methods in ASAG struggle to comprehend, reason about and evaluate complex mathematical answers involving symbols and formulas. These score-only methods fail to generalize across diverse responses and skill states associated with identical scores. Moreover, they ignore the potential upward bias introduced by generative Artificial Intelligence (AI) usage in assessing students' performance and knowledge mastery. To address these limitations, we propose AICG (AI-aware Cognitive-supervised Grading), a two-branch multi-task framework. The first branch combines RoBERTa-based regression, ordinal, and classification heads for individual-answer grading. The second branch uses an expert-designed Q-matrix to associate responses with assessed skills and jointly predicts G-DINA-derived skill mastery and learning gain. Through the shared encoder, these skill-level targets provide cognitive supervision for answer grading beyond score labels. AICG further uses a Gen-AI detector, trained through supervised contrastive pretraining followed by linear classification, to estimate assistance probabilities and convert them into reliability weights, attenuating unreliable cognitive evidence during learning assessment. We conduct experiments on 49,402 instructor-graded responses spanning 19 academic semesters. Relative to RoBERTa regression, AICG reduces mean absolute error (MAE) by 50.3% and improves quadratic weighted kappa (QWK) from 0.798 to 0.927. Ablations identify Q-matrix-guided cognitive supervision as the primary source of improvement, while AI-aware weighting further reduces overall MAE by 4.2% over the unweighted cognitive model and consistently improves MAE and QWK across the four most recent post-Gen-AI semesters.

arXiv AI Papershace 14 días

When AI Investigators Cannot Trust Their Tools: Conflict-Aware Evidence Verification for Cyber Threat Investigation

LLM agents increasingly triage security alerts by combining the outputs of several detection tools, implicitly treating those tools as trustworthy. Unlike prior work on conflicts between a model's knowledge and its context, we study conflicts among the tools themselves: what happens when this machine-generated evidence is conflicting or wrong. On the corrected CIC-IDS2017 dataset, four heterogeneous detectors emit labels and confidences that we corrupt under a threat model varying the number of corrupted tools, the persistence of the corruption, and the confidence attached to wrong outputs. Across open-weight LLMs from 3B to 14B parameters, unverified investigators are fragile: one corrupted detector out of four lowers naive accuracy to .50-.72, three lower every unverified investigator to .13 or less, and honest disagreement alone makes them abstain on up to 63% of flows. We propose Conflict-Aware Evidence Verification (CAV), built from established components: an equal-weight vote that ignores reported confidences, an evidence-conflict gate, and an independent nearest-neighbour trust check that tests the leading candidates. With its prompt frozen before testing, CAV-LLM improves on the naive investigator in every corrupted condition (p < 0.001), reaching .84-.88 with three corrupted detectors against at most .08, while calling the LLM on 0.4% of clean traffic. A fixed verification rule is as accurate or better unless most tools are corrupted. When we make the independent channel itself noisy, CAV inherits its errors: it stays ahead when two or more tools are corrupted, but adding the channel as an extra voter is better when at most one is. Verification thus moves trust from the tools to the check, which must be measured and protected in turn.

arXiv AI Papershace 14 días

Researcher-in-the-Loop Evolutionary Discovery of Entanglement-Assisted Quantum Code Families

LLM-guided program evolution can produce mathematical constructions, but validity, generality, and reproducibility require distinct evidence. We develop a researcher-in-the-loop discovery pipeline that combines program evolution, exact evaluation, mathematical generalization, certified nonexistence checks, and formal proof checking. We apply it to entanglement-assisted quantum error-correcting codes (EAQECCs). Starting from tasks that did not specify a family formula, evolved programs produced finite qubit constructions whose structure researchers generalized to the family [[n,1,n-1;n-q-1]]_q for prime powers q3 at n2q+1 and for q=2 at odd n5. Lean checks the algebraic certificate underlying this family. Together with entanglement lifting, the family closes 54 gaps in archived EAQECC tables. The released evidence includes 115 verified code records, ten nonexistence results with checked DRAT proofs, family implementations, and preserved search programs for exact checks and seeded replay. In a controlled cohort of eight paired generation blocks, validation-selected iterative programs achieve success in 46.1% and 31.6% of held-out executions at unseen lengths n=13,15, compared with 19.5% and 3.1% for independently generated programs. In this setting, program evolution and researcher analysis turn finite computational discoveries into a proved family with independently checkable evidence.

arXiv AI Papershace 14 días

TraceSpec: Understanding and Improving Speculative Decoding for LLM Agents

LLM agents execute tasks through repeated model calls over evolving histories of reasoning, actions, and tool feedback. Aggregate speculative-decoding metrics merge these distinct decisions, providing limited guidance on where draft supervision is needed. We present TraceSpec, a framework connecting trajectory evaluation with adaptive draft training. It combines dynamic execution with fixed-prefix replay and introduces TraceSpec-Traj, a corpus of 10,000 training states and 2,500 test states spanning browsing and software engineering. Analysis of five speculative baselines reveals that aggregate acceptance can rise with context length while declining within matched late browsing actions. Workload and phase further distinguish these trends, motivating supervision guided by trajectory conditions. TraceSpec trains on intermediate assistant turns and allocates a fixed prediction budget using state-conditioned estimates of prefix-aware acceptance gaps. Across three target sizes and two temperatures, it improves replay throughput by 12.5% to 22.6% over its DSpark base. Qwen3-4B ablations show that adaptive allocation increases accepted length by 3.3% to 10.0% over fixed full-turn supervision, with additional comparisons substantiating the roles of state conditioning and consecutive-token agreement. The resulting evaluation resources, trajectory findings, and training strategy connect the analysis of agent execution to practical draft improvement.

arXiv AI Papershace 14 días

ImplicitHop: Evaluating knowledge integration across context and memory

The ability to integrate new information and deduce its consequences is central to learning, understanding, and reasoning in large language models (LLMs). Whether information arrives during training or in-context, models must not only store local facts, but propagate their long-range implications globally. To explore information integration in LLMs, we introduce two complementary natural language benchmarks for large-scale evaluation of deductive reasoning as a function of inference depth. The first is a formal graph benchmark enabling controlled evaluation over a wide range of graph topologies and integration tasks; the second is a knowledge composition benchmark grounded in real-world knowledge graphs. We use these benchmarks to compare knowledge integration from information encoded in context versus information encoded in model weights during pretraining, evaluating a set of open models implicitly, without chain of thought. We find that accuracy decays sharply with increased composition length for both pretrained and in-context knowledge. Information presented in-context, however, substantially outperforms deduction over information encoded in weights. These results quantify a fundamental limitation of current models with direct implications for prompt-based control of agent behavior and for continual learning strategies, and they provide a foundation for developing training, prompting, and integration methods that support deeper reasoning.

arXiv AI Papershace 14 días

ROSTrans: Empirically Benchmarking LLMs for ROS1-to-ROS2 Migration

Although large language models have shown strong capabilities in domain-specific code generation, their performance in robotics software development, especially for the Robot Operating System (ROS), remains largely unexplored. ROS code generation presents distinctive challenges, including complex inter-node coordination, middleware configuration, and lifecycle semantics that go far beyond standard coding benchmarks, and many important correctness properties become observable only once migrated nodes are launched and communicate. We present ROSTrans, a benchmark for evaluating LLMs on ROS1-to-ROS2 code migration, a representative and practically important setting for ROS code generation. ROSTrans comprises 108 scaffolded completion tasks drawn from a diverse set of 70+ open-source ROS1 repositories and organized into a hierarchical taxonomy assessing interface-level, behavioral-level and system-level correctness. For every task we establish an executable reference: a verified ROS2 implementation, a runtime test suite that launches the nodes and exercises their topics, services, actions and parameters, and a container ensuring reproducibility. Candidates are scored by injecting only their target files into an otherwise fixed copy of the reference package, so a candidate can neither supply nor weaken the tests it is judged by. Evaluating six frontier LLMs over five stochastic generations per task, we find pass@1 below 10.9%, and that 56.5% of all samples build successfully and then fail at runtime,and 32.0% fail runtime assertions on the migrated interfaces, a class of error no compile-only oracle can detect. Execution-guided repair further improves task success, reaching 49.1% and 47.2% for Claude Opus 4.6 and GPT-5.5 after three rounds, respectively.

arXiv AI Papershace 14 días

LLM Detoxification via Transition-Aware Reward-Guided Representation Sampling

Large Language Models (LLMs) can generate toxic content, motivating effective test-time detoxification. Existing representation steering methods often rely on static or coarse interventions and provide imprecise guidance because they inadequately model transitions between toxic and non-toxic generations. We propose Transition-Aware Reward-Guided Representation Sampling (TARGRS), a test-time framework that learns reward guidance over toxicity transitions in latent representation space. During exploration, progressive steering along the toxicity direction constructs trajectories from non-toxic to toxic states. Toxicity scores along these trajectories turn sparse annotations into dense, listwise supervision for a transition-aware token-level reward model. During inference, TARGRS uses this model to guide coarse-to-fine sampling along the opposite, non-toxic direction: coarse sampling identifies the intervention needed to reach non-toxic regions, and refinement selects a detoxified representation with minimal perturbation. Experiments on 10 LLMs across RealToxicityPrompts, BOLD, and AttaQ show that TARGRS consistently outperforms state-of-the-art baselines, reducing toxicity by up to 69.38% on RealToxicityPrompts. TARGRS runs nearly as fast as the original LLM while preserving its core capabilities with minimal impact. By intervening in the language decoder space, TARGRS also extends to Large Vision-Language Models (LVLMs), reducing toxicity by up to 64.23% across 3 representative LVLMs.

arXiv AI Papershace 14 días

FrontierMat: Benchmarking Language Models at the Frontier of Materials R&D

To accelerate materials discovery, language models must be able to address problems at the research frontier. Yet textbook- and examination-derived benchmarks primarily assess established knowledge, leaving performance on frontier research problems largely untested. We introduce FrontierMat, a benchmark of 330 problems drawn from frontier academic and industrial research in fields including aerospace, semiconductors, embodied intelligence and nuclear fusion. Every problem is validated by leading experimentalists and industry experts in its field. Each problem presents a real study and asks the model to infer a withheld outcome or draw a conclusion from selected experimental evidence. Problems are annotated by functional domain, material system and the scale of the governing mechanism; 130 require cross-scale reasoning. The strongest general-purpose model scores only 53.7%, and on matched problems, cross-scale problems score 8.7 points below single-scale ones. Half of the problems are solved by at most one model, and models fail chiefly by endorsing what is plausible rather than what was observed. A domain-adaptation study with meta-weighted continued pretraining and scale-modular self-distillation raises an open-weight model from 36.7% to 55.4%, matching the strongest proprietary model, with improvements on public materials benchmarks and no decline in aggregate performance on general benchmarks. FrontierMat provides a testbed for measuring progress toward language models that support frontier materials R&D.

arXiv AI Papershace 14 días

ARK: Avoiding Routing Collisions for KV Cache Transfer in Disaggregated LLM Inference

Disaggregated LLM inference separates the prefill and decode phases across GPU pools, generating massive KV-cache transfers. Because these transfers last hundreds of milliseconds to seconds, they behave as mega elephant flows that dominate link bandwidth utilization. In this regime, stateless ECMP can perform poorly: hash collisions may overload one spine link while leaving others idle, stretching transfer times by seconds. Yet this same persistence makes coordination practical. Since these flows are long-lived, even lightweight one-to-all coordination can be amortized over their lifetime. We present ARK, a distributed elephant-flow path reservation mechanism. ARK coordinates senders to choose source ports whose hashes map concurrent flows onto distinct spines, without requiring switch changes or receiver-side packet reordering. Packet-level RDMA simulations show that ARK reduces mean and P95 FCT by up to 27.3% and 34.0% under moderate load, and further reduces mean TTFT by up to 12.9%.

arXiv AI Papershace 14 días

Lineage Is Memory: Parent Inheritance and Feedback Disclosure in Short-Budget LLM-Guided Optimization

Iterative language-model optimizers refine prompts and programs by evaluating a candidate, showing a proposer model a sample of failing cases, and keeping revisions that improve performance. Recent work attributes the limits of this loop to narrow single-round reflection, introducing full-set error clustering, targeted minibatches, or larger proposer models so the optimizer sees more failure modes per step. Yet every existing evaluation tests these mechanisms exclusively inside lineage-preserving search, where each revision edits a surviving parent, leaving the contribution of parent inheritance confounded with the breadth of per-round feedback. Across 1,750 pre-registered runs spanning nine prompt and program synthesis benchmarks, we separate parent inheritance from feedback disclosure under matched evaluation budgets. We find that greedy parent inheritance drives the vast majority of optimization progress, outperforming seed-reset resampling even when seed-reset proposals receive a clustered summary of all failures in every round or use a larger proposer model. While full-set error clustering and diverse failure selection improve single-step rewrites from the seed and accelerate the opening rounds of search, sequential parent inheritance rapidly closes this gap as solved error modes are retained in the candidate and filtered from subsequent failure lists. Moreover, decomposing token expenditure in GEPA reveals that proposer reflection accounts for only a small fraction of total cost whereas validation and minibatch gating consume nearly the entire budget, showing that greedy parent inheritance over raw failures captures the gains of complex reflective loops at a fraction of their compute.

arXiv AI Papershace 14 días

Counterfactual Pairwise Augmentation Improves Small Pointwise LLM Judges

Agentic systems increasingly gate individual steps with a pointwise judge that scores a response against a criterion. Small language models are attractive for this role because they can be served and specialized efficiently. Such a judge must locate the criterion's decision boundary, but pointwise labels under-determine that boundary in fine-tuning: labeled responses to different queries differ along many dimensions at once, so they do not reveal which features decide the label. We introduce CPA (Counterfactual Pairwise Augmentation), which makes the deciding difference explicit in training. For each pointwise item, CPA synthesizes and audits a counterfactual response to the same query intended to carry the opposite label. It frames the accepted pair as a pairwise task asking which of the two responses satisfies the criterion. The judge is fine-tuned jointly on both pointwise and pairwise datasets and deployed in pointwise mode only. A second-order analysis shows when this should outperform the prevailing strategy of merging separately trained pointwise and pairwise experts: merging moves every parameter direction toward the pairwise expert by the same fraction, whereas joint training can hold the pointwise optimum wherever the pairwise loss is flat and absorb its curvature where it is sharp. This is the regime that CPA's same-source, counterfactually minimal pairwise view is designed to produce. On separate Qwen3-8B LoRA adapters for six binary criteria, CPA improves criterion-averaged pointwise macro-F1 by 2.5 points over pointwise-only training and outperforms model merging on all six criteria at every tested merge weight. It also surpasses the strongest zero-shot frontier judge we evaluate, Gemini 3.6 Flash, by 11.8 points.

arXiv AI Papershace 14 días

HyperRepair: Recovering Forgotten General Capabilities of Fine-Tuned Language Models

Fine-tuning improves a language model's performance in a target domain but can weaken its general capabilities. Existing methods often face a trade-off between general capabilities and domain specialization, making it difficult to preserve both. We introduce HyperRepair, a framework for recovering general capabilities while retaining domain specialization at only inference cost. To make recovery generalizable across models sharing a pretrained backbone, HyperRepair learns a hypernetwork from repairs trained on source models. The hypernetwork uses activation fingerprints that summarize changes from the pretrained backbone to generate a tailored Repairing Side Net (RSN), whose prompt-conditioned gate applies corrections selectively. This enables recovery on models from unseen adaptation domains without target-domain examples or additional optimization. We evaluate HyperRpair across two model families, 12 LoRA adaptation datasets, and 15 general benchmarks. On the default split, HyperRepair achieves 93.4% of the base models' average general score while retaining 99.5% of the fine-tuned models' average domain score.

arXiv AI Papershace 14 días

TalentRL: Strategy-Guided Reinforcement Learning for Hierarchical Skill Routing

Natural-language skills enable large language model (LLM) agents to reuse procedures across tasks, yet most systems retrieve individual skills from a flat SkillBank without explicitly modeling the high-level strategy that governs their selection and use. We introduce TalentRL, a strategy-guided reinforcement learning framework for hierarchical routing over procedural memory. Coarse-grained talents represent recurring strategies for skill use, while fine-grained skills provide executable procedures. For each task, the policy model selects a talent that guides execution and softly prioritizes its associated skills during similarity-based retrieval. However, since task outcomes depend jointly on talent selection, retrieved skill context, and downstream execution, terminal rewards alone cannot isolate the contribution of the routing decision. TalentRL therefore trains talent routing with reward contrasts between rollouts of the same task under matched interventions on the talent-conditioned context, and optimizes downstream execution with group-relative advantages. Across seven open-domain and multi-hop question-answering benchmarks, talent-aware supervised fine-tuning improves average success rate by 7.53%, 6.71%, and 9.21% over the corresponding zero-shot baselines for Qwen2.5-7B, Qwen3-8B, and Llama-3.1-8B, respectively. Reinforcement learning further improves these SFT checkpoints by 7.58% to 9.32%, and TalentRL achieves the highest average success rate among all compared methods on all three backbones. On the agentic benchmarks ALFWorld and WebShop, TalentRL achieves the highest ALFWorld success rate (99.3%) and the highest WebShop score (90.7) among all compared methods.

arXiv AI Papershace 14 días

When Disclosure Is Compulsory: Evaluating Ride-Hailing Rebalancing Policies with Self-Interested LLM Driver Agents

The EU Platform Work Directive, to be transposed by December 2026, requires ride-hailing platforms to state in plain language the main parameters of their automated decisions and how a worker's behaviour affects them. Such rules are exactly what a large language model (LLM) can read, so drivers will be able to let one decide for them. We compare two main families of rebalancing policy once their rules are public: a reactive policy (R) that sends idle drivers to cover requests just lost, and an anticipatory policy (A) that plans moves on a demand forecast. Each driver is an LLM agent maximizing its own net income, given no rules (N) or the rules at the directive's level of detail (L). We test two LLMs, Claude Haiku 5.5 and Jev, which understand the rules equally well, in a three-zone model and in Manhattan. With the rules disclosed, drivers follow the platform more closely: in Manhattan, the share of relocation offers they decline falls from about four in five to one in four under both policies. Yet the effects differ by policy. For the platform, disclosure costs R 3.7% of its profit in Manhattan, as R now pays for moves drivers used to make for free, while A's profit is unchanged; the two policies, which tie on profit under full compliance, end 6.1% apart. For drivers, disclosure raises income in the small model under both policies and LLMs; for users, it raises the service rate there by 10 to 18 percentage points. In Manhattan, self-interested drivers also make A a win-win-win against full compliance, raising driver income, platform profit and service rate by 10 to 12% each. Rebalancing policies that perform equally well under full compliance can thus show opposite results under mandatory disclosure. The study offers a framework to evaluate such policies under the new requirement and is, to our knowledge, the first to do so with LLM agents.

arXiv AI Papershace 14 días

When Should AI Agents Ask? Clarification Requests Under Limited Human Attention

AI agents increasingly carry out delegated tasks in parallel, with one person overseeing several of them at once. In practice, an agent often lacks information its task requires, such as an unstated preference, constraint, or specification. It can ask the human to clarify, or it can search for the information itself, for example in past records, which takes time and effort and may fail. In this paper, we study how to coordinate these decisions across several agents that share one human with limited attention. We build a stylized model in which each agent can search, ask, or continue the parts of its task that do not depend on the missing information, and we formulate the problem as a dynamic program. We first show that whether an agent should ask depends on more than its own uncertainty: an agent with other work to do can be better off asking earlier, and which question the human should answer first depends on all waiting questions, so no ranking computed agent by agent is optimal. We then propose the Pairwise Coordination Policy (PCP), which uses exact one- and two-task problems to estimate how pairs of tasks delay each other when they share the human. Theoretically, PCP is exactly optimal when at most two requirements remain unresolved and costs at most twice the optimum when every unresolved requirement must be resolved by the human. We evaluate PCP on 864 synthetic cases and in a case study with LLM agents. On the synthetic benchmark, PCP's mean gap to optimal cost is 0.07%, and it reduces cost by an average of 18-31% relative to three standard rules for when to ask. In the LLM case study, agents completed every task correctly, and PCP achieved the lowest mean cost in all tested settings.