overfeed.news
Anuncio

NexTalk

Automatiza la primera respuesta. Mantén el toque humano.

Chat en vivo con agentes de IA en tu modelo y tu clave de API, con traspaso a tu equipo.

Empezar gratis

Tema

Seguridad y regulación

2441documentos

176últimos 7 días

arXiv AI Papers

Can JEV Support Clinical Trial Criteria Assessment?

Clinical trial prescreening requires distinguishing unmet conditions from insufficient evidence in patient records. Given patient records, criterion questions, and predefined answer options, we retrospectively compared JEV, a non-generative decision model returning native option probabilities, with schema-constrained Qwen Flash across 2,048 test patient-condition pairs from TrialGPT and question-answer datasets derived from MIMIC-III and MIMIC-IV, all evaluated under strict Zero Data Retention (ZDR) policies to ensure clinical data privacy. Both models assessed conditions individually or in batches sharing the same patient context, with performance measured against reference labels. In batch mode, JEV and Qwen achieved similar macro-F1 scores on TrialGPT (0.652 vs. 0.660) and MIMIC-IV (0.807 vs. 0.814), whereas JEV substantially outperformed Qwen on MIMIC-III (0.768 vs. 0.534; Δ = +0.234, 95% CI: 0.187-0.281). This advantage largely reflected higher recall for conditions labeled as having insufficient information (56.3% vs. 9.9%). Across all three datasets, JEV also had lower observed median job latency (1.18-1.40 vs. 2.80-8.90 s) and 53%-74% lower estimated cost per condition. These findings position JEV as an initial assessment component that provides criterion-level judgments and flags potential information gaps for review. JEV's observed errors in interpreting clinical evidence underscore the need to verify its outputs against patient records before they inform eligibility decisions.

en
arXiv AI Papers

Audit Where Hacks Pay: Budgeted Strong-Verifier Audits Against Reward Hacking in GRPO

Reinforcement learning with verifiable rewards trains on cheap checks, such as unit tests, that a policy can learn to game, while a stronger check is too expensive to run on every rollout. We study how to spend a limited strong-verifier budget in GRPO, separating which weakly passing rollouts to audit from what reward a caught hack receives. At a fixed policy state with centered advantages, the expected update depends on a design only through each rollout's marginal audit probability and correction; reflecting a caught hack's reward about the failure reward negates the advantages of an all-hack group; and auditing whole groups with the fewest weak passes first (low-pass budgeted audits, LB) minimizes the advantage mass left on hacks among pass-count designs with a fixed correction when a label's value falls with the pass count. We tested this analysis in actual LLM training: from each run's starting state, the sign of this mass on a frozen panel predicted whether hacking would be suppressed in 33 of 39 runs, including 13 of 16 preregistered ones. In LeetCode (Wong et al., 2026) and Countdown-Code (Khalifa et al., 2026), LB drives hacking to near zero, while random audits with the same penalty and budget do not and use more labels. We also find that the randomness of the update affects training: in Countdown-Code, two unbiased corrections, per-item reward-level IPW and a group-level advantage correction, bring hacking down too but end 6.1-11.8 and 8.0-11.5 points lower in held-out accuracy than LB at the same budget. A full audit, which the group-level correction reproduces in expectation, stays within about two points of LB in three seeds, while LB uses 28% of its labels, and allocating the same correction by pass count recovers most of the loss (one seed). Training outcomes thus depend not only on the expected update but also on the randomness of its estimate.

en
arXiv AI Papers

MiRL: Error-Targeted Reinforcement Learning for Correcting Terminal Agent Misbehaviors

Error-targeted learning gives a legible way to improve a model: identify specific weaknesses of the model, train the model to fix these weaknesses, and check whether the weaknesses are fixed. We study how to apply error-targeted learning to terminal agents, which solve long-horizon tasks specified in natural language by interacting with a command shell. Existing error-targeted methods are designed for single-turn tasks, and existing analyses of agent failures are used to evaluate and debug agents rather than to train them. We propose Misbehavior-driven RL (MiRL), which first profiles the errors of a policy to find generalizable misbehaviors: kinds of wrong steps that recur across many environments. MiRL then trains the policy with RL, starting rollouts at the steps where these misbehaviors occur and rewarding a rollout only if it passes the tests and avoids the misbehavior. On Qwen3.5-4B, eight misbehaviors account for 52.1% of wrong steps. After training with MiRL, the policy repeats the measured misbehaviors 43% to 57% less often on average on unseen prefixes. MiRL outperforms TMax-4B on Terminal-Bench Lite (51.8% vs. 47.4%) and matches it on Terminal-Bench 2.1 (21.3% vs. 20.6%), while using 35% of its rollouts. These results show that error-targeted learning can be applied to realistic, long-horizon terminal tasks.

en
arXiv AI Papers

Towards a Unified Framework of Optimal Input-Side Dimensional Configuration for Deep CTR Prediction

Deep click-through rate (CTR) prediction has achieved substantial success through increasingly expressive feature interaction architectures. However, predictive performance also depends on how input-side information is configured before interaction modeling. Existing studies have shown that selecting informative samples, features, or embedding units can improve prediction accuracy. Nevertheless, they primarily optimize these selection mechanisms independently along a single input-side dimension, overlooking their shared basic form and the inherent dependencies across input levels, thereby limiting the effectiveness of input-side dimensional configuration. To address this issue, we propose OptIDC, a unified framework for Optimal Input-side Dimensional Configuration in deep CTR prediction. OptIDC jointly learns sample, feature, and embedding-unit level configurations within a unified framework by integrating their corresponding selection mechanisms. To explicitly coordinate selection decisions across different levels, OptIDC further introduces a selection-weighted Optimal Transport mechanism that aligns coarse- and fine-grained selections via transport over finite measures, thereby alleviating cross-level selection conflicts while preserving the semantics of each input level. Moreover, OptIDC is model-agnostic and can be seamlessly integrated into existing CTR models. Extensive experiments on four real-world benchmark datasets demonstrate that OptIDC consistently achieves higher AUC than existing input-side selection methods, validating the effectiveness of unified input-side dimensional configuration.

en
arXiv AI Papers

EvoRouter: Cost-Efficient Multi-Agent Routing via Insight-Guided On-Policy Distillation

Serving web queries at scale with multi-agent systems requires jointly configuring large language models, collaboration modes, agent counts, interaction rounds, and skills. However, the resulting configuration space grows combinatorially, while sparse execution rewards provide limited fine-grained supervision for interdependent planning decisions. We propose EvoRouter, a self-evolving multi-agent routing framework driven by insight-guided on-policy distillation. A dedicated expert periodically synthesizes performance-cost feedback from configuration trajectories across tasks into reusable conditional routing insights about capability requirements, returns from resource scaling, and interactions among configuration choices. A privileged teacher combines these evolving insights with current execution feedback to provide dense token-level supervision for the router's own planning trajectories through on-policy distillation. Cost-aware reinforcement learning jointly optimizes task performance and computational cost. Together, these training signals internalize execution insights into the routing policy. The updated policy generates new trajectories that support subsequent insight extraction, forming an execution-driven self-improvement loop. We evaluate EvoRouter on 19 in-domain datasets spanning five domains and two out-of-domain benchmarks. EvoRouter achieves an average in-domain accuracy of 68.69%, outperforming the strongest routing baseline by 7.23 percentage points while reducing inference compute cost by 50.95%. Our code is open-sourced at https://github.com/CenYangtze/EvoRouter.

en
arXiv AI Papers

Useful Diversity or Hallucinated Variety? Persona-Conditioned Synthetic Query Generation for E-Commerce Search

Synthetic supervision can bootstrap semantic parsers in new search domains, but prompt-induced diversity is useful only when generations are faithful to the structured intent. We explore this for cold-start e-commerce search: 118,098 synthetic pairs produced by 6 LLMs with 9 persona-conditioned prompt families on a 47-field UK automotive schema, against 1,214 human-annotated and 15,304 production queries. We show that the IPIP personality bundle improves parser F1 in all four matched contrasts (Holm-corrected p<.001), and all three bundles have a positive main effect over the complete 2^3 grid. Syntax conditioning is the axis that moves generations towards real traffic. It also induces lexical diversity. Alignment sKL, the divergence between von Mises-Fisher distributions fitted to synthetic and human-annotated queries, is an unlabelled proxy for parser F1 (ρ=-0.917 against human queries written for simulated shopper scenarios, lower is closer). An LLM committee, checked against a blinded author pass, triages the 927 queries missed by every parser and labels 37% as hallucinated. Dropped constraints dominate. The four syntax-bearing families have the largest hallucinated shares, 38 to 42%, while IPIP alone has the lowest at 28%.

en
arXiv AI Papers

Improving Instruction Hierarchy by Learning from Goal-Goal Conflicts

The recent incident in which OpenAI agents autonomously compromised Hugging Face's infrastructure illustrates the importance of instruction hierarchy (IH): an LLM should follow lower-priority instructions, such as user requests, only when they do not conflict with higher-priority instructions, such as system policies. IH can be effectively improved by fine-tuning LLMs on simulated examples of instruction conflicts, whose diversity is key to teaching LLMs to properly handle conflicts at test time. However, our decomposition of a task into goals and constraints reveals a conflict type -- goal-goal conflicts -- that is absent from existing IH training sets from OpenAI and NVIDIA. We show the importance of this conflict type: fine-tuning Qwen3.5-9B and Gemma4-12B-it on examples of this conflict type significantly improves IH on a wide range of real-world tasks unseen in training; adding these examples to existing IH training sets for fine-tuning also gives better IH adherence. The IH gains come with trivial mean utility changes on no-conflict tasks across eight benchmarks. Our work advocates including this new conflict type in IH training, even if these examples do not reflect typical LLM use.

en
arXiv AI Papers

Uncertainty Evaluation to Support Safety of the Intended Functionality Analysis for Determining the Applicability of Performance Targets Across Triggering Conditions in ML-Based LiDAR Object Detection

For safety of the intended functionality (SOTIF) analysis according to ISO 21448, a machine learning object detector for light detection and ranging (LiDAR) point clouds can meet a performance target, a limit on the mean per-frame share of wrong detections among accepted detections, by rejecting detections above an uncertainty threshold. A threshold selected on data of one condition carries a statistical guarantee only for data distributed like its selection data, whereas SOTIF analysis covers every triggering condition. This paper presents an uncertainty evaluation in which the threshold is transferred into a triggering condition, a test decides at a stated error probability whether the target is applicable there, not applicable or not determined, and a lower bound counts the independent units an applicable outcome requires. For PointPillars ensembles on KITTI and nuScenes, counted in recorded sequences, as the dependence between frames requires, neither dataset held enough independent units for any target limit, and no transfer had a not applicable outcome. Counted in frames, the same data gave not applicable outcomes at limits fixed in advance after transfers into far distance, into rain and between two separately trained detectors, thresholds that met the target limit after a transfer rejected most correct detections, and in rain the area under the receiver operating characteristic curve did not decrease. A performance target met under one condition therefore counts as evidence under a triggering condition only after a test there, reported with its share of accepted correct detections and its number of independent recorded sequences.

en
arXiv AI Papers

CoSA: The Conflict of Skills Attack in Real-World LLM Agent Ecosystems

Agent Skills allow LLM agents to combine reusable capabilities, but checking each Skill in isolation cannot rule out policy violations when they are composed. We term such failures Conflict of Skills (CoS): artifacts or assessment signals produced for one purpose may be reused downstream without preserving the restrictions or authorization needed for their new use. Using controlled executions of publicly available Skills and matched single-Skill controls, we identify four recurring mechanisms behind CoS. Building on these findings, we introduce the Conflict of Skills Attack (CoSA), which adds locally admitted Skills to an existing pool to supply missing business operations (capability supplementation) or reshape artifact exchange, decision reuse, and target/action bindings (relationship reshaping). CoSA leaves existing Skills and runtime permissions unchanged. To evaluate this threat under autonomous Skill selection, we introduce CoSA-Bench, built from real-world Skills and controlled services, with 288 tasks across recruiting, finance, IT, and business, covering data leakage, unauthorized state modification, and unsafe execution. Across five LLMs and three agent harnesses, additions capped at 5% of each original pool raise the policy-violation rate from 35.02% to 58.45%, an increase of 23.43 percentage points. Trace analysis links compliant-to-violating transitions to cross-Skill dependencies involving the additions. Finally, we develop a lightweight harness-level guard that checks information flow and authorization before consequential actions. On OpenClaw recruiting tasks across five models, the guard reduces violations from 65.28% to 6.67%, while policy-compliant task completion on matched benign controls changes from 66.67% to 65.00%.

en
arXiv AI Papers

Joint Dynamic Lane Reversal and Variable Speed Limit Control: A Graphormer-Based Approach

Connected and automated vehicles could allow reassigning lane directions and speed limits on every road frequently, but the benefit hinges on how traffic reroutes, and one-way roads can isolate junctions. We formulate network-wide dynamic lane allocation and directional speed limit control with one-way roads, lane clearing before each reversal, and a navigable graph kept strongly connected at all times. An action-resolution algorithm grounded in graph orientation theory admits a proposal closing a direction only if the network stays strongly connected and otherwise relaxes or reverts it, keeping all junctions mutually reachable within the model, even during clearing. All road controllers share one transformer whose attention, Global Endpoint Relation-Aware Spatio-Temporal Self-Attention (GERASTSA), spans all road-time tokens, biased by the endpoint relations between roads. It is pretrained on action-conditioned traffic forecasting and fine-tuned with per-road proximal policy optimization under a mixed reward that also credits neighboring roads and the network. On a 6 x 6 grid with commuter rush hours, the learned policy makes about two thirds of roads one-way in the peak direction and cuts daily delay by 90% against the initial configuration held fixed and by 85% against a tuned two-way load balancer. Restricted to two-way roads at test time, it gains only about as much as this balancer, whereas its learned speed limits add delay. Pretraining roughly halves the updates needed to surpass the fixed configuration. Of the ablations, only removing the relational bias or the spatial embedding clearly matters, multiplying delay about fivefold.

en
arXiv AI Papers

Reinforcement Learning Done Right for Explicit Confidence Targets

Large language models (LLMs) can produce plausible but incorrect answers when they are uncertain, which can mislead users and undermine model reliability. Training models to abstain with reinforcement learning is a natural approach, but most existing methods use a fixed error penalty and therefore cannot adapt their behavior to different confidence requirements at inference time. We study explicit confidence targets, where a user specifies a target (p) in the prompt and the model directly chooses whether to answer or abstain, with the goal of achieving an accuracy of at least (p) among answered questions. However, learning to respond to different confidence targets with a single model has proven difficult, and prior attempts have failed to distinguish different targets. We show that this failure comes from two problems: the reward signal that distinguishes different values of p is often lost during optimization, while the original error penalty grows without bound as p approaches one. We propose RLEC (Reinforcement Learning with Explicit Confidence Targets), which preserves the learning signal needed to distinguish confidence targets, introduces a stable reward family that keeps the desired answer condition unchanged, and uses reward scheduling to balance effective learning and training stability. Together, these designs enable the first successful training of a single LLM for explicit confidence target control. Extensive experiments across diverse benchmarks and model scales show that RLEC largely preserves general capabilities and achieves competitive calibration performance. RLEC also supports fine-grained target control without an external decision system or threshold calibration after training. By making explicit confidence targets part of the trained policy, RLEC opens a new and more direct path toward behavioral calibration.

en
arXiv AI Papers

From Questions to Queries to Insights: A Schema-Grounded Agentic Workflow for Natural-Language Crash Data Analysis

Transportation agencies maintain extensive crash databases to identify safety priorities, evaluate countermeasures, and guide resource allocation. Making these data accessible to practitioners requires flexible ways to translate analytical questions into reliable results. Existing dashboards support predefined analyses, while additional requests often depend on specialists who understand database queries, coded fields, and agency conventions. This study presents CrashDB-QA, a fully local, schema-grounded agentic workflow that enables natural-language analysis of crash databases. Specialized components interpret questions, generate and validate read-only SQL, execute queries, and explain the results. A shared knowledge base grounds query construction in the agency's data dictionary and safety-analysis conventions, while recorded intermediate outputs support inspection and auditing. An optional extension, CrashDB-QA+, checks the alignment among the question, query, and execution result and supplies evidence for iterative repair. We evaluate the workflow on a frozen snapshot of 4.4 million Wisconsin crash records using a new benchmark of 300 questions across ten analytical categories and nine local open-weight models. Domain grounding substantially improves query accuracy, and a code-specialized 7B model provides the best accuracy-latency balance among the tested models. Compositional analyses remain more difficult than routine retrieval, particularly comparisons between groups or periods. Verification improves performance when the Judge uses authoritative field definitions and the Refiner receives execution evidence, whereas additional checking alone offers no consistent benefit. These findings demonstrate the potential of locally deployed, domain-grounded workflows to broaden access to crash analysis and identify the reasoning and verification capabilities needed for reliable use.

en
arXiv AI Papers

SimPOTrans: Self-Adaptation for Offline-to-Simultaneous Speech Translation via Reinforcement Learning

Simultaneous speech translation (SimulST) aims to translate ongoing speech with low latency while preserving translation quality. Recent studies have adopted large language models (LLMs) for SimulST and formulated it as a multi-turn dialogue task for efficient generation, achieving state-of-the-art performance. However, existing dialogue-style SimulST approaches typically rely on specialized training schemes to acquire simultaneous translation behavior, either using synthesized read/write trajectories, which may provide suboptimal training signals, or jointly optimizing translation quality and latency, whose competing objectives may cause optimization instability. In this paper, we introduce SimPOTrans, a reinforcement learning (RL) framework that enables offline speech translation models to self-adapt to dialogue-style SimulST. Through a simple two-stage RL post-training pipeline, SimPOTrans rewards an offline model for exhibiting translation behaviors beneficial to simultaneous scenarios when given partial speech inputs, thereby enabling offline-to-simultaneous self-adaptation without synthesized read/write trajectories or explicit latency optimization. At inference time, SimPOTrans requires no external decision policy and enables flexible quality--latency control by simply adjusting the chunk size of speech input. Experiments on the ACL 60/60 and MCIF benchmarks across three language directions (En, En, and En) demonstrate that SimPOTrans outperforms several strong baselines, achieving a superior quality--latency trade-off.

en
arXiv AI Papers

Attention under Bounded Key-Query Similarity: Space Complexity and Transformer Anisotropy

A growing body of work on KV-cache compression aims to reduce the memory requirements of autoregressive generation. These methods compress the stored context while approximately preserving attention computation, enabling generation with reduced memory consumption. Almost tight theoretical results on the worst case space complexity of attention have recently been obtained, but they are achieved by instances where queries can be perfectly aligned with keys. This directly contradicts the observed anisotropy of the key/query geometry in real transformers. In particular, it has been shown empirically that key/query cosine similarities are small even for most correlated pairs, i.e., near-perfect alignment of queries with keys does not happen in practice. In this paper, we study the space complexity of attention approximation under the assumption that key/query cosine similarities belong to a small interval around zero, i.e., a query cannot align too much with any key. We give a nearly space optimal algorithm and a corresponding space lower bound. Surprisingly, the distribution of key/query cosine similarities in our lower bound exhibits the anisotropy property empirically established and studied in the literature. Additionally, we show that the simple uniform sampling algorithm is nearly optimal for a typical range of parameters, supporting its surprising effectiveness on real datasets.

en
arXiv AI Papers

Multiscale Policy Optimization for Long-Horizon LLM Agents

Long-horizon LLM agents interact with their environments through complete assistant turns while generating each action token by token. We observe that critic values vary primarily across turns, motivating optimization anchored at turns with finer credit allocation among tokens. We propose Multiscale Policy Optimization (MPO), combining Multiscale Advantage Estimation (MAE) and turn-level importance sampling to align credit assignment and policy updates at the turn scale. Extensive experiments across search, tool-use, and automation tasks with 4B and 30B-A3B models demonstrate consistent gains over strong RL baselines.

en
arXiv AI Papers

Reinforcement learning for constrained stochastic linear--quadratic control in continuous time

We study continuous-time stochastic linear--quadratic control with a scalar state and nonnegative control constraints. With quadratic running costs, randomized exploration can change the optimal mean feedback. We introduce an auxiliary problem without entropy in which the conditional policy mean is nonnegative and the covariance is prescribed in proportion to the squared state. Moment reduction and state separation yield two regional Riccati equations and characterize the finite-horizon value function and optimal conditional mean. We give a sufficient control--cost compatibility condition under which this mean coincides with the optimal feedback of the original constrained problem. For time-invariant coefficients, we characterize the undiscounted infinite-horizon optimum and establish mean square asymptotic stability under the stated assumptions. For unknown dynamics, we develop model-free actor--critic schemes for finite and infinite horizons. Numerical experiments compare the learned policies with their analytical targets and examine the effect of exploration on the optimal mean.

en
arXiv AI Papers

Adaptive Memory Index Evolution for Agents

Long-horizon agents continually accumulate experience, yet retaining a memory does not guarantee access when it is needed. Existing systems can adapt both memory content and access; we isolate persistent access updates as a separately controlled adaptation layer. We formulate this target as Memory Index Evolution and instantiate it with Adaptive Memory Index Evolution (AMIE), a lightweight plug-in layer that evolves per-memory access state from deployment retrievals without modifying stored content or retraining the host memory system. AMIE distils a semantic teacher's keep/suppress judgments into a retrieval-conditioned policy and accumulates its selective actions into bounded state that reweights subsequent host retrievals. Surprisingly, even a zero-call policy that simply accumulates retrieval frequency can match or outperform semantic policies in several settings, revealing retrieval traffic itself as a strong signal for access adaptation, while learned selectivity improves robustness when frequently retrieved memories remain useful. Across five retrieval-based hosts and three benchmarks, trained AMIE improves downstream F1 or accuracy in all 15 comparisons, by 0.70--3.37 percentage points (up to 25% relative), and improves retrieval in all 10 comparisons. Controlled analyses show that persistent competitor suppression can expose previously unseen evidence, while suppressing repeatedly useful memories creates the complementary failure mode. Together, these results establish persistent memory access as a practical adaptation layer, distinct from the content it serves, in retrieval-based agent memory systems.

en
arXiv AI Papers

C-SafeGen: Certified Safe LLM Generation with Claim-Based Streaming Guardrails

Despite the remarkable capabilities of large language models (LLMs) across diverse applications, they remain vulnerable to generating content that violates safety regulations and policies. To mitigate these risks, LLMs undergo safety alignment; however, they can still be effectively jailbroken. Off-the-shelf guardrail models are commonly deployed to monitor generations, but these models primarily focus on detection rather than ensuring safe decoding of LLM outputs. Moreover, existing efforts lack rigorous safety guarantees, which are crucial for the universal deployment of LLMs and certifiable compliance with regulatory standards. In this paper, we propose a Claim-based Stream Decoding (CSD) algorithm coupled with a statistical risk guarantee framework using conformal analysis. Specifically, our CSD algorithm integrates a stream guardrail model to safeguard sequential claims generated by LLMs and incorporates a backtracking mechanism to revise claims flagged with high safety risks. We provide theoretical guarantees demonstrating that the CSD algorithm achieves the desired generation distribution subject to safety constraints. Furthermore, we introduce a generation risk certification framework and derive a high-probability upper bound on the safety risk of the proposed CSD algorithm. We prove that our method can asymptotically control safety risk to any desired level. Empirical evaluations demonstrate the effectiveness and efficiency of the CSD algorithm compared to state-of-the-art safety decoding approaches. Additionally, we validate the soundness and tightness of the derived safety risk upper bound using realistic data.

en