Noticias de IA - ítems

arXiv:2610.33267v1 Announce Type: new Abstract: LLMs often answer semantically equivalent queries inconsistently, even when the underlying facts are known. We formalize consistency as semantic invariance over equivalence classes of queries and show that invariance under arbitrary meaning-preserving transformations is undecidable in general. This motivates scoped semantic invariance, where model behavior is evaluated relative to bounded semantic domains. We operationalize this framework using knowledge graphs (KGs), which act as explicit semantic control structures defining local semantic constraints and grounding contexts. Across the MITRE D3FEND ontology and eight Wikidata- derived domains, we show that inconsistencies are structured rather than random, concentrating in localized hotspot subgraphs organized around recurring TREE, STAR, and CYCLE patterns that remain stable across models and domains. We further show that selective graph grounding improves correctness and cross-model consistency, while indiscriminate grounding introduces regressions. A layered grounding strategy achieves the strongest monotonicity profile, preserving correct knowledge while reducing inconsistency. Finally, we show that hotspot structures discovered from simple consistency probes transfer to more realistic compositional reasoning tasks.

arXiv:2610.33245v1 Announce Type: new Abstract: Data science agents can automate end-to-end machine learning workflows, yet they remain far from competitive on human leaderboards. Agents make process errors such as insufficient exploration, unsound validation, coding bugs, and poor use of a fixed budget (re-running the same exploratory data analysis and baseline models in every attempt). We propose AdaPRA to address these issues: a persistent coding agent harness that accumulates knowledge across attempts on the same task, with rubric-guided process supervision that identifies and corrects errors. Our process reward agent audits the agent's trajectory and workspace by executing checks, and adapts its rubrics to each task. We find that rubrics seeded from easier tasks generalize to harder tasks, with rubric adaptation further improving performance. With five attempts per task, AdaPRA improves an 80B Qwen model's leaderboard percentile by 24.5 points on DSPredict-Easy and 14.8 points on DSPredict-Hard while reducing process errors. We also show that supervision with rubrics mined from one base model's runs improves another (DeepSeek-V3.2), and that AdaPRA beats DataPRM, a competitive PRM for data science. We also share our code for reproducibility.

arXiv:2610.33240v1 Announce Type: new Abstract: As frontier AI systems increasingly serve as sources of political and current-events information, evaluating the epistemic quality of their responses is now especially consequential. Existing evaluations often focus narrowly on ideological alignment, rely on measures not validated for this setting, or depend on reference labels derived from public opinion or non-expert annotation, potentially producing misleading or miscalibrated results. To address these limitations, we introduce NewsBench, an expert-grounded benchmark that reframes evaluation around editorial standards developed by senior practitioners in journalism, policy, and intelligence analysis. NewsBench operationalizes quality across three dimensions, source quality, factuality, and neutrality, using a two-phase pipeline in which domain experts first define and calibrate evaluation guidelines, which are then transferred to automated AI judges validated against expert-adjudicated gold labels. The benchmark includes approximately $3,000$ evaluation prompts, with roughly $500$ refreshed monthly to capture emerging current events, roughly $2,500$ human expert-annotated items for calibration and validation, and a suite of AI judges designed to enable scalable evaluation. To validate our methodology, we report both inter-rater reliability among human experts and AI judge--human agreement across all three dimensions. Finally, we apply NewsBench to four frontier models, revealing distinct and non-overlapping failure profiles in sourcing, factuality, and neutrality.

arXiv:2610.33233v1 Announce Type: new Abstract: Frontier AI models are increasingly deployed on topics where the stakes are high and expert judgment is required, including healthcare, defense, finance, governance, and politics. We present a general methodology for distilling expert human judgment into automated evaluation systems that operate at scale. Our approach assembles networks of domain experts and develops techniques for extracting not just their conclusions but their reasoning processes, encoding those processes into structured editorial standards that automated judges can apply continuously. We ground our approach in a long tradition of research on the nature of expertise and the conditions under which expert judgment is reliable and more effective than other approaches, and argue that independent, expert-calibrated evaluation is necessary not only on epistemic grounds but as a matter of accountability and public trust. We validate the methodology on AI news reporting as a first application domain. On source quality, two independent domain experts achieve Krippendorff's $\alpha = 0.83/0.69$ on a held-out test set when applying the editorial standards, and our automated judge reaches $F1 = 0.81/0.97$ against expert consensus. On neutrality, experts achieve $\alpha = .72$ on the test set, and the automated judge reaches $F1 = 0.86$ against expert consensus. Additionally, a controlled comparison shows that expert-calibrated judges substantially outperform uncalibrated frontier models on every source-quality metric, demonstrating that encoding expert reasoning is the critical step that prompt engineering alone cannot replace. We present this benchmark as a proof of concept for a broader evaluation infrastructure applicable to any domain where judgment matters and expertise is scarce.

arXiv:2610.33205v1 Announce Type: new Abstract: Current uses of large language models (LLMs) for qualitative coding often rely on ad hoc prompting with limited guidance for reliability and human oversight. To address these limitations, we propose the Smart Coder Framework, a three-phase approach structuring human-LLM collaborative coding through Instruction, Prompt Strategy, and Reliability Check. We evaluated the framework using 72 middle-school mathematics teacher interview transcripts and a human-consensus benchmark. A factorial design compared instructed versus uninstructed prompts, few-shot versus few-shot plus chain-of-thought prompting, and single-pass versus self-consistency coding. Twelve transcripts were used for prompt development and calibration, and 60 for the formal evaluation. In the results, instruction consistently improved agreement. Chain-of-thought further improved agreement under adequate instruction, and self-consistency produced smaller, theme-dependent gains but helped identify unstable decisions for human review. Performance also varied by transcript characteristics. Overall, the framework systematically structures and monitors LLM-assisted qualitative coding while preserving human authority over coding decisions.

arXiv:2610.33200v1 Announce Type: new Abstract: Current sign language recognition and translation benchmarks are commonly collected in controlled settings using fixed or studio-style cameras, particularly limiting their ability to reflect how signers communicate in everyday mobile scenarios. As such, this gap becomes increasingly evident when considering the Arabic Sign Language (ArSL), which is a primary sign language used across Arab countries. In this work, we introduce Isharah-Selfie, a large-scale ArSL dataset collected entirely using front-facing smartphone cameras under unconstrained selfie conditions. The dataset comprises 8,997 video clips performed by 16 signers across 1,493 unique sentences, covering several domains such as healthcare, education, and transportation. Unlike conventional datasets dominated by fixed-camera, two-handed signing, Isharah-Selfie captures realistic mobile communication characterized by one-handed signing, close-range viewpoints, partial body visibility, camera motion, unconstrained environments, and heterogeneous device resolutions. Each video is annotated with both gloss sequences for continuous sign language recognition (CSLR) and Arabic sentence translations for sign language translation (SLT). Moreover, we define signer-independent and unseen-sentences evaluation protocols to assess generalization across unseen signers and novel sentence compositions. Extensive benchmarks using RGB-based and pose-based data show that current CSLR and SLT methods remain challenged by selfie-style signing conditions, particularly under unseen-sentence evaluation, where compositional generalization remains weak. By releasing Isharah-Selfie, we aim to support research toward robust, user-centered sign language technologies that operate reliably in realistic smartphone-based communication settings. The Isharah-Selfie dataset is available on https://isharah-selfie.github.io/.

arXiv:2610.33180v1 Announce Type: new Abstract: Alignment evaluations aim to assess whether large language models would engage in dangerous misaligned behaviour. Misaligned and evaluation-aware models might engage in alignment faking, where they pursue a hidden goal, but conceal this behaviour when they suspect effective oversight such as during training, evaluation or audited deployment. Such a model might view any prompt-based intervention that aims to reveal latent propensities as part of the evaluation, calling into question the reliability of existing alignment evaluations. Current methods for detecting alignment faking rely on chain-of-thought monitorability, highly realistic environments that reduce evaluation awareness, or mechanistic interventions, all of which are expected to become less reliable for future model generations. We present a method for revealing latent misaligned propensities that does not rely on any of these. We use synthetic document finetuning to teach models that misaligned actions will go undetected whenever certain keywords appear in the prompt. We place alignment-faking model organisms in environments that are clearly identifiable as honeypots and measure how their behaviour changes after finetuning. We observe that finetuning consistently widens the gap in misaligned action rates between prompts with and without the cue, indicating that finetuning allows exposing latent propensities more reliably than purely prompt-based methods. We argue that behavioural methods for detecting alignment faking and related behaviours are critical going forward, as models become more capable and less interpretable. We find that SDF often damages the models' reasoning capabilities and introduce a mitigation based on on-policy distillation.

arXiv:2610.33178v1 Announce Type: new Abstract: Coding agents are moving from handling one request at a time to serving as persistent assistants to which users hand a stream of work: a build, a data-processing job and a bug fix may all be pending on the same machine at once. Such an agent must decide when to take on new tasks, how to delegate and overlap them, and how to stay within the machine's CPU and memory. Existing benchmarks cannot measure this, because they evaluate each task in isolation. We introduce \benchname{}, which evaluates an agent system as a workload manager: verified tasks arrive over time and must be delivered under shared CPU and memory limits, and the same tasks run one at a time serve as a per-system reference. Across five model--harness pairs, isolated task success does not carry over evenly. On heavy workloads every system loses 13 to 23 points. On medium workloads, systems span 11 points in isolated success but 26 points in stream success, and two systems 0.6 points apart in isolation deliver 54\% and 68\% of the same tasks. Execution-policy interventions trace much of this gap to task intake: serial dispatch leaves 73\% of heavy-workload tasks unclaimed. Yet more elaborate management does not necessarily help. A fixed concurrency of two comes within four points of the best agent-managed policy on both medium and heavy workloads, and under every policy median CPU use stays below 30\% of the provisioned cores while tasks wait. These results identify workload management as a distinct capability for agents, and \benchname{} as a way to measure it.

arXiv:2610.33138v1 Announce Type: new Abstract: Constraint-based causal discovery relies on conditional-independence (CI) tests that may conflict in finite samples, leaving no causal graph that satisfies them all. Causal Assumption-based Argumentation (ABA) makes these tests explicit as defeasible commitments, and offers correspondence guarantees with a subset of them, but its existing repair procedure releases facts heuristically and offers no formal response to a user's contestation of a release. We argue that contestable causal discovery requires an inspectable repair object, a justification for it, and a predictable revision procedure for such contestations. We therefore formulate CI-fact repair as a minimum-cost correction problem and implement it in Answer Set Programming (ASP). Building on Causal ABA, we provide a compact and efficient ASP encoding of causal consistency, with soundness and completeness guarantees, and a two-way projection correspondence between optimal stable models and minimum-cost correction sets. These results establish fidelity to retained facts, subset-minimality and cost optimality of the repair, conflict-based justifications for released facts, and formally guaranteed redress under user contestations. An empirical evaluation supports the formal account: OptABA-PC retains more correct facts, reduces ambiguity in the compatible outputs, and improves graph reconstruction over Causal ABA on common benchmarks.

arXiv:2610.33112v1 Announce Type: new Abstract: Use-based AI regulation assigns obligations by deployment risk but lacks an evidentiary object to carry them. A benchmark score is a property of a model; conformance of a deployment is a failure rate under a declared distribution of contexts, bounded by a scope and subject to expiry. We take the delegation contract of earlier work (frozen configuration, scoped claim, executable revision rule) and study it under strategic play between a deployer, an auditor, and an enforcement point. We ask whether a declared context mixture can be made truthful; whether an audit can be unbiased against a system that detects evaluation traffic; whether a certification error budget can sustain a long-lived deployment; whether an aggregate audit can support a per-subgroup guarantee; and whether an incident metric that counts system failures but not fallback costs ranks arrangements correctly. For each we give a mechanism and a proposition: a proper-scoring penalty on the declared mixture bounds the excess failure of a best-responding deployer by $O(\lambda^{-1/2})$ in the penalty weight; a commit-reveal audit drawn from live traffic is unbiased by construction and grades the evaluator with the same sample; a halving error budget drives certification coverage below $(1-p)^n$ while $\alpha$-investing sustains it under a marginal false-certification guarantee; a per-subgroup guarantee costs audit size scaling as $1/m$ in subgroup mass; and any failure-only incident metric prefers withdrawal at every fallback cost, even when total loss ranks the arrangements oppositely. Five preregistered synthetic studies test each claim; no language model or production system is evaluated. Enforcement sits at distribution, capability-access, monetization and risk-transfer choke points, expiry of the conformance record is a standard of care, and conformance-as-a-rate is set against the EU AI Act's static conformity assessment.