Sección

Investigación

Papers y preprints de los feeds de investigación que seguimos.

6415documentos

Profiling a Frontier Agent with EBench: From Direct Control to Harness with Policy

Frontier multimodal large language models (MLLMs) can control robots without robot-specific training, but how their capabilities compare and interact with post-trained vision-language-action (VLA) and world action model (WAM) policies across manipulation demands remains to be thoroughly evaluated. In this report, we evaluate GPT-6-Astra on EBench as the core of an agent that controls a robot through an agent harness under the same visual and proprioceptive observation protocol as post-trained policies. We study two agent configurations: direct control, in which the agent generates waypoint commands directly, and harness with policy, in which it can also invoke a post-trained VLA or WAM policy for action execution. Under direct control, the agent achieves competitive overall performance relative to post-trained policies. It leads on short-horizon tasks and exhibits the least performance variation across generalization dimensions, whereas it underperforms post-trained policies on long-horizon tasks, tabletop manipulation, and high-precision contact-rich tasks. In the harness with policy, invoking a post-trained VLA or WAM combines the complementary strengths of both components, raising overall task success to the level of the strongest compared post-trained policies while substantially reducing token cost. However, performance remains below the per-task upper bound of the standalone components due to suboptimal policy invocation, fixed-instruction execution without subgoals, and grasp disruption when the agent resumes control. Overall, our study characterizes the capability boundary of a frontier MLLM agent and suggests directions for improving future agent harness design. Project page: https://internrobotics.github.io/EBench-Astra-Tech-Report/

arXiv AI Papershace 18 días

Reva: Real-Time Vulnerability Detection through Contextual Probing of Large Language Models

A vulnerability costs more to fix the later it is discovered. As coding agents generate code faster than developers can review it, software needs a first line of defense screening each new function in real time. LLM agents have recently become capable of finding real-world vulnerabilities by working as a security expert would: they run analysis tools, explore the surrounding code, test hypotheses and weigh the evidence before forming their own judgement. This reasoning, however, spans many model calls, so that each verdict takes minutes of compute, far too long for real-time screening. Faster detectors, which fine-tune deep learning models or prompt an LLM for a direct verdict, are less accurate: they judge a function without the context an agent gathers, and an LLM's generated verdict is unreliable even when given some of it. We present REVA (REal-time Vulnerability Analysis), an AI vulnerability detector that brings the security context and code understanding components behind agentic detectors into a single forward pass of an LLM. For each new function, REVA gives the LLM a pre-computed threat model of its module, lightweight static analysis findings and call-graph information. REVA leverages a trained probe to read the LLM's internal representations of the new function in this context. On PrimeVul's held-out test set, REVA on GLM-5.3-Flash reaches 60.0% accuracy in about one second per function. This matches VulTriage (59.2%), the state-of-the-art agentic detector, run on the same LLM, while being over 250 times faster and less than a tenth as costly. On the same prompt, the probe is 14% to 17% more accurate than the LLM's own direct verdict. Across three open LLMs, REVA is at least as accurate as VulTriage, and about 10% more accurate on two of them. An LLM's internal representations thus encode much of what agentic detectors obtain through reasoning.

arXiv AI Papershace 18 días

Understanding Jev-Style System One Models: Does Structure Reduce Selection Bias?

Selection bias in large language models (LLMs) can persist despite instructions to ignore presentation order or prompts that elicit reasoning. Recent efforts to reproduce Jev-style System One behavior have produced a growing family of models that predict probabilities directly over supplied candidates. Yet systematic comparisons of selection bias across Jev versions, Jev reproductions, and generative LLMs remain limited in multi-option tasks. This leaves a central question: does Jev-style decision structure reduce selection bias? We address this question with a simple test that shuffles options while keeping question and candidate content fixed. Across two Jev versions, twelve Jev reproductions, and six generative LLMs, we trace individual answers through eight presentations and compare them with fixed-order repetition. On our primary collection of 1,700 questions from two benchmarks, Jev's accuracy varies by less than two percentage points, yet its cumulative failure set expands by 27--29\% relative to the original order, with first failures continuing into the final presentation. This loss of initial successes is substantially larger than under fixed-order repetition, even for highly repeatable Jev reproductions. Nor does Jev show a general advantage over the LLM baselines in preserving correct choices. Aggregation offers limited protection: voting yields modest gains for Jev, while even unanimous answers are frequently wrong. These results expose a gap between repeatable decisions and decisions that remain correct across equivalent presentations, showing why stable accuracy alone cannot establish resistance to selection bias.

arXiv AI Papershace 18 días

Pretraining Knows What, Simulation Teaches How: Compositional Generalization in a Sim-Trained Robot Agent

A robot policy needs to understand what a task is asking for in a scene, and how to move the robot to accomplish that task. To achieve these two skills, we introduce Zero-real Agent Policy (ZAP), a recipe that teaches a small off-the-shelf vision-language model to control a robot via a simple tool harness. We use a pretrained 27B parameter model to bootstrap world understanding and reasoning, followed by supervised fine-tuning and GRPO on simple simulated tasks to teach it how to move the robot. Without any real-world robot data during training, ZAP reaches 96.7% success on real-world tasks, matching GPT-6 Astra in our harness and outperforming the MolmoAct2 VLA (33.3%), while running 3x faster and 17x cheaper than GPT-6 Astra. We demonstrate that ZAP also generalizes to completely new objects, goals, strategies, and language by seating a test tube in a rack (and choosing a different hole in the rack when prompted), handing a test tube between arms mid-air, drawing an X on a tic-tac-toe board, and playing chess. These goals and types of manipulation don't appear in our training data. Furthermore, ZAP also transfers to an unseen embodiment. On a single-arm Franka, ZAP reaches 55.9% on RoboLab-120, 4.9 percentage points ahead of the leading VLA/WAM baseline.

arXiv AI Papershace 18 días

Does VGGT Know What It Does Not Know?

Feed-forward 3D reconstruction models such as VGGT regress cameras and geometry in a single forward pass, but they report no uncertainty for the camera poses they predict: an incorrect relative pose looks no different from a correct one. We ask whether the frozen model nonetheless knows when its pose is wrong. On HardMatch, a benchmark of 1,000 deliberately difficult image pairs on which VGGT fails on 35% of pairs, a linear probe on VGGT's frozen features predicts pose failure on held-out categories with ROC-AUC 93.5%, compared with 79.7% for VGGT's own point-map confidence and 77.7% for a classifier trained on all of VGGT's outputs. The probe catches failures VGGT is confident about, lets a downstream system abstain on the pairs it distrusts, and transfers without retraining to rotation failures on PhotoTourism and ScanNet-1500, beyond what co-visibility explains. This knowledge is absent from the per-image encoder: it is computed partway through the layers that fuse the two views, and the same holds for a second model, π^3. In essence, we find that VGGT knows more about its own failures than it reports.

arXiv AI Papershace 18 días

Learning When to Commit: Schedule Distillation for Fast Diffusion Language Models

Diffusion large language models (dLLMs) can generate multiple tokens in parallel, but often require many decoding rounds to complete a response. Efficient generation depends not only on which tokens are easy to predict, but also on how committing to them helps predict the remaining positions. A highly predictable token may provide little useful context, while revealing a less predictable token can help several other positions become predictable sooner. We propose dSpark, a new schedule distillation-based framework for learning when to commit tokens for faster diffusion language generation. To achieve this, we construct unmasking schedules that prioritize tokens whose reveal makes other positions easier to predict, reducing uncertainty earlier in decoding. We then distill these schedules into the model to enable faster parallel sampling, with a KL-divergence regularizer to help preserve generation quality. We extensively evaluate our approach across mathematical reasoning and code generation tasks. On GSM8K, our method achieves remarkable 10.12and 5.12speedups over the various baselines on LLaDA-8B-Instruct and Dream-7B-Instruct, respectively, while maintaining superior task performance. Our code is available at https://github.com/VILA-Lab/dSpark.

arXiv AI Papershace 18 días

Game theory of academic effort and hiring in the age of AI

Many academic disciplines (in particular, mathematics and the theoretical sciences) face difficult questions about how to adapt their practices to the rise of highly capable artificial intelligence (AI). Before AI, a well-written research paper with a strong theoretical result acted as a reliable signal that its authors understood their field in depth. Now that AI is widely accessible, such signals have become cheaper to produce. As a result, they no longer reliably sort academics by their level of understanding. This has direct consequences for academic hiring. The age of AI raises a question for each side of the academic job market. Should an early-career researcher invest in a deep understanding of their field, or instead use AI to build a longer resume with many papers? Should a recruiter favor applicants with fewer papers who pass an interview aimed at probing understanding, or insist on both publication volume and a pass? In this paper, we develop a game-theoretic model of this two-sided dilemma. We classify all pure-strategy equilibria that can arise in this game and use learning simulations to identify the behaviors that populations of applicants and recruiters may learn to adopt. Our analysis provides a framework for reasoning about how academic hiring can adapt to AI and for evaluating practices that preserve the value of deep understanding.

arXiv AI Papershace 18 días

FOCUS: Active Critic Learning for Long-Horizon Language Model Agents

Sparse outcome rewards make it difficult for long-horizon language-model agents to assign credit to earlier actions. Learned critics provide value estimates for intermediate states, but these estimates can be inaccurate, while estimating intermediate values with repetitive continuation samples from points within a trajectory incurs substantial costs. To mitigate these issues, we introduce FOCUS, an RL method that uses selective branching to help the critic network learn its value estimator. The rollout controller facilitates active learning for the critic by sampling continuations from states that are expected to most reduce the critic's estimation error. These refined value estimates then score the segment between consecutive states with a TD(λ) advantage, while the sampled evidence supplies training targets for the critic. FOCUS supports long-horizon agentic reinforcement learning through a modular context-management interface that decouples context management from rollout allocation. Our experiments show that using compaction-induced segment boundaries for value estimation and credit assignment, FOCUS outperforms GRPO and VinePPO with matched per-step generation budgets on legal reasoning, deep search, and code localization benchmarks, improving held-out accuracy at Qwen3.5-9B scale by 7.2 points over the strongest baseline on BrowseComp-Plus (55.4 vs. 48.2) and by 4.1 points on DeonticBench (56.5 vs. 52.4).

arXiv AI Papershace 18 días

Quantifying Claim Smoothing by LLM Research Agents

Deep research, or any use of large language models (LLMs) to read a large document collection and produce insights, requires synthesizing information from different sources. In this work, we find that LLMs produce long-form research reports that read as confident syntheses but which fail to accurately reflect the underlying documents. We call this failure smoothing. In collapsing diverse and potentially conflicting sources into a single narrative, the model discards the evidential "jaggedness" of its corpus and asserts conclusions more decisive than the citations can support. We introduce a claim level scoring pipeline, SmoothScore, to quantify this effect. Our pipeline extracts key conclusive claims from model generated reports, then finds evidence relevant to each claim and clusters that evidence along high level aspects. Finally, each claim is scored with respect to the evidence to quantify the degree of smoothing done by the agent. Across 7 systems and 4 datasets, we find that agents consistently produce stronger claims than warranted.

arXiv AI Papershace 18 días

Selective Tie Learning: Mitigating Spurious Correlations in Preference Optimization

Preference optimization can learn spurious correlations, and augmenting the preference data with ties (equally preferred responses) can reduce spurious reliance. However, ties differ in their corrective effect, making their selection for augmentation important. For log-linear policies, we derive each tie's influence on spurious reliance and show that taking the largest predicted reductions is first-order optimal among subsets of equal size. Computing these influences requires knowledge of the spurious features and the curvature of the training objective, which is often unavailable or expensive to estimate. To this end, we explore using the magnitude of a tie's implicit reward margin and characterize the conditions under which it is a useful proxy for predicted influence. To test the theory and its qualitative transfer to neural models, we estimate influence with a counterfactual proxy for spurious reliance and use a generalized Gauss--Newton approximation to curvature for language models. Across log-linear policies, neural models, and large language models, our selection rules consistently achieve comparable reductions in spurious reliance with fewer ties than uniform selection.

arXiv AI Papershace 18 días

A theory of noisy linear recurrent neural network learning dynamics

Biological neural networks are noisy: their dynamics involve difficult-to-predict fluctuations which are likely due not just to chaos, but also to a variety of low-level biophysical processes. Despite the omnipresence of intrinsic noise, its theoretical consequences are incompletely understood, especially in the context of learning. For example, does noise meaningfully shape representation learning, or is it better viewed as a nuisance factor whose presence or absence changes little about what solution is found and when? Here, we argue that intrinsic noise can significantly modulate representation learning in recurrent neural networks (RNNs), since it can strongly break solution-space degeneracies and in some cases control whether a non-degenerate solution exists at all. We show that this is true for an analytically tractable model of noisy RNN learning with a linear activation function and white noise inputs. We find that, in the so-called rich regime (where weights are initialized with small values), noise strongly modulates the directions in which the read-in, readout, and recurrent weights align. Moreover, we find that noise produces a kind of phase transition: for sufficiently high noise, a non-degenerate solution is not even optimal if parameter scales are limited (e.g., via regularization). These results extend recent analytic work on RNN representation learning and show that noise can act as a fundamental determinant of representations.

arXiv AI Papershace 18 días

Atomic Security Concerns: Towards a Foundational Language for Security Knowledge Representation

Cybersecurity governance, risk management, compliance, and security architecture rely on numerous concepts, including security properties, domains, controls, risks, implementation measures, capabilities, and compliance requirements. Although these concepts are widely used, they are often represented through heterogeneous abstractions and vocabularies, making relationships among them difficult to express consistently and leaving much of the associated reasoning dependent on expert knowledge. This paper introduces the concepts of Security Concerns and Atomic Security Concerns (ASCs) and explores the hypothesis that cybersecurity knowledge may be represented through a finite set of foundational concerns. Inspired by the role of atoms in chemistry, the proposed approach views cybersecurity concepts as compositions of more fundamental concerns rather than as independent abstractions. Hence, Atomic Security Concerns could be viewed as a common abstraction layer for the representation of cybersecurity concepts, independently of their domain, nature, or level of granularity. Building upon this perspective, the paper explores how Atomic Security Concerns may provide the vocabulary of a cybersecurity language and how cybersecurity concepts may be represented as concern sets supporting explicit relationships among concepts. The contribution of this work is intentionally conceptual: it introduces a framework and research hypothesis for investigating the foundations of a common language for cybersecurity knowledge representation, reasoning, and future formalization.

arXiv AI Papershace 18 días

A Unified Arena for LLM Agents across Diverse Eurogame

Interactive environments offer a practical way to evaluate large language model (LLM) agents, and competitive games extend this evaluation to decisions shaped by competing goals and other agents' choices. Yet existing evaluations confined to individual games or execution frameworks leave unclear which competitive advantages persist across settings. We introduce DECE (Diverse Eurogame Competitive Environments), a unified arena spanning ten Eurogames with shared interfaces and evaluation protocols. These games combine lasting strategic commitments, competition over shared resources, and multiple routes to victory. Our primary study evaluates 12 model-harness configurations, spanning nine models and four execution harnesses, in 480 three-player matches across ten games. The main and supplementary studies comprise 1,032 completed matches in total. These comparisons reveal broad cross-game ranking agreement alongside game-specific strengths. Matched studies show that the same model can perform differently across execution harnesses, adding another source of variation to these profiles. These performance advantages do not consistently translate into better token or cost efficiency, making resource use an additional consideration when comparing configurations. Together, the findings support evaluating complete agent configurations by both their competitive performance across games and the resources needed to achieve it.

arXiv AI Papershace 18 días

Free to Look, Free to Reason: Simple LLM Harnesses Outdiagnose Hand-Crafted Root Cause Analysis Agents

Root cause analysis (RCA) of microservice failures has attracted many LLM agents whose investigation is hand-crafted, through standard operating procedures, fixed stages and role agents. Each is evaluated on its own data against few baselines, so it is unknown whether this design beats letting a capable model investigate on its own. We compare three simple harnesses that prescribe none of the model's calls (one call over an evidence package, a SQL agent and the Codex coding agent) with five published LLM RCA agents and with statistical and causal methods on RCAtlas, 90 validated cases from eight public datasets, across six models and three reasoning efforts. The simple harnesses are the most accurate (PR@1 65-77% on three mid-tier models), ahead of the published agents (16-53%) and the non-LLM methods (6-39%), on every model on which both families were run. The published agents fix from a tenth to all of their model calls in the program, and a case study shows how these prescriptions skip the decisive evidence. A stronger model mainly improves reliability, a multi-turn harness mainly improves aptitude, and more reasoning effort changes accuracy by a few points at most while multiplying the time by up to seven. Among runs more accurate than the best non-LLM method, every Pareto-optimal run in cost or time uses a simple harness.

arXiv AI Papershace 18 días

Improving Signals, Declining Encoder: A Known-Answer Test of Label-Free Monitoring in Self-Supervised Learning

Self-supervised training is monitored without labels, through the loss and statistics of the learned features. We report a case where we know the encoders lost part of what they had learned, so we can check whether these signals noticed. We trained LeJEPA encoders on synthetic images with known generating factors. In nine of ten seeds, the readability of those factors fell abruptly and stayed down, nearly always along the same direction in factor space. Meanwhile the loss kept falling, and most of the eleven label-free signals we tracked kept improving and rated the declining runs better than runs that did not decline. Standard change-point detection rarely flagged the decline near its onset, and selecting checkpoints by these signals mostly chose ones after it.

arXiv AI Papershace 18 días

Do Similarity Measures Measure Similarity? How User Activity Creates Phantom Correlation in Recommender Systems

Item-item similarity is one of the oldest and most foundational signals in recommender systems: two items are assumed to be similar when the same users consume them. Despite the sparsity and scale of interaction data, the implicit assumption across the field is that item co-occurrence reflects genuine shared preference, capturing what users like together. In this paper, we demonstrate that a substantial fraction of observed item-item similarity is an artifact of variance in user activity rather than shared taste. Even if every user consumed items completely independently, differing only in their total consumption volume, standard similarity measures still detect strong associations between items. We term this phenomenon phantom correlation: spurious item association that emerges strictly from heterogeneous user activity in the complete absence of preference structure. Starting from minimal formal assumptions, we show that independent consumption under varying activity is described by a Rasch model, and we derive how the resulting association depends on item popularity for each similarity measure. By comparing our theoretical baseline against real-world benchmark datasets, we reveal the extent to which popular similarity measures reflect phantom correlation, which bears on how item co-occurrence is interpreted and used in collaborative filtering.

arXiv AI Papershace 18 días

Beyond Physical Illness: Can Clinical Narratives Reveal Emerging Mental Health Risks in Chronically Ill Patients?

Mental-health risks such as depression, dementia, and suicidal ideation can emerge during hospitalization among chronically ill patients and may worsen without timely recognition and intervention. However, most existing mental health studies focus on patients with established psychiatric conditions rather than patients hospitalized primarily for physical illness. In this work, we investigate whether latent mental-health risk in chronically ill patients can be inferred from longitudinal clinical narratives. We construct a weakly supervised cohort around 4,000 patients from MIMIC-III using mental-health evidence extracted from discharge summaries. A hierarchical state-space framework derives mental-health representations from day-wise clinical narratives and models their temporal evolution for risk estimation throughout hospitalization. Preliminary results demonstrate improved binary risk prediction using the proposed representations and show distinct model derived risk trajectories between risk and no-risk cohorts, providing a foundation for studying emerging mental-health risks in chronically ill patients.

arXiv AI Papershace 18 días

EEGaze: Multimodal Attention Prediction in Visual Search via EEG-Based Saliency Refinement

Visual search is ubiquitous in daily life, from locating keys in a messy apartment to finding a tool in a cluttered workshop or an application icon on a desktop screen. While visual attention is to a large extent driven by image content, research in psychology indicates that user cognitive states have a significant influence on attention allocation. A promising way to access information about cognitive states is via Electroencephalography (EEG) recordings. Despite this potential, previous works in computational attention modeling have largely disregarded EEG, and related studies in psychology operate with abstract, simplistic stimuli, or even artificially guided search procedures. We bridge this gap with two contributions. First, we present EEGaze 140, the largest publicly available multimodal dataset of free visual search with realistic stimuli. The dataset contains synchronized EEG and eye tracking recordings from 36 participants across 140 naturalistic scenes, covering two scenarios: searching for desktop icons and finding tools in cluttered workshop scenes. Second, we introduce EEGazeNet, the first method for human attention prediction in visual search that integrates EEG signals. Our approach consists of a EEG guided refinement stage that adapts the output of a state of the art saliency prediction method to the user current cognitive state. In comprehensive evaluations on EEGaze 140, we show that EEGazeNet consistently outperforms all baselines and ablation conditions.

arXiv AI Papershace 18 días

CLOSURE: Cross-Channel Closure for Multivariate Time Series Anomaly Detection

Multivariate time series produced by industrial installations and server clusters are under continuous monitoring, and anomaly detection is the first line of defense before a local fault escalates into a system-wide failure. In the absence of labels, the dominant paradigm trains a reconstruction model on normal data and scores the residual to reach a verdict. Recent work has steadily enriched the expressiveness of the recovery map through Transformers, diffusion models, and cross-scale associations. A residual, however, only certifies that an observation can be recovered; it does not certify that the variables still explain one another as they do under normal operation. A coordinated yet unfamiliar operating transition inflates the residual without breaking any relationship, whereas a sensor drifting away from its peers may leave the residual silent. Methods that fold inter-variable dependencies into the reconstruction objective inherit this ambiguity, because their relational evidence is ultimately supervised by the same recovery loss. We therefore propose CLOSURE, which evaluates recoverability and cross-channel consistency as two normal-data prediction tasks learned in separation. The recoverability task is solved by a diffusion reconstructor conditioned on a discrete memory of normal temporal bases. The consistency task is solved by a relation-closure operator that predicts every channel exclusively from the others, where the target's own trajectory is withheld and sparse signed gates decide which peers are actually used. The two scores are reconciled only through calibration on normal data. Extensive experiments show that CLOSURE consistently outperforms recent reconstruction-, frequency-, and topology-based detectors on public benchmark streams, confirming the value of examining consistency alongside recoverability.

arXiv AI Papershace 18 días

Model Selection for Off-Policy Evaluation via Pairwise Bellman Validation

Model selection and hyperparameter tuning pose a central challenge in off-policy evaluation (OPE) because errors of fitted Q-functions and occupancy-ratio estimates cannot be directly evaluated from held-out transitions. Bellman-residual and minimax validation methods often require auxiliary regressions or optimization over separate critic classes, adding computational and tuning costs. We introduce pairwise Bellman validation (PBV), a tuning-free minimax procedure for selecting among finite classes of fitted Q-functions and occupancy-ratio estimators using held-out transitions. PBV constructs critics from pairwise candidate differences, requiring only finitely many empirical moment calculations without auxiliary validation models or separate critic optimization. We establish finite-sample oracle inequalities relative to the best available candidate, even under candidate-class misspecification. Our key insight is that Q-function errors and log ratios between candidate and true occupancy ratios provide oracle critics whose Bellman and adjoint Bellman moments control estimation error. Pairwise candidate differences approximate these unknown critics, while penalties account for comparator approximation error. For Q-function selection, we first develop unweighted PBV, which evaluates pairwise Bellman moments under the offline distribution and admits oracle guarantees under a pairwise coercivity condition relating the offline distribution to the target-policy dynamics. We then develop target-aligned PBV (TA-PBV), which uses marginalized importance weighting to provide error control without pairwise coercivity. For occupancy-ratio selection, we develop adjoint PBV (A-PBV), which evaluates adjoint Bellman moments against log-ratio critics. On Hopper and Deep OPE, unweighted PBV performs competitively, while target-occupancy weighting can improve performance in settings with controlled overlap.