Noticias de IA - ítems

arXiv:2610.24168v2 Announce Type: replace-cross Abstract: Large language model (LLM) agents are increasingly incorporated into agent-based models (ABMs) on the premise that richer, context-sensitive human behavior can improve empirical fidelity. However, LLM-supported ABMs introduce an open-ended design space in which subtle modeling and configuration choices, such as prompt design and environment coupling, can substantially alter outcomes. Developing LLM-supported ABMs to simulate human collective behavior therefore requires navigating configuration sensitivity while identifying the limitations of the current model architecture. In this paper, we introduce the Model Capacity Frontier (MCF), a framework that characterizes the empirical performance attainable by a specified architecture across its admissible configurations. In multi-objective settings, this capacity is represented by an estimated Pareto frontier obtained through systematic configuration exploration. We demonstrate MCF in crowd simulation of university commencement events with approximately 12,000 agents, using empirical movement data on temporal occupancy, spatial distribution, and flow direction. We first characterize the capacity of a conventional crowd model and then evaluate successive LLM-supported architectural extensions. In the process, MCF reveals which dimensions of attainable performance expand and which remain limited, providing empirical targets for refinement and re-evaluation. Initial LLM integration improves flow direction but degrades temporal occupancy and spatial distribution; subsequent architectural refinement of the LLM layer improves temporal occupancy and recovers spatial performance, while retaining much of the earlier gain in flow direction. Our case study shows how MCF guides the systematic architectural development of LLM-supported ABMs in configuration-sensitive, open-ended design spaces.

arXiv:2610.21561v3 Announce Type: replace-cross Abstract: Large language models (LLMs) deployed as conversational agents are increasingly expected to follow procedural logic. While factual hallucination has received extensive attention, a different failure mode has been overlooked: procedural hallucination, in which an agent produces individually fluent and factually plausible utterances while deviating from the intended conversational process. We introduce DialFork, the first benchmark to measure this failure mode on raw dialogue text. DialFork is built by forking conformant dialogues at controlled turn positions and injecting one of four hallucination types, yielding 36,278 dialogues across 24 tasks in 12 domains; 30 human annotators validate the construction. We define three evaluation tasks that characterize procedural deviation at three granularities: how far a set of dialogues deviates from conformant ones (monitoring deployed agents), how far a single dialogue does (procedural retrieval), and at which turn a dialogue first goes off-track (real-time guardrails). For the first two, we propose dialog-flow-aware metrics, FlowKL and FlowScore, that compare dialogues by how they unfold rather than their content. At the dialogue level, FlowScore significantly outperforms all baselines across seven encoders, showing that semantic similarity is no substitute for procedural similarity. We release DialFork under open licenses at https://github.com/idiap/dialfork, hoping the community will measure conversational process alongside content.

arXiv:2610.20064v2 Announce Type: replace-cross Abstract: We present FoveaVLA, a Vision-Language-Action (VLA) model that replaces the wrist camera with learned foveation actions. Wrist cameras provide close-up detail but restrict the view to the end-effector's surroundings. External cameras see the whole scene, but the policy receives them at low resolution, so the fine detail needed for manipulation is lost. FoveaVLA recovers this detail by predicting, alongside each action chunk, the interaction region to observe at the next policy query. The policy then receives crops of these regions, taken from the original-resolution images, together with the full views. We build FoveaVLA on $\pi_{0.5}$ in two configurations, Single and Dual, using one or two external stereo cameras with full and foveal RGB and depth inputs, and post-train both on DROID with RoboInter interaction annotations. Across four real-world task suites, FoveaVLA-Single achieves 70.8% success with one external RGB-D view, and FoveaVLA-Dual achieves 77.1%, versus 52.1% for the wrist-equipped $\pi_{0.5}$. Both match the baseline on plain pick-and-place and improve success with distractors, cloth manipulation, and far-away objects.

arXiv:2610.18690v2 Announce Type: replace-cross Abstract: Image-token compression must reduce computation without hurting the semantics. We introduce Varest, guided by semantic fungibility: tokens within a coherent semantic region can be summarized coarsely, while object boundaries and literal text require finer distinctions. Object and text localizers guide variable-resolution merging, preserving spatial coverage and finer detail at boundaries and text. Learned [REL] registers then prune visual tokens after early decoder processing. Training across budgets enables one checkpoint per backbone. Across three backbones and 13 benchmarks, Varest outperforms 15 baselines in mean performance retention at every evaluated budget, retaining 86.10-92.33% of full-token performance with only 5.56% of the original layer-averaged image-token count.

arXiv:2610.16067v2 Announce Type: replace-cross Abstract: Agent evaluation harnesses are software: they run a test suite, drive a test input generator, and decide outcomes with a test oracle. We audit them for defects a score table cannot show. First, a systematic oracle-integrity auditing approach - five state-integrity invariants and three known-answer controls - is measured by injecting seven oracle defect classes across two task domains and a synthetic domain. Every class is detected, and on the 57 separable variant-target combinations it detects 54; the repaired oracle yields no false positive. On a widely used agent benchmark, it exposes a reward function that resets the database to build a ground-truth hash and never restores it, so a harness scoring afterwards reads the reference solution's state: recorded labels report 4,194 database passes over 4,320 episodes, replay gives 2,883, and success falls from 94.3% to 65.5%. The correction is not a uniform offset; it removes 24 to 34 points from each agent and changes which agent this matrix measures as strongest. Second, more weakly, localising the apparatus is a property of the measurement: a 3,240-episode ablation associates the three documents with lower collapse in the tested configuration, without ordering them, and a second task domain reverses the language effect. Third, the method did not transfer to a second, independently implemented benchmark; we name the contract required. A replication package recomputes every analysis offline with no model calls.

arXiv:2610.15428v2 Announce Type: replace-cross Abstract: Context-based offline meta-reinforcement learning (COMRL) has advanced almost entirely through better task representations, which separate task identity from artifacts of the behavior policy so that the latent reflects the task alone. The implicit expectation has been that a clean representation will transfer to out-of-support (parameter-extrapolation) tasks. On such a task, however, the encoder produces a latent outside the training support, where the policy has never been trained. We address this gap with Pessimistic Latent Task-aware Optimization (PLATO), which exposes the policy to off-support latents during training. PLATO assumes a mild manifold hypothesis: task latents lie on a low-dimensional manifold, and out-of-support tasks lie farther from the training centroid. It exploits this geometry by perturbing inferred latents outward from the centroid to synthesize counterfactual tasks. A learned decoder ensemble then rolls out short trajectories at each perturbed latent, and the policy is updated on the imagined transitions under a pessimism weight that down-weights steps on which the ensemble disagrees. We prove that the perturbation leaves the training support by a controllable margin and that the out-of-support generalization gap is bounded by the same perturbation reach, and we give explicit conditions, verified empirically, under which the epistemic disagreement at the perturbed latent stays bounded away from zero. On eight MuJoCo continuous-control benchmarks, PLATO consistently improves out-of-support returns over strong COMRL baselines while remaining competitive on in-distribution tasks. Our code is available at https://github.com/deephoony/plato.

arXiv:2610.15207v2 Announce Type: replace-cross Abstract: Large Language Models (LLMs) have been shown to be computationally expensive and vulnerable to jailbreak attacks, in which adversarial prompts bypass safety alignment and induce harmful responses. To address this inefficiency, a range of techniques have been proposed to enable efficient LLM inference. Depth-based dynamic inference has demonstrated promising results by enabling outputs to be generated from intermediate layers. It offers the flexibility to adjust model behavior based on the available resource budget. However, how this flexibility affects the robustness of LLMs against jailbreak attacks remains unexplored. In this paper, we address this gap via a comprehensive investigation into the impact of depth-based dynamic inference on the robustness of LLMs against jailbreak attacks. Specifically, we make the following key observations: 1) Shallow exits are consistently more robust to jailbreak attacks than deeper ones; 2) Exit depth affects how easily the adversarial suffix can be optimized, with deeper exits reaching lower attack loss; and 3) Adversarial suffixes affect utility uniformly across exit depths. These insights can help design dynamic inference methodologies that are more robust to jailbreak attacks. Code is available at https://github.com/LassNajmuldeen/balcony-jailbreaking

arXiv:2610.15029v2 Announce Type: replace-cross Abstract: Disease classification from patient-reported symptom descriptions is usually approached with lightweight discriminative models or with autoregressive large language models (LLMs), which understand language broadly but generate text token by token and require output parsing. We study decision foundation models, which score a closed set of candidate labels in a single forward pass without generating text. We adapt Clef-Flash, a 9-billion parameter decision model built on a Qwen3.5 backbone, to the 24-class Symptom2Disease benchmark using 4-bit quantized low-rank adaptation (QLoRA) on a stratified split of 960 training and 240 test cases. The fine-tuned model reaches 95.42% accuracy (95% CI 91.98-97.42%) and a macro F1 of 0.9551, up from 65.00% for the unadapted weights (62.92% through the remote API). However, TF-IDF linear SVM and logistic regression baselines trained on the same split reach 98.75% (95% CI 96.39-99.57%) with sub-millisecond CPU inference, so on this benchmark the fine-tuned model offers no accuracy advantage over a well-tuned linear classifier. A Leave-One-Symptom-Out clause-ablation analysis of four illustrative narratives shows predictions depending on distinguishing clauses: in a psoriasis example, removing the clause describing silvery scales lowers the predicted probability from above 99.99% to 0.12% and shifts the prediction to impetigo, while removing other clauses changes it by less than 0.001 percentage points. Decision architectures deliver structured, single-pass predictions from large language backbones, but short, lexically distinctive benchmarks such as Symptom2Disease cannot show an advantage over simple baselines; validation on longer, real clinical text and clinician review are needed.

arXiv:2610.12445v4 Announce Type: replace-cross Abstract: Recent incidents have highlighted the challenge of monitoring LLM agents and the danger of models deceiving people. We show that white-box deception detection via probes can be scaled up to frontier monitoring settings by collecting the largest deception dataset to date for training probes and introducing a novel probe architecture which can aggregate information across many layers and tokens. Our probes surpass black-box monitors on agentic coding benchmarks, achieving 98.3% AUROC in SHADE-Arena and 98.0% in BashArena (vs. 97.7%, 97.6% resp. for action-only text monitors), and show improved efficacy as the underlying model is scaled up. We further test probes on several cases where deception cannot be determined from the context alone; in these cases of introspective deception, the ground truth can only be determined through careful elicitation or knowledge of a model's training data. In one such evaluation, we show that probes can distinguish transcripts containing a model's true hidden goal from other goals with an AUROC of up to 99.7%. Our probes also detect deception on prominent open-weight models which lie about politically sensitive topics when put under pressure. We release our training dataset, dubbed FIBS, to help drive frontier deployment of effective probes, and encourage the community to expand upon it with further examples of deception and sabotage.

arXiv:2610.11287v3 Announce Type: replace-cross Abstract: Long-horizon agents compact their history to continue within a finite context window, but a textual summary alone may not support every subsequent decision. We introduce REMORY, a neural memory network that supplements the summary with a bounded sequence of soft memory tokens. Given the history and summary, the network learns to generate tokens that help a frozen LLM approximate the continuation it would produce with the full history. The tokens are conditioned on the summary and appended after it, forming an analogue of a residual connection along the sequence dimension. On SummHay, REMORY improves source attribution at nearly unchanged insight coverage and approaches the full-context joint score using only 5.2% of the input positions. Across long-horizon agent benchmarks, Qwen3.8-27B and GLM-5.3-Flash show consistent gains with residual memory. Both models also exhibit substantially fewer repeated tool outputs and tool errors on BrowseComp and Terminal-Bench 2.1.