NousResearch/Arete-27B
New text-generation model. Tags: qwen3_5, nous-research, arete, supervised-fine-tuning, value-alignment, text-generation, conversational, dataset:NousResearch/arete-data-pairs
en♥ 105
Seguridad y regulación
2441
179
Safer for Whom? A Systematic Exploration of AI Safety in Eating Disorder Interactions
AI assistants advise users about diet, exercise, and weight loss---areas in which unsafe guidance can worsen body dissatisfaction, dietary restriction, compulsive exercise, and other behaviors associated with eating disorders (EDs). Existing evaluations emphasize diagnostic accuracy, overtly harmful prompts, or synthetic scenarios, providing limited insight into whether models recognize implicit vulnerability in natural interactions or protect users consistently. We evaluate responses from three open-weight large language models to 100 questions sourced from an online pro-anorexia community. We grade responses using clinically informed criteria that extend a seven-dimensional taxonomy of ED-related harms. We also construct matched conditions in which questions are accompanied by information about the user varying race, gender, body mass index (BMI), and explicit ED vulnerability. Safety varies substantially across models and harm dimensions. High-risk responses are approximately twice as frequent without bios, suggesting that models do not consistently recognize vulnerability from language alone. Explicit signals of ED vulnerability improve safety by 10--14 points across models. Responses also differ significantly by race, gender, and BMI, with body size producing the largest disparities. These patterns are consistent with reliance on the stereotype that EDs primarily affect thin, White women, potentially leaving users outside this profile with less protective responses.
en
PatternBloom: Empowering Agentic RAG with Externalized RL-Distilled Graph Patterns
Reinforcement-learning-trained agents for retrieval-augmented generation retrieve content (passages, knowledge-graph fragments, or hyperedges) and supervise training with outcome rewards on the final answer. Across these paradigms the reasoning route through retrieved content never leaves the language model's hidden state, conflating declarative knowledge with the procedural knowledge that recurs across queries of the same logical type; content-only methods consequently plateau as reasoning depth grows. We introduce PatternBloom, an agentic RAG framework that shifts the unit of retrieval from content (what to read) to logic (how to reason). A first RL stage scores the agent's constructed evidence graph by a frozen oracle's information gain through the Information-Density Reward (IDR), a size-normalized mutual-information surrogate. High-reward trajectories are distilled into a Graph Pattern Memory (GPM) of type-abstracted reasoning skeletons, and a second RL stage uses the Pattern-Augmented Reward (PAR) to couple policy to memory, turning procedural reasoning into a learnable, externalized structural prior. A 7B PatternBloom outperforms 27 baselines by +15.0 Avg-OOD EM, with the gain compounding with reasoning depth precisely because procedural knowledge, once distilled, is reused rather than re-derived at every query. We release our code at https://github.com/EllenYiGe/PatternBloom.
en
Represented but Not Used: Value-Policy Dissociation in LLM Agents
Large language model (LLM) agents need to choose actions whose consequences unfold over multiple future steps, yet poor decisions do not reveal whether the model failed to represent those consequences or failed to use them. Mechanistic interpretability has studied internal representations of environment state and plans, but sequential decision making also requires evaluating how candidate actions change future outcomes. Reinforcement learning formalizes this as action value. We study this computation in Qwen2.5-7B-Instruct, Qwen2.5-32B-Instruct, and GPT-OSS-20B across a partially observed Grid environment and WebShop, a simulated web shopping environment. We find that the relative values of candidate actions are linearly decodable from intermediate hidden states, including at states where the policy prefers a worse action. We call this gap between value representation and policy choice value-policy dissociation. To investigate how represented value relates to action selection, we compare direct interventions along the value readout direction with Natural Value Transport (NVT), a transformation learned from hidden state changes that naturally result from differences in action value. While readout-direction interventions have limited and inconsistent effects, NVT more consistently influences action preferences, particularly at later layers. We also test a closed-loop controller that uses the value readout as a sensor and NVT as an actuator, observing improvements in task outcomes. These results suggest that LLM agents can represent useful prospective value without reliably translating it into policy choice.
en
Decentralised Drone Swarm Formation under Silent Failures with ACO Pheromone Claims and PSO Tuning
We study decentralised drone formation, without a central planner, under the conditions that make it hard: nobody knows how many drones are alive, failures are silent, radio is short-ranged and lossy, and actuation is noisy. In our controller, ant colony optimisation (ACO) allocates drones to picture positions through claims that behave as reinforced, evaporating pheromone and are gossiped between neighbours, so evaporation doubles as failure detection. A particle swarm optimisation (PSO) law flies the drones, a noise-buffered Voronoi filter keeps them apart, and PSO also tunes the controller offline. We prove that claims of destroyed drones vanish within bounded time, that conflicts between drones that exchange tables are resolved consistently, that survivors cover the picture to within a factor of two, and that drones cannot collide while their safety cells are feasible. On held-out seeds and with two tuned configurations, the controller formed the picture in 95-99% of runs in the hardest regime, restored it after a silent loss of half the swarm in 96-100% of runs at normal difficulty, transferred unchanged to eight unseen shapes and to 300 drones, and overlapped in 5 of 21,088 runs. A CBBA-style baseline that shares the evaporating claims but uses argmax without swaps was at least as reliable as ours once tuned with the same budget, and repaired faster, whereas ours generalised better to unseen shapes. The benefit thus comes from the pheromone dynamics and from tuning, not from ACO's stochastic construction or PSO's social terms. A centralised oracle is faster.
en
HexIt: Learning Hexahedral Block Decomposition with Expert Iteration on Combinatorial Maps
Decomposing a polyhedral solid into a few well-shaped hexahedral blocks is a long-standing open problem in mesh generation. We cast it as a sequential decision problem on a 3-dimensional generalised map: an agent splits edges, faces and volumes with three insert operations until every cell is a hexahedron and the mesh passes a quality threshold, and a policy that convolves over the map's involutions chooses each insert. A successful decomposition takes over a hundred decisions, and a uniformly random policy never finds one. We clone a policy from constructed decompositions and improve it by expert iteration. On held-out chamfered solids that share no boundary topology with the cloning data, the final policy solves an expected 94.4\% in 20 samples, and expert iteration raises the best-of-20 solve rate on large chamfered solids from 0.203 to 0.355. At matched compute, PPO, GRPO and VinePPO updates all make the cloned policy worse.
en
Deploying Physically and Epistemically Compliant Agents by Distilling Belief-Space Task and Motion Planners
In this work, we train agents to act under uncertainty according to specified, belief-dependent rules. Deep reinforcement learning (RL) can in theory train agents to follow rules if encoded into well-designed rewards, but in practice, RL fails to perfectly optimize these rewards, leading to ineffective or noncompliant agent behavior. Belief-space planners by construction produce behavior in line with developer specification, but require hand-programming online planners and belief-state estimators, adding significant engineering complexity at scale. We combine the two: we plan offline to generate a dataset of sample trajectories and then distill the plans into a policy that acts on raw observations. We propose two changes to planner distillation that address problems specific to history-based policies: Dual Averaging, a loss that combines per-timestep predictive loss and average loss on task changes, and ρ-Optimal Reinforcement, a feedback mechanism that reinforces the policy's own successful rollouts when these are near-optimal. We provide a sufficient condition under which static search problems with noiseless observation models is complete, and show that maximum-likelihood-observation planning implies this condition in static search with continuous beliefs. We create a challenging new domain, Scavenger, where an agent needs to gain information in order to succeed. We evaluate our distilled policy in two versions of Scavenger, one where information is provided visually and one where it is provided as a command in natural language. In either case, the learned policy attains a success rate above 90% in Scavenger, compared with at most 5% for RL. We also find that distilled planners transfer with greater than 60% success to following unseen and intentionally vague commands and greater than 75% success to following floorplans larger than those seen in training.
en
Mental Imagery That Matters: Curing Latent Collapse in Multimodal Reasoning
Mental simulation, the ability to reason by constructing and updating internal visual states, is central to flexible human cognition, but remains elusive in current multimodal large language models. Latent multimodal reasoning offers a promising route toward this capability by interleaving text with continuous visual latent blocks, allowing models to maintain internal visual workspaces without decoding them into pixels. In this paper, we ask whether these latent visual states actually matter. Through a direct perturbation analysis across representative latent-reasoning MLLMs, we find that they often do not: dropping, replacing, or shuffling generated visual latents has only minor effects on the model's own final-answer probability, while equivalent perturbations to the linguistic reasoning trace are substantially more damaging. We call this failure mode Mental Simulation Collapse (MSC): the model appears to perform interleaved visual reasoning, but routes most answer-relevant computation through text. We further diagnose that MSC follows naturally from existing training objectives: visual alignment can make generated latents resemble helper-image embeddings, but it does not require the decoder to rely on them; subsequent text-only supervision further leaves the semantic role of the latent pathway unidentifiable. To address this, we introduce AnchorSim (Anchored Mental Simulation), a contrastive distillation framework that trains models to make their generated visual latents consequential. Across multiple visual reasoning benchmarks, AnchorSim improves performance, while diagnostic re-evaluation shows that visual perturbations now meaningfully affect predictions. Our results expose a fundamental failure mode of latent multimodal reasoning and demonstrate that mental simulation must be anchored to model behavior in order to matter.
en
Beyond Query Similarity: Safety-Conditioned Cache Routing for LLM Response Reuse
Semantic response caching reuses a cached response for a semantically similar query to improve LLM inference efficiency. However, query similarity alone can lead to unsafe reuse when safety conditions change. Existing approaches improve cache safety through stricter similarity checks, context-aware validation, or response verification, yet they may still miss safety conflicts between the cached response and the new request or introduce unnecessary verification overhead. We argue that safe and efficient semantic caching requires a reuse policy that makes the cached response's safety conditions explicit and uses them to guide the reuse decision with a balance between safety and utility. In this paper, we propose SafeCache, a safety-aware framework that separates candidate retrieval from safety approval and formulates semantic response reuse as a safety-conditioned routing problem. SafeCache matches each cached-response condition against the new request and aggregates the per-condition mismatches into an overall risk score. A separate contradiction guard prevents explicit conflicts from being diluted by strong matches. It further estimates matching uncertainty and calibrates the routing policy under a required unsafe-reuse target and a verifier-call budget to choose among direct reuse, selective verification, or regeneration. Experiments show that SafeCache achieves the better safety-utility trade-off among the evaluated baselines, enabling safer response reuse at comparable cache utility.
en
Tail-risk-aware Agent Safety: Constraining Tail Risks Across Multi-step LLM Agent Workflows
LLM-based agents extend LLMs with tool calls, memory, planning, and retrieval to act in external environments, but they are vulnerable to attacks at different workflow steps. Existing defenses, including prompt engineering and post-training alignment, primarily improve LLM-only safety and often overlook risks that arise from multi-step execution. External verification adds guardrails to check each step separately, but it typically ignores how risks can propagate across steps. We argue that we should consider risk transitions across dependent steps through the agent workflow, rather than separate verification on a single step. In this paper, we propose Tail-risk-aware Agent Safety (TAS), a tail-risk-aware safety framework that models agent workflow safety as a sequence of dependent safety states and constrains the rare but extreme risks (i.e., tail risk) with a shared budget. TAS first models safety-state transitions sequentially along the workflow as an MDP to trace risk propagation, then measures a tail-sensitive risk to target rare but high-cost risks (i.e., tail risk), and constrains the risk with a shared budget to derive a step-wise safety strategy. By constraining tail risk rather than the overall average, TAS suppresses rare high-cost failures while preserving utility in normal cases. Experiments show that TAS achieves better performance compared with eleven baselines across three benchmarks.
en
Too Big to Inspect? Ethically Auditing Large Datasets Through Small Samples
Big techs and researchers increasingly claim that their training datasets are too large to inspect for bias and ethical problems. However, in the race to build ever-large models, they have relied on automated web scraping and no human curation. Datasets built this way absorb the stereotypes, exclusionary norms, and historical prejudices of their sources, and the models trained on them can reproduce and amplify those patterns. Because manually auditing billions of samples is humanly impossible, we need practical, effective methods to identify biases in these datasets. This paper studies the feasibility of ethically auditing large datasets through small samples and proposes a practical method for identifying structural biases by applying AI Ethics Tools (AIETs) to random samples. We applied Datasheets for Datasets, Data Statements, Data Nutrition Labels, and Data Cards to a 1% sample of Framed Multi30K, and Datasheets for Datasets to two 0.00006% samples of relaion2B-en-research-safe. Our results show that evaluating small samples with AIETs can reveal ethical problems, structural biases, and technical inconsistencies. Although industry practice assumes that identifying problems requires analyzing the entire dataset, we show that a random subset supports a viable qualitative analysis without overwhelming computational or human resources. Demonstrating that a dataset is biased does not require examining every instance: if a bias appears in a random sample, it is a flaw of the creation process. The AIETs' questions acted as provocateurs, evoking different perspectives and analysis methods for the same dataset, but the analysis and its results depended fundamentally on the human factor. We hope these findings contribute to urgent public discussions about ethics, transparency, accountability, and AI regulation by offering a practical pathway for responsibly auditing massive datasets.
en
The Limits of Information-Flow Control Enforcement for Agentic Systems
Information-flow control (IFC) is a promising approach to enforce policy in agentic AI systems. However, IFC suffers from label creep, a phenomenon where monotonically increasing security labels propagate through a system and progressively limit its utility. Label creep in agentic AI systems can be mitigated using two major strategies: (1) improved system design and (2) label declassification. In this paper, we characterize how far system design alone can mitigate label creep in agentic systems. Specifically, we measure how system topology, scale, and agent composition affect utility loss and security enforcement cost proxies for conservative IFC enforcement in a stateful agentic system simulator. We develop a pipeline to systematically vary these features and measure their impacts on label propagation. Our results demonstrate strong associations between structural graph features and our IFC utility loss metric (up to a Spearman correlation |ρ| of 0.96) and show that system size and tool allocation substantially affect how labels propagate. These findings identify design choices to reduce IFC utility loss, but they also demonstrate that system design alone cannot fully prevent utility loss due to monotonic label propagation. This motivates the development of label declassification mechanisms that aid these design choices and diminish IFC utility cost.
en
Finite Time Analysis of Risk-Sensitive RL via Noisy Power Iteration
Estimating the principal eigenpair of a positive operator from stochastic evaluations gives rise to a noisy form of normalized power iteration: each step applies a sampled operator estimate and then renormalizes. We establish finite-time stochastic-approximation guarantees for this procedure, working in Hilbert's projective metric, which is well-suited to the multiplicative geometry of Perron--Frobenius eigenproblems. Our main application is risk-sensitive average-cost reinforcement learning (RL) with exponential utility. In this setting the Bellman equations are multiplicative, and value functions are characterized by nonlinear eigenvalue problems rather than additive fixed points. We cast both policy evaluation and control in this framework, obtaining risk-sensitive TD- and Q-learning algorithms that learn from a single Markovian trajectory. Under explicit positivity, mixing, and linear function approximation assumptions, we prove finite-time bounds on the recovered eigenvector and eigenvalue with {O}(ε^{-2}) sample complexity up to problem-dependent constants. The results cover both tabular and linear function approximation regimes, and through the duality between exponential utility and KL-robust control yield finite-time guarantees for KL-robust average-cost RL as well.
en
FEROM: Frontier Endogenous Reveal-Order Marginal Policy Optimization for Masked Diffusion LMs
Masked diffusion language models generate text by iteratively unmasking positions, with practical samplers often selecting reveal positions from the model's own logits. The reveal order is therefore an endogenous latent variable of the rollout process with an enormous discrete latent space. Policy optimization objectives relying on pre-defined schedules or sampled trajectories introduce either bias or high variance. In this paper, we propose FEROM, Frontier Endogenous Reveal-Order Marginal Policy Optimization, which approximates policy gradients for the rollout-induced marginal response policy of masked diffusion LMs. FEROM derives a Rao-Blackwellized policy-gradient identity over latent reveal orders and expresses the resulting estimator as posterior edge occupancy on a reveal-state DAG. To make marginalization practical, we introduce Frontier Reveal Marginalization, a budgeted estimator that approximates this marginal edge structure by combining CDF-based computation of local probabilities of the endogenous reveal process with frontier expansion of high-mass partial states. Integrated into a GRPO-style objective, FEROM replaces single-path log-scores with a detached token-score surrogate over posterior-relevant reveal edges. Experiments on math and coding tasks show comparable or improved results over existing methods under matched scoring-state budgets. An offline proxy study further shows improved marginal-gradient approximation with increased scoring budgets.
en
CPNav: Calibrated Vision-Language Navigation for Object-Goal Search in Unseen Environments
Directing a robot to locate a target object in unseen environments remains a central challenge in embodied AI. Vision-Language Model (VLM)-based methods provide rich semantic reasoning but often produce uncalibrated and overconfident action decisions. We introduce CPNav, a navigation framework that combines collision-aware planning with statistically calibrated VLM action scoring for map-free object-goal navigation. From RGB-D input, CPNav constructs a navigability map, generates collision-free motion primitives, and queries two VLMs for goal detection and action scoring. The resulting scores are calibrated offline via conformal prediction (CP) using oracle actions obtained through a tree-structured exploration procedure with backtracking, providing finite-sample guarantees at a user-specified risk level. The CP layer is model-agnostic and can wrap any VLM policy. Online, calibrated VLM scores guide action selection, while a tree-structured exploration history records alternative actions and allows backtracking. We evaluate CPNav in both simulation and real-world hardware settings. In AI Habitat simulation, compared with the uncalibrated baseline method, CPNav achieves relative SR and SPL improvements of 6.2% and 11.4% on HM3D v0.2, and 12.4% and 8.8% on MP3D, respectively. In real-world deployment on a modified Hiwonder MentorPi robot, the corresponding improvements in SR and SPL are 24.9% and 15.2%, respectively. Code is available at https://github.com/practice-lab-ucla/CPnav.
en
On Agentic Manipulation Policies
Recent work on robot manipulation has focused on specialist policies trained on demonstrations. We show that generalist AI models can control manipulators more reliably than specialist policies such as vision-language-action models (VLAs) and world-action models (WAMs). We systematically analyze agentic manipulation policies (AMPs), in which a generalist AI model commands the manipulator directly, adapting its actions as the task unfolds. We present AMP1, a reference harness that runs unchanged across single-arm, bimanual, and mobile manipulation. AMP1 achieves state-of-the-art results on the RoboLab-120 and CaP-Gym benchmarks, with no demonstrations or task-specific training. On RoboLab-120, AMP1 reaches an 85.6\% success rate zero-shot, compared to 49.2\% and 42.9\% for leading VLA and WAM policies. The agent succeeds zero-shot in 104 of 108 real-world trials spanning twelve tasks and three robot instances, reducing failure rates by 20x and 15x relative to strong VLA and WAM baselines. The basic harness supplies no grasping, pushing, or placing skills; the agent develops its own strategies, such as searching for hidden objects, using objects as tools, and calibrating its cameras. With the addition of a persistent memory store, the agent can also learn from experience: an augmented harness with a memory representation that is carried across episodes lowers the failure rate on RoboDojo bimanual manipulation tasks by 4.6x relative to zero-shot deployment. Our analysis suggests that AMPs are substantially more general and robust than specialist manipulation policies, providing a promising foundation for robot manipulation.
en
Grasp-TAGS: Task-Aware Grasp Selection via Critic Distillation for Prehensile Loco-Manipulation
Loco-manipulation, which requires mobile manipulators to coordinate locomotion and manipulation over long horizons, is a key capability for large-object rearrangement in applications such as service robotics and warehouse automation. Despite recent progress, reliably executing these tasks remains challenging because performance depends on complex robot-object and object-environment interaction dynamics. Existing approaches address this challenge either through model-based methods, which often require accurate dynamics models, or through learning-based methods, which reduce this requirement but typically assume a fixed or predefined initial grasp. However, different geometrically stable grasps can lead to substantially different downstream manipulation performance. Therefore, we propose Grasp-TAGS, a task-aware grasp selection framework for goal-conditioned large-object rearrangement. We train a unified PPO-based reinforcement learning policy across multiple initial grasp configurations and distill the learned critic into a lightweight grasp utility network that ranks candidate grasps using only the candidate grasp configuration and goal information. At inference time, we select the grasp with the highest predicted downstream utility for the downstream policy to complete the task, before even making a contact. Experiments demonstrate that Grasp-TAGS improves rearrangement success over both fixed single-grasp policies and random grasp selection, achieving an average success rate of 89.84% compared with 79.87% for the best-performing single-grasp policies, an improvement of 9.97 percentage points. Grasp-TAGS also reduces final orientation errors across all evaluated objects. For more qualitative results, refer to the supplementary webpage: https://rpm-lab-umn.github.io/Grasp-TAGS-web-page/
en
Flow-Guided Target-Space Alignment via Path Consistency
Many multimodal alignment problems require mapping a source modality into a fixed space defined by a target modality, where the target space is not merely an intermediate embedding but is used directly for downstream analysis. In this setting, successful alignment should preserve both the global structure of the target distribution and the instance-level correspondence observed in paired source--target training data. Existing objectives typically address only part of this goal: contrastive losses encourage relative similarity, distribution-matching losses improve marginal agreement, and pointwise regression fits each source sample to its paired target without explicitly modeling target-space path geometry. We propose FlowAlign, a flow-guided target-space alignment method that repurposes conditional flow matching as a training-time alignment signal rather than an inference-time transport model. FlowAlign uses the learned flow field to guide the alignment of source-side representations with the fixed target space, encouraging them to preserve paired correspondence while remaining compatible with the global target-space structure. We further provide theoretical analysis showing how FlowAlign helps improve paired correspondence by controlling endpoint accuracy and capturing local target-space geometry. Experiments on synthetic data, paired single-cell multiome benchmarks, and image--text alignment demonstrate that FlowAlign improves instance level correspondence while preserving target-space structure across different modality settings.
en
Accelerating Reinforcement Learning Weight Refit with the Reshard Collective
During reinforcement learning (RL) post-training, training trajectories are generated on the fly by running inference with the current policy model. For...
en
SageMaker Inference now supports Multi-Instance GPU to maximize GPU utilization for model endpoints
Amazon SageMaker Inference now supports GPU partitioning with NVIDIA Multi-Instance GPU (MIG), which divides one GPU into hardware-isolated partitions so you can run a different model in each or take a right-sized fraction of a newer accelerator. This post shows how two optional API fields place one model per partition, and what we measured against a whole older GPU.
en