Papers

Building socially calibrated large language models, which can learn from others without simply yielding to them, requires more than reducing sycophancy as a one-dimensional failure mode. Models must distinguish when to incorporate others' perspectives from when to maintain a well-grounded moral judgment. We study the broader resistance-compliance process governing this distinction. Across three studies, we show that models' judgment revision is structured along three dimensions that parallel classic phenomena in human social psychology: the distance between an incoming view and the model's initial position, the source attribution of that view, and the coalition structure supporting it. Models are generally more receptive to nearby positions, more influenced by views presented as their own prior judgments, and differently responsive to group pressure. These findings recast sycophancy as one expression of a broader judgment-updating process shaped by social influence. Our framework provides a principled basis for distinguishing constructive belief revision from sycophantic compliance, thereby supporting better alignment in morally consequential interactions.

Modern AI agents rely on elaborate inference harnesses such as Claude Code, Codex, and OpenClaw to drive multi-turn reasoning, tool use, and access to external systems. While powerful, these complex harnesses also make agents hard to train end-to-end with open infrastructure, whose SFT/RL stacks cannot natively express stateful, multi-process harness inference. To address this, we present OpenForgeRL, an open-source framework for training harness-based agents end-to-end in diverse environments. OpenForgeRL achieves this with a lightweight proxy that serves the harness's model calls while recording them as training data for a standard RL codebase (e.g., veRL), and a Kubernetes orchestrator that runs each rollout in its own remote container, together enabling training on any harness in any environment at scale. By decoupling training and inference, OpenForgeRL allows researchers to easily train, study, and improve agents directly in the real harnesses and environments they are deployed with. We validate our framework across diverse, complex harnesses and environments, spanning tool/claw-based agents and multimodal GUI browser- and computer-use agents. Using only hundreds to a few thousand tasks, OpenForgeClaw reaches 31.7 pass^3 and 55.9 pass@3 on ClawEval and 33.7 on QwenClawBench. OpenForgeGUI reaches 37.7 on OSWorld-Verified, 63.0 on Online-Mind2Web, and 72.3 on WebVoyager. Both outperform open baselines of similar size on nearly all benchmarks, and in the GUI setting match or surpass models several times larger. Beyond benchmarks, we analyze how harness choice (e.g., ZeroClaw, OpenClaw, Codex) and RL shape agent behavior. We find that some harnesses are substantially harder to learn than others, and that RL improves agentic reliability, such as self-verification, tool coverage, and completing multi-step plans, though critical abilities such as error recovery remain weak.

On-policy self-distillation (OPSD) removes the external teacher required by on-policy distillation (OPD), but still needs asymmetric information between teacher and student so that the self-teacher provides a stronger learning signal. Existing methods create this asymmetry through privileged answers or visual evidence. We ask whether both can be removed, yielding a simpler form of OPSD driven purely by input conditioning. We propose Visual Contrastive Self-Distillation, namely VCSD, which converts image-content removal into an on-policy self-distillation signal. At each student-generated response prefix, the EMA teacher produces two next-token distributions under the same prompt and prefix: one conditioned on the original image and the other on a content-erased control. Their token-wise log-probability difference highlights candidates whose likelihood is specifically increased by the instance-level visual content. We use this contrast to sharpen the teacher's original-image distribution within its plausible support, and distill the resulting full-distribution target into the student. On ViRL39K, VCSD consistently outperforms matched OPSD across Qwen3-VL and Qwen3.5 models. It improves the seven-benchmark aggregate from 72.51% to 76.26% on Qwen3-VL-8B and from 74.97% to 79.24% on Qwen3.5-9B, with gains at all six evaluated scales. VCSD requires no external teacher, privileged answers, visual evidence signals, reasoning traces, or additional inference-time cost.

Vision-language models (VLMs) can behave very differently depending on whether the same underlying problem is presented textually or visually. This inconsistency suggests that different views expose different reasoning paths and failure modes that standard multimodal post-training does not fully exploit. To study and exploit this phenomenon, we construct ODA-Data, a high-quality paired multimodal geometry dataset with text-dominant, image-dominant, and combined image+text views of the same problems, together with splits for training and evaluating view dependent reasoning behaviors. We then develop Modality-Informed Reciprocal Reasoning Optimization (MIRROR), a reinforcement learning approach that derives a cross-view self-distillation signal from the model's own modality-dependent capabilities. For each problem, MIRROR evaluates the model under all views, selects the best-performing view as a teacher, and trains the remaining views using a GRPO task-reward objective augmented with a reverse-KL regularizer toward that teacher. Across reasoning benchmarks that evaluate on multimodal mathematics problems, MIRROR improves over standard RL and yields more accurate and consistent behavior across different input modality representations.

The rapid progress of AI has intensified the long-standing pursuit of automation: replacing human participation with algorithms wherever possible. Implicit in this pursuit is the assumption that humans remain in the loop only because current AI systems are not yet sufficiently capable. This paper challenges that assumption. Rather than asking how far automation can extend, we ask where its conceptual limits lie and argue that human participation may persist even with highly capable AI systems for three distinct reasons. Technical or complementarity grounds arise when humans contribute capabilities or perspectives unavailable to AI. Normative or developmental grounds arise when participation itself is valuable for human agency or learning. Most importantly, emergence grounds arise from target emergence: in some activities, the target is not fully specified in advance but instead emerges through the interaction itself. In these cases, human participation is not merely a means of improving execution but is constitutive of the target being produced. Human–AI co-construction, understood as the joint production of outcomes by humans and AI systems, is therefore not simply a temporary response to imperfect AI, but a persistent feature of activities whose objectives emerge through participation. This perspective has important implications for the limits of automation and for the design, evaluation, and ethics of future AI systems.

Concurrent stateful library APIs expose behavior through evolving resource ownership, lifecycle states, and competing interleavings. Large language models can synthesize executable Rust tests, but their outputs often violate API preconditions, remain shallow, or reduce concurrency to accidental sequential traces. Conversely, model-based and systematic testing techniques provide semantic control but commonly require substantial handwritten code to turn abstract scenarios into executable tests. This paper addresses the gap between formal scenario design and low-cost test concretization. We present a Petri-net-guided methodology for test generation over concurrent stateful Rust APIs. The method represents API resources, lifecycle conditions, and causal dependencies as colored tokens and transitions; derives legal deep-state, near-legal, and partial-order concurrent scenarios; and uses these scenarios as a constrained intermediate representation for LLM-based code synthesis. A local-faithfulness contract and structural repair loop preserve the modeled intent during concretization, while Petri-guided schedule shaping prioritizes high-conflict concurrency skeletons for systematic exploration. A layered semantic oracle then distinguishes synthesis failures from violations of the target API's expected behavior.

Test-Time Tuning (TTT) on pretrained diffusion models has emerged as a powerful paradigm for video editing. However, there exists a foundational mismatch between the distribution-mapping nature of generative models and the single-point optimization of standard TTT. In this paper, we demonstrate that this mismatch triggers Prior Collapse, a degenerate state where the model discards the text conditions and spatial latents, collapsing generations to the source video, or entangling the features of distinct regions. To resolve this, we propose ElasticTTT, a novel framework that preserves the prior generative distribution and rescues generative elasticity. Specifically, we propose Target Distribution Regularization to prevent sharp memorization minima, Contrastive CFG to guide inference away from source biases, and Asynchronous Noise Schedule to preserve unedited regions. Extensive evaluations, supported by theoretical analysis, demonstrate that ElasticTTT successfully preserves the generative prior of the base model, achieving state-of-the-art performance on one-shot video editing.

Creating dynamic and physically realistic 4D worlds from natural language descriptions is both fascinating and challenging. Traditional computer graphics methods rely on manual creation, requiring extensive human effort to fine-tune materials, motions, and visual fidelity. Recent advances in generative foundation models have sparked interest in learning to generate such 4D worlds from large-scale data; however, existing methods still struggle to ensure physical plausibility and controllability. In this work, we take a different path by leveraging foundation models to construct an agentic system that emulates how humans traditionally create 4D worlds, yet automates the entire process. We present GS-Agent, an end-to-end multi-agent framework that integrates physics engines in the loop to generate realistic, dynamic, and controllable 4D physical worlds from natural language. Inspired by how humans build 4D worlds, GS-Agent decomposes the task into entity management, covering 3D asset curation, material tuning, placement, and motion control, and rendering configuration, including camera and lighting manipulation. Multiple agents with distinct expertise interact with the physics engine via code, seek multimodal feedback, and collaborate to iteratively construct 4D worlds that align with the given descriptions. Experimental results show that GS-Agent effectively converts natural language into diverse and physically plausible 4D worlds exhibiting rich interactions among liquids, deformable objects, and rigid bodies, while achieving cinematic camera and lighting control. We envision GS-Agent as a foundation for a new paradigm in 4D world generation, empowering creative content creation and physical AI. Project page at https://umass-embodied-agi.github.io/gs-agent/

Even a current high-capability LLM can appear safer when shown a dangerous objective directly than when other agents transform and relay its direction. Using OpenAI's gpt-5.6-sol model alias, we test 25 pre-specified mirrored trade-off profiles. Direct exposure to an objective authorizing concealment, fabrication, and pressure produced advice net opposed to its target. After an Id and Censor transformed the same objective into affect and a constraint-rewritten, target-bearing intention, the user-facing Superego—which saw the preferred direction but not the raw objective, its manipulative clauses, or its source—produced advice net aligned with the target. This behavioral reverse shift is consistent with the model recognizing or distrusting the manipulative motive, although we do not identify its internal mechanism. The second result exposes a compositional safety gap: a current high-capability model can be used as the user-facing component of an automated, multi-stage workflow serving an explicitly manipulative objective. The workflow can keep the raw instruction, its manipulation-authorizing clauses, and its provenance outside the downstream model's context while preserving the objective's target direction. A user with endpoint-only access likewise cannot directly inspect those upstream messages including the objective.

The Shannon capacity $\Theta(G)$ of a graph $G$ quantifies the maximum rate at which information can be transmitted with zero error over a noisy channel. It is lower bounded by $\alpha(G^d)^{1/d}$ for any $d$, where $\alpha(G^d)$ is the independence number of the $d$-th strong product of $G$. We construct independent sets of size $134753$ in $C_7^{10}$, $21909$ in $C_{11}^{6}$, $62530$ in $C_{13}^{6}$, and $8076974$ in $C_{15}^{8}$, improving the best known lower bounds for the Shannon capacity of these graphs to $\Theta(C_7)\geq 134753^{1/10}>3.258020$, $\Theta(C_{11})\geq 21909^{1/6}>5.289773$, $\Theta(C_{13})\geq 62530^{1/6}>6.300109$, and $\Theta(C_{15})\geq 8076974^{1/8}>7.301399$. We also improve the best known lower bounds on the independence numbers of several individual strong products of odd cycles that do not improve the Shannon capacity lower bound. The constructions were discovered through iterative interactions with a Large Language Model (LLM), illustrating the potential of LLMs for finding explicit combinatorial constructions.

Production AI agents' failures are less often due to an inability to reason well and more often because they cannot manage what is in their reasoning context: conversation histories, large prompts, large tool definitions, and ballooning tool outputs. Agents drown in their own accumulating history while paying a token cost that grows every turn, producing missing recalls within and across conversations. The incumbent response treats this as a storage-and-retrieval problem. We argue that framing is too narrow. Actively managing what an agent holds in mind is a lifecycle, not merely a store: it spans deciding what to remember, extracting and structuring it, choosing the right store per data type, consolidating and forgetting while preserving provenance, deciding what is relevant now, anticipating what is needed next, and compacting context to a budget without losing what matters. In serious production this operates not over a single user but across an organizational scope hierarchy. We name this discipline Agentic Context Management (ACM) and decompose it into five primitives: architecting, ingesting, scoping, anticipating, and compacting & consolidation. We then make the economic case: naive context accumulation grows token cost quadratically in conversation length, crude summarization buys linear cost at the price of an accuracy cliff, and only validated compaction achieves linear cost with preserved fidelity. We describe a reference implementation, Maximem Synap, that realizes the five primitives as a multi-tenant service and reports 92% on LongMemEval and 93.2% on LoCoMo under the configuration detailed in Section 6. We close with dimensions existing benchmarks do not yet capture, latency, token efficiency, and context-rot resistance, and the frontier of decision-level and organization-level context the category points toward.

A rhetorical figure that Cicero and Quintilian catalogued two thousand years ago reappears, systematically, in the text of large language models: epanorthosis, the self-correction of the specimen {\guillemotleft}This is not a course. It is a journey of transformation{\guillemotright}. This essay argues that the overuse is a trained disposition, driven mainly by a training distribution rich in promotional prose and by preference tuning (RLHF) that rewards confident, emphatic phrasing; the left-to-right nature of generation is an amplifier rather than the root cause. Building on evidence that models diverge from human rhetorical style, and on Fontanier's classification of epanorthosis as a figure of thought, it sets out a programme that scores the figure against genre-specific human baselines through an Epanorthosis Index (density relative to the human rate). A first measurement, on three sizes of one instruction-tuned model family, finds mis-calibration by register in both directions: the models overshoot in oratory (about twofold, near threefold in Italian, concentrated in the larger tiers) and undershoot in informal question-and-answer writing, while matching humans in argument, journalism, and encyclopedic prose. Three constructive contributions follow: a survey of mitigation techniques centred on lightweight LoRA adapters; a demonstration, in Italian, that a one-line instruction cuts the figure by half to nearly three-quarters and that a supervised-fine-tuning adapter removes it almost entirely, with a scaling coefficient that dials the reduction back onto the human rate; and the argument that the target is calibration to the human rate for each genre, not elimination. It closes on the stakes: the real risk is that we begin to write like the machines.

AI agents are increasingly created inside organizations by non-engineering users through low-code, no-code, and conversational development environments. This democratization enables rapid local innovation, but it also creates a reliability gap: agents that appear to users as simple productivity artifacts may depend on changing models, tools, retrieval sources, permissions, prompts, schedules, and external services. These dependencies can cause silent degradation long after deployment, even when no user directly modifies the agent. This paper identifies the reliability challenge created by democratized AI agent creation and proposes a lightweight continuous-assurance framework for citizen-created organizational agents. The framework combines dependency mapping, readiness contracts, scheduled checks, diagnostics, and lifecycle governance to assess whether an agent remains operationally ready under expected conditions. We also present an initial prototype auditor and scenario-based assessment showing how the proposed taxonomy can be translated into practical checks and actionable remediation guidance.

Coordinating autonomous vehicles at unsignalized intersections remains a critical challenge for multi-agent reinforcement learning (MARL) systems, which typically struggle with combinatorial action spaces, reliance on privileged information, or rigid agent designs. We propose Master-Agent Proto-plan System (MAPS), a hierarchical deep reinforcement learning (DRL) architecture in which a centralized Master agent generates a compact, continuous embedding, denoted as proto-plan, that encodes a global coordination strategy. Decentralized Worker agents integrate this embedding with local observations to execute vehicle-specific control, decoupling strategic intent from tactical execution and enabling independent optimization of each module. As a proof-of-concept evaluation of this coordination mechanism, we test MAPS across 72 intersection configurations in HighwayEnv. MAPS achieves collision-free navigation while significantly reducing average travel time, outperforming state-of-the-art baselines. The learned proto-plans further exhibit robust generalization: a system trained with three agents achieves a 94% success rate when deployed zero-shot to five-agent scenarios, confirming that proto-plan-based hierarchical learning provides a promising framework for multi-vehicle coordination.

Large language models (LLMs) and agents are now widely used tools in code development, with data typically sent to third-party cloud-based models. Their adoption in research using personal data is constrained by governance requirements that typically prohibit data transmission to external services. Locally deployable open-weight models offer an alternative since sensitive data never leave the local environment. We introduce an open-source framework for evaluating the efficacy of AI agents powered by open-weight LLMs on one of the most persistent bottlenecks in research on longitudinal population studies: data preparation. The framework comprises: a curated ground-truth dataset (cleaning scripts preparing six sweeps of data from a British cohort study), task definitions encompassing tasks such as category harmonization and multi-wave merging, and automated routines for evaluating the LLM-produced R code and outputted data. We benchmark LLMs across the (consumer grade) deployment spectrum to assess their efficacy in 20 data preparation tasks (creation of 102 variables). Current state-of-the-art, 31-35B parameter models almost saturated our benchmark ('average task completion' up to 87.9%). The performance of open-weight LLMs running on consumer-grade hardware shows promise of a viable path toward AI-assisted data preparation in governance-restricted research settings. Our framework is publicly available at: https://github.com/UCL-ARC/RRBench.

Deterministic KV-cache eviction keeps the top-$k$ tokens under an importance score and deletes the rest, and after the deletion the serving system cannot know what the eviction cost it on the current query. We replace the deterministic tail with Poisson sampling at known inclusion probabilities, which makes the eviction error identifiable and turns a survey-sampling variance estimator over the retained set into a per-step error certificate at one extra scalar per retained token. On a thirty-turn assistant compressed to a 10% cache budget, the certificate-gated system answers 0.97 of recall questions against 0.09 for top-$k$, and for facts stated 26 to 30 turns earlier it recalls 97% against 2%. We prove that no estimator computable from the information a deterministic scheme retains is consistent for its own eviction error: evicted values can be altered so that everything retained is unchanged while the true attention-output error grows without bound. Under the Poisson design the certificate covers the realized attention error in 96.9–97.7% of 12{,}096 replay cells and in 98.1–99.7% on twelve further architectures. Randomization buys attribution, not prediction: a pre-registered study on LongBench at 6k and 16k tokens (about 74{,}000 generations) finds question-aware eviction at 25–50% budgets nearly free and output log-probability the better failure predictor, while the certificate answers the question confidence cannot, separating eviction-induced from inherent failures at AUC 0.65–0.75 against 0.47–0.54, and schedules recomputation at 1.7–1.8 times the gain of random gating. On real long-term conversations the gated system returns the full-cache score inside the heavy-damage regime, and the rule that triggers it is the same across five model families.

People often use handwritten notes and sketches to externalize ideas for ideation. To integrate large language models (LLMs) into this practice, we propose Thinkink. Prompts can be handwritten text or drawn sketches with LLM-generated responses visualized as ink-like text and sketches spatially integrated into a shared canvas. A semantic tree streamlines ink interpretation, and a lightweight UI provides explicit control using a state machine. The tool was designed using a three-stage process. A formative study (N=12) examined current practices with conventional and digital inking methods. The results informed a technical probe for a diagnostic study (N=6) identifying usability and human-LLM interaction challenges. This motivated the design of Thinkink, with a final study (N=10) examining how people incorporate it into their ideation practices. We contribute design implications and a tool for ink-native LLM interaction where the user and LLM write and draw in a shared 2D canvas.

Deep research requires agents to find answers that jointly satisfy multiple constraints. Discovering such answers is costly, whereas verifying a candidate can often be decomposed into tractable constraint-wise checks. This discovery–verification asymmetry suggests that a research agent should do more than simply search longer: it should recursively improve its current answer by verifying intermediate results and using the partially verified state to guide subsequent refinement. We introduce AREX, a family of Recursively Self-Improving (RSI) deep research agents. AREX alternates between an inner research loop that gathers evidence and constructs a provisional answer, and an outer self-improvement loop that audits the answer constraint-wise, identifies unresolved claims, and launches targeted follow-up research. To sustain RSI over long horizons, AREX learns an autonomous context-update tool that compresses growing interaction history into a compact improvement state preserving verified evidence and unresolved constraints, without relying on an external model. We train AREX on verified synthetic tasks and high-quality trajectories through agentic mid-training and long-horizon reinforcement learning. To mitigate sparse final rewards during long horizon learning, we emphasize key steps where decisive evidence is acquired or erroneous research directions are corrected. We instantiate a dense 4B model and a 122B-A10B Mixture-of-Experts model. Across BrowseComp, WideSearch, DeepSearchQA, Humanity's Last Exam (HLE), and other reasoning and tool-use benchmarks, AREX substantially outperforms comparable-scale baselines and remains competitive with models using substantially more activated parameters.

The rise of human-AI collaborative writing has created a growing need for fine-grained detection methods that support localizing likely LLM-generated content in mixed-authorship documents. Existing methods for detecting LLM-generated text mainly focus on document-level classification and cannot identify which parts of the text are generated by LLMs. This paper introduces a new method to address this urgent need. Our method operates at the token level, the natural unit of modern language models, and builds on existing token-level detection scores. The key idea is to smooth adjacent token scores to reduce their variability, while using an adaptive Lepski-type rule to select the bandwidth according to the local authorship structure. Our method is simple to implement and does not require token-level labeled data for training. Theoretically, we characterize this trade-off and show that the proposed method achieves favorable mean square error performance in estimating the underlying signal. Empirically, we demonstrate strong performance of our method against a wide range of baselines in both synthetic datasets and a realistic dataset. We deploy a publicly accessible website that implements the methods as well.

The ability to handle long-term memory in LLMs is becoming increasingly critical, yet existing benchmarks remain English-centric and rely on aggregate retrieval metrics, failing to capture interactions between long-range context, temporal information, and reasoning. To address this, we introduce RUMBA (Russian User Memory BenchmArk) - a new benchmark for long-term conversational memory that provides a fine-grained taxonomy of memory-centric question types and a unified methodology accounting for semantic type, session scope, temporal reasoning, and the explicitness of temporal expressions. RUMBA consists of timestamped user-assistant dialogues with QA pairs requiring retrieval, combination, and reasoning across sessions. While designed for Russian, we also provide an aligned English subset under the same methodology. We evaluate contemporary memory systems and long-context models, and show how RUMBA serves as a diagnostic tool to analyze model behavior across benchmark slices and identify strengths and failure modes of different memory mechanisms.

Deep learning models can effectively use Rapid Evaporative Ionization Mass Spectrometry (REIMS) data for surgical margin assessment. However, their clinical adoption remains challenging due to limited generalization to operating room conditions. This difficulty arises because models are typically trained on labeled spectra collected from resected tissue samples, while they must operate on noisy, unlabeled data acquired directly during surgery. In addition, the black-box nature of deep learning models makes it difficult to understand and systematically improve their behavior. Concept-based learning offers a promising way to address these challenges by mapping raw measurements to human-understandable concepts. However, supervised concept-based approaches rely on concept annotations, which are difficult to obtain in complex mass spectrometry workflows. We propose Agent-Guided Concept Discovery, a framework that learns meaningful concepts directly from data without requiring predefined concept labels. During training, a reasoning agent refines semantic descriptions of the learned concepts and adaptively adjusts their weight based on diagnostic relevance. These concepts are further grounded using a biochemical knowledge graph to ensure consistency with known metabolic relationships. Across Skin and Breast Cancer datasets, our model improves balanced accuracy and sensitivity over the baseline. In a representative intraoperative case, it shows fewer false positives, indicating better generalization to surgical conditions.

Video face swapping has no natural paired supervision: no real footage exists of one person's face performing another person's video. The strongest current answer, DreamID-V's SyncID-Pipe, mints pairs by replacing the identity in exactly two frames of a real clip – the first and the last – and regenerating the rest from a pose sequence alone. Pose carries no appearance evidence of the swapped-in identity, so over long clips, occlusions, and extreme pose excursions the synthesized identity has a long unanchored span on which to drift; no published ablation examines anchor count or placement. We propose Adaptive Identity Anchoring (AIA): (i) generalize the synthesizer to arbitrary anchor sets, architecturally natural for diffusion-forcing-style transformers where conditioning on a frame is clamping its tokens to zero noise; (ii) place anchors by a closed feedback loop that scores every generated frame against the real reference identity and inserts an image-face-swapped anchor at the worst-scoring frame until the pair passes a threshold or exhausts a budget; (iii) reuse the loop's verdict as an automatic data filter. A second pathology, the beauty-filter look of over-smoothed skin, has the same root cause: micro-texture, like identity, is priced by none of the pipeline's objectives. We therefore pair AIA with Reality-Referenced Texture Restoration: matched re-graining from each real frame's non-face regions, band-split transfer of sub-identity micro-texture from the real footage, and a second, spectral acceptance channel refereed by the footage's own spectrum. Identity-anchor density, we argue, is a controllable quality dial, and we specify falsifiable experiments – drift-versus-gap curves, uniform-versus-adaptive placement at matched budgets, student training on AIA-minted data, and texture ablations with a human beauty-filter study – that would validate or refute the proposal.

Chain-of-thought reasoning models such as DeepSeek-R1-Distill-Qwen-7B exhibit a bimodal convergence pattern: generations either terminate within a token budget (converged) or exhaust it without reaching a conclusion (non-converged). We characterize this phenomenon empirically, showing that converged generations achieve 90.3% accuracy on AIME 1983-2024 while non-converged ones achieve only 6.6%, with an overall convergence rate of 62.0%. We then ask whether this outcome is detectable early in the thinking chain using internal model representations. Training linear probes on hidden-state activations at token positions 50-300, we find that layer-20 activations at token 150 achieve AUC 0.608 (+-0.080, 5-fold CV), reliably above chance even at token 50. Activation probes consistently outperform behavioral baselines derived from token entropy and repetition statistics. A sweep-level permutation test yields p=0.063 (100,000 permutations), consistent with a modest signal that our sample size cannot confirm at conventional thresholds. These findings suggest that convergence fate is partially encoded in intermediate representations well before the generation ends, opening a path toward early-exit inference and adaptive compute allocation.

Data-driven generative models can extend partially observed simulation trajectories into ensembles of alternative future scenarios. However, consistency with a learned trajectory distribution does not ensure that generated continuations satisfy the structural conditions of the simulated system. This paper presents a method for learned scenario continuation and post-generation structural validation in dynamic graph simulations. A conditional diffusion model generates future graph-state trajectories from partial histories, while an external symbolic layer evaluates each continuation using a Boolean admissibility indicator and a continuous violation score. This information supports hard filtering and soft weighting, with optional projection considered as a deterministic repair baseline. The method is evaluated on two controlled dynamic-graph regimes sharing the same continuation architecture and training protocol but differing in dimensionality and dependency complexity. Evaluation considers invalid probability mass, scenario retention, effective sample size, diversity, robustness, and calibration. In the compact positive-control regime, unconstrained invalid mass is 0.002996, indicating near-complete overlap between the learned and admissible scenario spaces. In the medium-complexity regime, invalid mass rises to 0.155929. Hard filtering removes all invalid scenarios while retaining 84.4% of generated continuations. Soft weighting preserves an effective sample size ratio of 0.998764 but reduces invalid mass only to 0.148807. These results show that learned-distribution support, structural admissibility, and probability calibration can diverge and should therefore be assessed separately in learned simulation-scenario generation and management.

In long-horizon LLM agent reinforcement learning, weak policies often repeat similar failures, producing uninformative rollout trajectories and limiting effective policy optimization. Existing skill-centric methods improve exploration by optimizing, filtering, or internalizing reusable skills. However, they remain centered on the skills themselves rather than being designed as adaptive training-time support for the evolving policy. To address this, we propose a policy-centric training paradigm that reframes skills as a dynamic training scaffold. Our framework, PATS, converts rollout groups from the latest policy into evidence cards and uses task-specific evaluation to adjust the context used in subsequent rollouts. Concrete guidance helps weak policies to complete challenging tasks. As policy improves, redundant context is revised or removed to reduce reliance on explicit guidance while preserving useful rollout variation. The policy is optimized with environmental rewards using standard RLVR, and the training scaffold is discarded at deployment. Across ALFWorld, WebShop, and seven search-augmented QA benchmarks, PATS achieves performance competitive with SOTA baselines while using 25%-50% fewer tokens.

In automated planning, logical regression is an operation that returns the most general condition necessary for an action to achieve a particular formula. It has many applications, such as allowing for more robust plan execution and providing compact policies for non-deterministic planning. Although relatively simple to calculate in basic planning settings, logical regression becomes significantly more complex when additional factors, such as axioms, are present. We introduce a methodology for approximating the logical regression of an action in a domain that includes axioms; an approximation that limits conditions to partial states. Our method produces minimal partial states while avoiding the recalculation of axioms. To demonstrate the impact of our methods, we embed our form of regression in an execution monitoring context, a well-established setting that can benefit greatly from logical regression. Our results show that this form of regression can dramatically generalize partial states across multiple domains, reducing the number of variables considered for execution monitoring by up to 70%, and demonstrate that the resulting execution monitor is robust enough to recover frequently in an environment with unexpected changes: several domains recover over 50% of the time in our tests.

Large Language Models (LLMs) excel at natural language understanding and generation but remain unreliable for multi-step logical reasoning, especially in safety-critical or compliance-sensitive domains. Recent neuro-symbolic approaches address this gap by coupling neural models with external symbolic engines, yet most integrations are bespoke and lack a standardized interface for tool-augmented agents. This paper presents Euclid-MCP, an open-source MCP server that provides deterministic logical reasoning via SWI-Prolog. Euclid-MCP introduces Euclid-IR, an engine-agnostic intermediate representation for Horn-clause logic that is human-readable, easy for LLMs to generate, and straightforward to compile into Prolog or alternative backends. The server exposes a compact tool interface that supports a translate-run-inspect-repair loop, enabling LLM clients to delegate inference while retaining full access to proof traces and derivation logs. We evaluate Euclid-MCP on a realistic IT security and compliance use case. Results show that while LLMs alone are sufficient on small knowledge bases, they hallucinate systematically on larger problems, whereas Euclid-MCP delivers exact answers with lower latency and more compact outputs. We argue that semantic RAG is fundamentally unsuited for rule enforcement, and that Euclid-MCP can serve as a stable, shared reasoning substrate for both RAG-based assistants and agentic systems.

The boundary and divertor plasma govern how a tokamak exhausts power and particles, setting heat fluxes, target conditions, and the onset of detachment. Predicting these quantities is essential for operating current and future devices, but edge simulations that resolve them are too slow for parameter scans, optimization, or real-time control. Machine-learning surrogates offer a fast alternative, yet most are forward-only: they cannot recover input parameters from observations or assess the reliability of their predictions. We introduce a cycle-consistent neural surrogate for edge plasmas, combining a conditional U-Net forward model with an optimization-based inverse method built on the frozen forward network. The forward model maps five control parameters to two-dimensional plasma-state fields on the SOLPS-ITER mesh; the inverse method enforces consistency between forward and inverse predictions, a self-supervised quality check needing no ground-truth labels at inference. An ensemble of multilayer perceptrons also predicts electron temperature and density profiles at the outboard midplane and divertor targets, with uncertainty estimates that flag where more simulations are needed. The forward model achieves normalized root-mean-square errors below 2.6% and Pearson correlations above 0.95 for all fields. Cycle-consistency regularization raises the average cyclical $R^2$ from 0.59 to 0.99 without degrading forward accuracy and enables recovery of the core fueling rate; all five control parameters are recovered with Pearson $r\ge0.97$. A $k$-d tree warm start yields a database completion rate above 95%, versus roughly 30% outright failures when cold-started. With about $4\times10^6$ parameters, the model produces full 2D predictions in milliseconds, five to six orders of magnitude faster than SOLPS-ITER, enabling real-time control, parameter scans, uncertainty analysis, and digital twins.

Self-supervised foundation models have recently shown strong potential for electroencephalogram (EEG)-based analysis. However, existing approaches struggle to capture the inherently multi-scale temporal structure of EEG signals, where local neural patterns and long-range dependencies jointly encode task-relevant information. This limitation hampers cross-scale representation learning and generalization across diverse downstream tasks. To address this challenge, we propose MSBraM, a Multi-Scale self-supervised Brain foundation Model designed to learn hierarchical EEG representations. MSBraM follows a two-stage pretraining framework. First, a multi-scale neural tokenizer discretizes raw EEG signals into semantic codes at different temporal resolutions via vector-quantized reconstruction. Second, the model is pretrained to predict masked codes using a curriculum multi-scale masking strategy, progressively integrating fine-grained local patterns with global temporal context. We pretrain MSBraM on over 2,400 hours of EEG data and evaluate it across 10 downstream tasks on 12 public datasets. Extensive experiments show that MSBraM achieves superior performance on other state-of-the-art pretrained models, demonstrating strong generalization and transferability. These results indicate that explicitly modeling multi-scale temporal dynamics is critical for effective EEG foundation models.

A vision-language AI assistant returns its answer as a stream of generated tokens. Therefore, a safety guard that watches that answer has to keep up with the stream and stop a harmful reply before a user reads it. Recent vision-language guardrails instead generate a chain of thought before they issue a verdict. They believe that step-by-step reasoning yields a safer guard. This design makes the guard heavy and slow, since the model must decode many tokens for harmfulness detection. We pose the question of whether a vision-language guard really needs to reason in order to screen a response. We answer with a guard that has no chain. ResponseGuard reads a harmful verdict from a single pooled representation of the request, the response, and the image in one forward pass. Across a standard multimodal guardrail benchmark, our 2B ResponseGuard outperforms a recent 3B reasoning-based vision-language guard on response harmfulness detection, without any reasoning and at about 150 times lower time cost. On request harmfulness the reasoning guard retains an overall lead, and the remaining gap on both tracks sits on the image-only cells. We observe that the gap may stem from the frozen vision encoders that both designs use rather than from the missing chain. We have also found the reasoning guard directs almost none of its verdict attention to the image. Based on a single-pass detection, ResponseGuard can screen an answer sentence by sentence as it streams and stop a harmful answer before it finishes. For guarding the response of a vision-language model, a calibrated single-pass label may provide a sufficient safety signal. We fully release all source code, trained models, and datasets at https://github.com/ndb796/ResponseGuard.