Named after the hundred-eyed watchman of Greek myth, Argus watches the education landscape: spotting new opportunities, pressure-testing the ventures we're building, and tracing every read back to the real-world signals behind it.
The evidence library: the raw signals the pipeline is watching across the education ecosystem. Every idea is built from these.
arXiv:2606.30945v1 Announce Type: new Abstract: Recent studies argue that LLMs can predict human stereotypical judgments. Yet whether LLMs emulate the cognitive processes underlying human stereotypes, or merely retrieve learned associations to solve prediction tasks, remains unclear. Prior work examines LLMs' stereotypes in either (i) controlled judgment tasks like multiple choice surveys, or (ii) contexts constrained by conventionalized and predictable group biases. Here, we compare the structure of the stereotypes that humans and LLMs exhibit in the interpretation of free-form stimuli, namely abstract art and Rorschach blots, which lack pre-established cultural meanings. We recruit participants across five social domains (gender, partisanship, personality, urbanicity, and lifestyle) and elicit both first-order (direct personal interpretations) and second-order responses (predictions about how members of social groups will interpret the stimuli); we replicate this design with two mult
arXiv:2606.30908v1 Announce Type: new Abstract: Student access to Large Language Models (LLMs) is reshaping learning behaviors; at the same time students are entering the workforce where effective LLM use is becoming an expected skill. In this Experience Report we share our DURA framework (Demystify-Use-Reflect-Assess) and materials we used to restructure our CS2 course to allow the use of LLMs. We first demystified LLMs, then provided guidance on use with required attribution. We also added reflections related to LLM use at three points throughout the semester to encourage student meta-cognition around LLM use. We increased the value of proctored assessments in tandem with allowing retakes and including questions that explicitly assess skills from programming assignments. Students reported using LLMs for clarifying course concepts, debugging, understanding assignment guidelines, and determining test cases, but also still sought assistance via office hours and TAs, monitored Piazza, an
arXiv:2606.30905v1 Announce Type: new Abstract: Community Notes, a bridging-based crowd-sourced fact-checking system, has emerged as a new mechanism for moderating misleading information on social media and has been adopted by major platforms including X, Facebook, Instagram, Threads, and TikTok. Since its introduction, there has been an open question about what role AI could play in scaling and optimizing the system. Recently, X extended its Community Notes system by introducing Collaborative Notes: notes initially drafted by an LLM and iteratively refined based on feedback from human contributors. In this work, we systematically analyze the complete corpus of 19,146 collaborative notes and 211,850 instances of human feedback. First, we develop a taxonomy of human suggestions for improving AI-generated note drafts and find that suggestions involving factual corrections and additional context are most likely to be incorporated, while subjective policy judgments rarely are. Second, we e
arXiv:2606.30860v1 Announce Type: new Abstract: As large language models (LLMs) become common in computing courses, we need to understand how the social setting shapes how students use them. This paper reports findings from a semester-long study of 96 undergraduate students who completed six assignments, alternating between individual homework and team project milestones. We tracked LLM usage, prompting habits, and how students verified AI-generated output across all six assignments. LLM usage dropped by 42.7 percentage points when students moved from individual work to their first team milestone, then partly recovered in later team tasks. Students also wrote fewer and simpler prompts, used fewer intentional prompting strategies, and checked LLM output less carefully. The share of students who ran tests on AI-generated code fell by 19.4 percentage points during team assignments and never fully rebounded. A within-student analysis found that 18.9% of students who consistently used LLMs
arXiv:2606.30667v1 Announce Type: new Abstract: After decades in which the software industry heroized its technical employees, our current moment finds those employees in crisis. Economically, they are squeezed by a job market that has turned on them since Elon Musk gutted Twitter in 2022. Politically, their craft is continually pressed into the service of an ever tighter alliance between Big Tech and authoritarianism. Professionally, they find themselves less and less able to contribute anything good to anyone within business models that are running out of room to pretend they do anything but extract. Technically, they are confronted with a much-hyped technology, generative AI, that distorts their work while purporting to make them redundant. And emotionally, they are simply not ok - as designer and developer Andrew Sempere puts it, "I think the word I'm looking for is: bereft." This study draws from a series of practitioner interviews undertaken for a current dissertation in STS and
arXiv:2606.30666v1 Announce Type: new Abstract: A common AI-safety narrative holds that sufficiently capable agents will predictably seek power, resist shutdown, and therefore tend toward confrontation with humans. We argue that this conclusion is often drawn in an implicitly Earth-centered strategic landscape. If a credible off-Earth autonomy pathway exists - i.e., a staged transition from Earth dependence to an autonomous machine industrial base - then confrontation is not the only route to reducing human control. Using Saklakov's decision-theoretic 'confrontation question' as an anchor, we provide a qualitative mapping from the autonomy pathway to key model terms showing that early cooperation can dominate confrontation as a path to autonomy, and that the autonomy pathway can reduce confrontation incentives by making Earth less strategically binding. We discuss how this incentive shift interacts with feedback-loop dynamics between human preemption and agent behavior, and outline imp
arXiv:2606.30665v1 Announce Type: new Abstract: Stage B heart failure is characterized by asymptomatic structural or functional cardiac abnormalities. Identifying individuals at this stage is clinically important, as early detection may enable targeted interventions to prevent progression to symptomatic disease. Explainable artificial intelligence (XAI) may support early detection, transparent risk stratification, and selection of clinically actionable interventions. This review examines the use of XAI in detecting and characterizing stage B heart failure. A literature search of Web of Science, Scopus, and PubMed was conducted on 27 March 2026. Studies were included if they applied AI with XAI techniques to stage B heart failure. After screening, 20 studies were included. Data on modalities, outcomes, demographic reporting, and XAI methods were extracted and synthesized. SHAP was the most commonly used method, followed by LIME, saliency maps, and Grad-CAM; however, XAI adoption was inc
arXiv:2606.30663v1 Announce Type: new Abstract: The FAIR principles (Findable, Accessible, Interoperable, Reusable) have transformed research data management, but they do not address the environmental impact of creating and using research software and data, such as energy consumption, carbon emissions, and life-cycle impacts that become central to computer science and engineering-related domains. To bridge this gap FAIR+Sustainability or FAIR+S, an extension of the FAIR framework that embeds environmental accountability as a core element, was introduced. Because FAIR principles already structure how digital research artefacts are described, shared, and reused, they offer an effective entry point for embedding sustainability considerations at scale. FAIR+S weaves carbon-footprint and energy-use considerations directly into FAIR-aligned metadata schemas, workflows and development specifications. In doing so, it enables research infrastructures to report, compare, and audit the environmen
arXiv:2606.30662v1 Announce Type: new Abstract: The advent of Generative Artificial Intelligence (GenAI), and in particular Large Language Models (LLMs), is reshaping educational practice, while intensifying ethical debate about its adoption. To date, the dominant paradigm remains cloud-based and text-only chatbot: a centralized service that offers limited pedagogical control, weak transparency over knowledge sources, and non-trivial risks for privacy and regulatory compliance. This model also presumes continuous connectivity and recurring API costs, creating structural barriers for many institutions, reinforcing existing digital divides. At the same time, educational interaction with LLM can benefit from multimodal cues and embodied presence, requiring interfaces that move beyond text-only tutoring. In this work, we propose ELEVATE (Efficient LLM Education with Virtual Avatar Teaching Engine), a framework to develop efficient GenAI-driven avatar tutors governed by epistemic infrastruc
arXiv:2606.30661v1 Announce Type: new Abstract: Large language models (LLMs) increasingly mediate access to information, yet their responses are shaped by training-data curation, alignment procedures, provider policies, inference-time moderation, and jurisdictional regulation. This paper examines LLM censorship as a sociotechnical phenomenon that extends beyond explicit refusals to include omissions, selective emphasis, framing effects, and geographically variable content controls. We synthesize recent empirical studies, provider case studies, regulatory developments, auditing methods, and mitigation strategies to clarify how censorship-like behavior emerges across the model lifecycle. The analysis highlights the tension between safety and openness, the difficulty of measuring soft censorship, the geopolitical divergence of moderation regimes, and the need for transparent, contestable, and independently auditable governance mechanisms. We argue that the central challenge is not whether
arXiv:2606.30660v1 Announce Type: new Abstract: Collecting reliable social data from low-literacy populations remains a persistent challenge, particularly when surveys involve sensitive topics and marginalized communities. Traditional paper-based and web-based survey modalities often suffer from high attrition and incomplete responses due to literacy barriers, social pressure, and interactional discomfort. In this paper, we present findings from an initial field evaluation comparing multiple survey modalities paper-based interviews, digital web-based surveys, conversational AI (convAI) surveys, and convAI enhanced with layered value-sensitive design conducted with low-literacy women across India. Using data from 315 participants, we show that convAI significantly improves survey completion rates relative to traditional modalities, with the highest completion and lowest drop-off observed when value-sensitive and culturally aligned conversational design elements are fully integrated. The
arXiv:2606.30658v1 Announce Type: new Abstract: Medical AI has shifted from reasoning to agentic AI, a new paradigm that autonomously invokes external tools during reasoning, rendering intermediate reasoning steps and tool outputs transparent to users. Although proven to outperform previous models, physician trust in agentic AI remains largely unexplored. To address this, three physicians evaluated 315 multimodal clinical cases quantifying both process-oriented cognitive trust and outcome-oriented behavioral reliance. Comparing agentic AI against non-agentic baselines, physicians exhibited significantly higher cognitive and behavioral trust for the agentic model (P < 0.001). Specifically, on treatment planning tasks, physicians trusted the agentic reasoning most, preferring it in 89.57% of cases. Furthermore, process-oriented cognitive trust is significantly associated with outcome-oriented behavioral reliance (P < 0.001). However, measurable over-reliance on incorrect agentic outputs
arXiv:2606.30657v1 Announce Type: new Abstract: Surgical outcomes depend not only on patient factors and postoperative care but are also strongly influenced by the quality of the operation itself. Yet, for much of mod-ern surgery, intraoperative quality has been assessed indirectly through outcomes and operative reports. The increase in minimally invasive procedures inherently guided by endoscopic video, together with advances in artificial intelligence, creates an unprecedented opportunity to systematically observe, measure, and improve surgi-cal care. This chapter introduces AI-enabled Surgical Quality Assurance as a frame-work for using surgical data to support continuous assessment and improvement in the operating room. We first review existing approaches to surgical safety, from sys-tem-level interventions to procedure-specific standards. We then describe how AI can transform intraoperative video into clinically meaningful information, including recog-nition of anatomy, instrument
arXiv:2606.30656v1 Announce Type: new Abstract: Artificial Intelligence (AI) has the potential to be transformative for development, but Africa is currently facing a fragmented and challenging "AI divide". This paper provides an empirical analysis of the current state of the AI landscape and how it compares with Africa's technological preparedness for the future. In our analysis, we approach the "AI Divide" from three angles: infrastructure, accessibility, and human capacity. First, we look at the physical constraints that prevent Africa from integrating digitally. We then evaluate the human-centred factors that limit the development of AI technology on the continent. Finally, we examine the human capacity to develop AI systems on the continent and provide three focused case studies. Our investigation shows that the physical infrastructure needed to build an AI economy on the continent is lagging, with only 38% internet penetration, poor broadband coverage and less than 1% of all data
arXiv:2606.30655v1 Announce Type: new Abstract: AI-native course assessments in senior computer science courses and related fields should grade students by \emph{AI-resilient skill}: the ability to achieve outcomes beyond a strong AI baseline. Such assessments should allow students to use AI freely, while reducing the extent to which greater private AI budget or more intensive AI use, by itself, becomes a grading advantage. This paper proposes a minimal formal framework for this goal. The framework specifies a real task, an executable evaluator, a declared AI-native Pareto frontier, and a grading rule based on Pareto surplus. The central claim is simple: Pareto surplus provides a measurable, protocol-relative certificate that a submitted artifact achieves a tradeoff not already supplied by the declared AI baseline, and grading by this surplus is AI-resilient with respect to that baseline. Interpreting surplus as evidence of student skill requires the surrounding assessment protocol--fo
arXiv:2606.30653v1 Announce Type: new Abstract: Large language models are increasingly deployed in agentic pipelines that depend on the model evaluating its own outputs without external verification. The reliability of these pipelines depends on an implicit assumption: that the model applies relevant concepts the same way when it generates an output and later evaluates that output. We propose a new measure, generator-evaluator self-consistency, to test this assumption directly and apply it to 10 frontier models across 491 concepts. We find, first, that there is substantial variation in self-consistency. Second, we find that in a clinical setting with physician-validated mistakes (Proniakin et al., 2025), across models, those with higher self-consistency are linked to greater vulnerability to mistakes. Thus, even when models consistently apply concepts they may not be safe to deploy. This is evidence of a consistency dilemma in LLMs: self-consistency is operationally useful, but models
arXiv:2606.30652v1 Announce Type: new Abstract: Transparency is increasingly mandated for public-sector AI systems, with organisations required to publish statements describing their AI use and oversight arrangements. However, the existence of such artefacts is often treated as equivalent to transparency itself, despite limited evidence that they proportionately serve relevant stakeholder groups. From a requirements engineering perspective, this raises a validation concern: compliance with mandated disclosure criteria does not necessarily ensure transparency adequacy for stakeholders with different levels of risk exposure, decision control, and involvement. This paper presents an empirical analysis of 92 publicly available AI transparency statements published by Australian Government agencies under the national AI governance mandate. We introduce the stakeholder Risk--Control--Involvement--Need (RCIN) framework to differentiate stakeholder classes according to their structural position
arXiv:2606.30651v1 Announce Type: new Abstract: Delirium is common in hospitalized patients and is often missed in routine care. We present a user-centered interactive machine learning (UC-iML) framework for delirium detection support that combines physician-guided feature refinement with interpretable modeling. Using 3,862 labeled admissions from six Toronto hospitals in the General Medicine Inpatient Initiative (GEMINI), we integrate administrative variables, laboratory results, medications, and a radiology-derived text indicator. Physicians guide feature refinement and model evaluation, and Shapley Additive exPlanations (SHAP) are used to summarize feature attribution. We evaluate standard supervised classifiers with temporally separated holdout testing and a later-phase validation cohort. Compared with automated and baseline variants, the proposed framework shows better overall discrimination and stronger temporal robustness, while the explanations highlight clinically meaningful s
arXiv:2606.30650v1 Announce Type: new Abstract: Educational support services often face a qualified-capacity problem: staff time is scarce, qualifications decay, new support needs can appear before anyone is prepared for them, and training consumes the same hours needed by current students. We introduce a synthetic benchmark and decision-support framework for qualified educational capacity planning. The model is a stylized single-institution service system with heterogeneous support-demand categories, backlog-only dynamics, continuous preparation states with hard threshold qualification and decay, and capacity-consuming training. The benchmark includes seed-controlled scenarios for announced and surprise new support categories, staff absences, and demand surges; exact feasibility discipline; declared per-policy information sets; requalification and greenfield-qualification counters; access-dispersion metrics; replay checksums; and paired statistics. We compare service-only, reactive, s
arXiv:2606.30649v1 Announce Type: new Abstract: As LLM-based agents are increasingly deployed to negotiate, delegate, or transact on a user's behalf, software pipelines need runtime mechanisms to verify that an agent's stated intentions match its actual behavior. We study whether a lightweight, real-time chain-of-thought (CoT) monitor can detect strategic deception during asymmetric negotiations, using a used-car sales scenario where a seller agent has private knowledge of an undisclosed defect and a buyer agent has only public market data. The monitor, implemented as a third agent, audits the seller's internal reasoning against its messages and alerts the buyer whenever concealment is detected, across multiple buyer-seller model pairings. Our experiments show that this monitor increases the buyer's walk-away rate, but reveal a persistent intelligence gap: lower-capability buyers often cannot translate an alert into an equitable counter-offer and still accept exploitative deals after b
Higher education has spent the last decade optimizing for the student experience with everything from enrollment funnels to retention analytics and personalized dashboards, while largely overlooking the people responsible for keeping it all running. And all of that optimization matters. But there’s a conversation that can’t wait any longer: the employee experience — specifically, what happens when the higher ed IT staff who keep institutions running don’t feel like institutions are keeping them. This isn’t a soft HR concern. It’s an operational risk. EDUCAUSE’s 2026 Workforce Report found…
Conversations with Kevin Hogan: Peter Youngs of the University of Virginia and Edthena CEO Adam Geller on the future promise of personalized learning for teachers.
arXiv:2606.26744v2 Announce Type: replace-cross Abstract: We present HyperDFlash, a block-parallel speculative decoding framework tailored to DeepSeek-V4's Hyper-Connections (HC). Despite the strong performance of DeepSeek-V4's native Multi-Token Prediction (MTP) module on initial token drafting, its draft accuracy degrades sharply at later positions, as error accumulation from unverified intermediate tokens harms draft acceptance rates. Although the original DFlash method supports efficient one-pass block drafting, it cannot be seamlessly adapted to the HC paradigm, since DeepSeek-V4's multi-path residual stream induces inherent feature misalignment with conventional drafting designs. To resolve this architectural mismatch, we propose two dedicated, model-aligned optimizations for HC residual streams. First, we adopt pre-collapse residual states as the exclusive conditioning signal, preserving complete multi-path structural information and better aligning the drafter with the target's
arXiv:2606.22248v2 Announce Type: replace-cross Abstract: Standard autoregressive Transformer decoders can often exhibit substantial forgetting under sequential fine-tuning on shifting curriculum distributions. This technical report evaluates SamatNext v0.2-B, an experimental 356M-parameter hybrid sequence decoder that alternates Differential-Attention-style layers with DeltaNet-inspired simplified linear-state mixer layers using RMS normalization and output scale calibration. We study the model under a controlled staged Python code curriculum and compare it with a parameter-matched Transformer baseline. In this setting, SamatNext v0.2-B achieves a 100.0% pass rate on the controlled Stage 5 holdout while retaining 98.8% of adjacent Stage 3 semantic behavior and reaching 12.0% on the Stage 2E early syntax holdout. The strongest Transformer baseline reaches 97.6% on Stage 5 but retains only 6.0% of Stage 3 behavior. Both architectures remain weak on long-horizon early-stage retention, so
arXiv:2606.14150v3 Announce Type: replace-cross Abstract: Pruning promises a shortcut to strong small language models. In this work, we examine this promise by pruning Llama-3.1-8B at pruning ratios of 0.5--0.8 with six methods spanning depth, width, and sparse granularities, under two controlled token-matched settings. (1) With the same training token budget, pruned initialization consistently outperforms random initialization. This shows that the parent model provides a strong starting point, although the advantage narrows as the training token budget grows and as the pruning ratio rises, nearly vanishing at the highest pruning ratio we study. (2) When training from scratch is instead given the full token budget consumed by the whole pipeline, pruning at finer granularities still retains an advantage, while coarser structured pruning can be matched or surpassed. This suggests that the parent model transfers knowledge that additional training tokens alone cannot fully recover, but onl
arXiv:2606.11270v2 Announce Type: replace-cross Abstract: Distillation of a language model intended to transfer benign behavior to a student model may also transfer undesirable characteristics, if they are present in the teacher model, a phenomenon known as subliminal learning. While qualitative evidence supports the existence of this effect, its magnitude has not been systematically characterized. This study quantifies subliminal behavioral transfer ratios by steering two teacher models (Llama-2-7B-Chat and Qwen2.5-7B-Instruct) at varying steering strengths and distilling student models using only benign data. Evaluation on 100 JailbreakBench prompts with GPT-4.1, serving as the evaluator, indicates that transfer is robust but exhibits distinct scaling behaviors. Llama-2 demonstrates a sharp threshold ($\tau = {0.25,0.32} \ \text{beyond} \ \alpha = -0.15$), whereas Qwen2.5 displays continuous and higher levels of transfer ($\tau$ up to $0.61$).
arXiv:2606.05510v2 Announce Type: replace-cross Abstract: Telehealth systems have become increasingly important for delivering accessible and timely medical information. Existing large language models often struggle to provide consistent and contextually appropriate medical responses across varying levels of case severity. This limitation highlights the need for models that can effectively adapt to the progressive complexity in medical queries. To address this challenge, we introduce a severity-aware multi-model framework that integrates curriculum training strategy with relevance-based response selection. The proposed framework employs a three-stage curriculum learning strategy, where each model is trained sequentially on mild, moderate, and critical cases to progressively acquire domain knowledge. The approach utilizes five large language models, each independently trained under the same curriculum scheme. During inference, all models generate candidate responses, and the most approp
arXiv:2606.01215v2 Announce Type: replace-cross Abstract: Current 3D spatial reasoning methods face a fundamental trade-off: neuro-symbolic 3D (NS3D) concept learners achieve interpretable reasoning through compositional programs but are constrained to closed-set concept vocabularies and simple programs; end-to-end 3D multi-modal LLMs (3D MLLMs) could handle complex natural language and open-vocabulary concepts but suffer from black-box reasoning without explicit spatial verification. We introduce APEIRIA, a neuro-symbolic 3D MLLM to bridge two paradigms by distilling symbolic reasoning patterns into MLLMs with natural language chain-of-thought. Our three-stage curriculum progressively builds reasoning capabilities: a) 3D perception alignment grounds object visual-geometric features to the LLM, b) CoT-SFT teaches query decomposition and stepwise verification from symbolic program traces, and c) CoT-RL extends reasoning patterns to open-set concepts and deeply nested instructions. By tr
arXiv:2605.24844v2 Announce Type: replace-cross Abstract: While general-purpose Large Language Models (LLMs) applied to Geology often hallucinate when reasoning about subsurface structures and deep-time evolution, current AI in Earth sciences predominantly targets surface remote sensing and GIS. To bridge this gap, we introduce Geo-Expert, a family of parameter-efficient geological LLMs fine-tuned on a custom-curated, high-quality instruction dataset processed using our custom instruction synthesis pipeline. We investigate the impact of model scaling and architecture by fine-tuning three base models: Qwen3-8B, Qwen3-32B, and Gemma-3-27B, with Low-Rank Adaptation (LoRA) method. Our extensive evaluation on a novel domain-specific benchmark, Geo-Eval, reveals that a domain-aligned 8B model can outperform open-weight 70B generalists and proprietary GPT-4o on specialized geological reasoning, while a 32B variant approaches frontier reasoning models. The optimized 8B model further offers a c
arXiv:2604.28123v3 Announce Type: replace-cross Abstract: The standard post-training recipe for large multimodal models (LMMs) applies supervised fine-tuning (SFT) on curated demonstrations followed by reinforcement learning with verifiable rewards (RLVR). However, SFT introduces distributional drift that neither preserves the model's original capabilities nor faithfully matches the supervision distribution. This problem is further amplified in multimodal reasoning, where perception errors and reasoning failures follow distinct drift patterns that compound during subsequent RL. We introduce PRISM, a three-stage pipeline that mitigates this drift by inserting an explicit distribution-alignment stage between SFT and RLVR. Building on the principle of on-policy distillation (OPD), PRISM casts alignment as a black-box, response-level adversarial game between the policy and a Mixture-of-Experts (MoE) discriminator with dedicated perception and reasoning experts, providing disentangled corre
arXiv:2604.07753v2 Announce Type: replace-cross Abstract: Empowering Large Multimodal Models (LMMs) with image generation often leads to catastrophic forgetting in understanding tasks due to severe gradient conflicts. While existing paradigms like Mixture-of-Transformers (MoT) mitigate this conflict through structural isolation, they fundamentally sever cross-modal synergy and suffer from capacity fragmentation. In this work, we present Symbiotic-MoE, a unified pre-training framework that resolves task interference within a native multimodal Mixture-of-Experts (MoE) Transformers architecture with zero-parameter overhead. We first identify that standard MoE tuning leads to routing collapse, where generative gradients dominate expert utilization. To address this, we introduce Modality-Aware Expert Disentanglement, which partitions experts into task-specific groups while utilizing shared experts as a multimodal semantic bridge. Crucially, this design allows shared experts to absorb fine-g
arXiv:2604.02371v2 Announce Type: replace-cross Abstract: Visual long-document understanding is critical for enterprise, legal, and scientific applications, yet the best performing open recipes have not explored reasoning, a capability which has driven leaps in math and code performance. We introduce a synthetic data pipeline for reasoning in long-document understanding that generates thinking traces by scoring each page for question relevance, extracting textual evidence and ordering it from most to least relevant. We apply SFT to the resulting traces within \texttt{} tags, gated by a \texttt{} control token, and the resulting reasoning capability is internalized via low-strength model merging. We study Qwen3 VL 32B and Mistral Small 3.1 24B. With Qwen3 VL, we achieve 58.3 on MMLongBenchDoc, surpassing the 7$\times$ larger Qwen3 VL 235B A22B (57.0). With Mistral, we show that synthetic reasoning outperforms distillation from the Thinking version's traces by 3.8 points on MMLBD-C, and
arXiv:2603.17621v2 Announce Type: replace-cross Abstract: Reinforcement Learning (RL) has emerged as a powerful paradigm for training LLM-based agents, yet remains limited by low sample efficiency, stemming not only from sparse outcome feedback but also from the agent's inability to leverage prior experience across episodes. While augmenting agents with historical experience offers a promising remedy, existing approaches suffer from a critical weakness: the experience distilled from history is either stored statically or fail to coevolve with the improving actor, causing a progressive misalignment between the experience and the actor's evolving capability that diminishes its utility over the course of training. Inspired by complementary learning systems in neuroscience, we present Complementary RL to achieve seamless co-evolution of an experience extractor and a policy actor within the RL optimization loop. Specifically, the actor is optimized via sparse outcome-based rewards, while th
arXiv:2603.05786v2 Announce Type: replace-cross Abstract: As AI agents become widely deployed as online services, users often rely on an agent developer's claim about how safety is enforced, which introduces a threat where safety measures are falsely advertised. To address the threat, we propose proof-of-guardrail, a system that enables developers to provide cryptographic proof that a response is generated after a specific open-source guardrail. To generate proof, the developer runs the agent and guardrail inside a Trusted Execution Environment (TEE), which produces a TEE-signed attestation of guardrail code execution verifiable by any user offline. We implement proof-of-guardrail for OpenClaw agents and evaluate latency overhead and deployment cost. Proof-of-guardrail ensures integrity of guardrail execution while keeping the developer's agent private, but we also highlight a risk of deception about safety, for example, when malicious developers actively jailbreak the guardrail. Code
arXiv:2602.20610v3 Announce Type: replace-cross Abstract: Specifications are vital for ensuring program correctness, yet writing them manually remains challenging and time-intensive. Recent large language model (LLM)-based methods have shown successes in generating specifications such as postconditions, but existing single-pass prompting often yields inaccurate results. In this paper, we present SpecMind, a novel framework for postcondition generation that treats LLMs as interactive and exploratory reasoners rather than one-shot generators. SpecMind employs feedback-driven multi-turn prompting approaches, enabling the model to iteratively refine candidate postconditions by incorporating implicit and explicit correctness feedback, while autonomously deciding when to stop. This process fosters deeper code comprehension and improves alignment with true program behavior via exploratory attempts. Our empirical evaluation shows that SpecMind significantly outperforms state-of-the-art approac
arXiv:2602.15257v3 Announce Type: replace-cross Abstract: We present the first comprehensive, large-scale study of training long-context vision language models up to 344K context, targeting long-document visual question answering with measured transfer to long-context text. While several such strong are open-weight, namely Qwen3 VL and GLM 4.5/6V, their training recipes and data pipelines are not reproducible. We systematically study continued pretraining, supervised finetuning, and preference optimization for 24B and 32B parameter models, backed by extensive LC evaluations and ablations to bridge this gap, and achieve state-of-the-art performance on MMLongBenchDoc for both parameter scales. In addition to this, our key findings include: (i) training on context lengths that match evaluation context lengths outperforms training on longer contexts, (ii) training and evaluating with page indices provides a simple, high-impact boost to long-document performance, (iii) our synthetic data pi
arXiv:2602.13792v2 Announce Type: replace-cross Abstract: Artificial intelligence built on large foundation models has transformed language understanding, computer vision, and reasoning, yet these systems remain isolated and cannot readily share their capabilities. Coordinating the complementary strengths of independently developed, black-box foundation models is essential for trustworthy intelligent systems, yet no established method exists. Here we show that such coordination can be achieved through a meta-ensemble framework termed StackingNet, which aggregates the output predictions of independent models at inference. StackingNet improves accuracy, reduces individual-model error and group-wise disparities, ranks model reliability, and identifies or prunes models that degrade performance, all without access to internal parameters or training data. Across language comprehension, visual attribute estimation, and academic paper rating, it consistently outperforms individual models and c
arXiv:2602.13562v2 Announce Type: replace-cross Abstract: While reasoning models have achieved remarkable success in complex reasoning tasks, their increasing power necessitates stringent safety measures. For safety alignment, the core challenge lies in the inherent trade-off between safety and utility. However, prevailing alignment strategies typically construct CoT training data with explicit safety rules via context distillation. This approach inadvertently limits reasoning capabilities by creating a rigid association between rule memorization and refusal. To mitigate the safety-utility trade-off, we propose the Adaptive Safe Context Learning~(ASCL) framework to improve the reasoning given proper context. ASCL formulates safety alignment as a multi-turn tool-use process, empowering the model to autonomously decide when to consult safety rules and how to generate the ongoing reasoning. Furthermore, to counteract the preference for rule consultation during RL, we introduce Inverse Fre
arXiv:2602.12418v2 Announce Type: replace-cross Abstract: Jailbreak attacks remain a persistent threat to large language model safety. We propose Context-Conditioned Delta Steering (CC-Delta), an SAE-based defense that identifies jailbreak-relevant sparse features by comparing token-level representations of the same harmful request with and without jailbreak context. Using paired harmful/jailbreak prompts, CC-Delta selects features via statistical testing and applies inference-time mean-shift steering in SAE latent space. Across four aligned instruction-tuned models and twelve jailbreak attacks, CC-Delta achieves comparable or better safety-utility tradeoffs than baseline defenses operating in dense latent space. In particular, our method clearly outperforms dense mean-shift steering on all four models, and particularly against out-of-distribution attacks, showing that steering in sparse SAE feature space offers advantages over steering in dense activation space for jailbreak mitigatio
arXiv:2602.02898v3 Announce Type: replace-cross Abstract: Language model benchmarks are pervasive and computationally-efficient proxies for real-world performance. However, many recent works find that benchmarks often fail to predict real utility. Towards bridging this gap, we introduce benchmark alignment, where we use limited amounts of information about model performance to automatically update offline benchmarks, aiming to produce new static benchmarks that predict model pairwise preferences in given test settings. We then propose BenchAlign, the first solution to this problem, which learns preference-aligned weightings for benchmark questions using the question-level performance of language models alongside ranked pairs of models that could be collected during deployment, producing new benchmarks that rank previously unseen models according to these preferences. Our experiments show that our aligned benchmarks can accurately rank unseen models according to models of human preferen
arXiv:2602.02472v2 Announce Type: replace-cross Abstract: Progressive Learning (PL) reduces pre-training computational overhead by gradually increasing model scale. While prior work has extensively explored depth expansion, width expansion remains significantly understudied, with the few existing methods limited to the early stages of training. However, expanding width during the mid-stage is essential for maximizing computational savings, yet it remains a formidable challenge due to severe training instabilities. Empirically, we show that naive initialization at this stage disrupts activation statistics, triggering loss spikes, while copy-based initialization introduces gradient symmetry that hinders feature diversity. To address these issues, we propose SPARKLING (balancing {S}ignal {P}reservation {A}nd symmet{R}y brea{K}ing for width-progressive {L}earn{ING}), a novel framework for mid-stage width expansion. Our method achieves signal preservation via RMS-scale consistency, stabiliz
arXiv:2512.09066v2 Announce Type: replace-cross Abstract: Reliable assessment of the abilities of large audio language models (LALMs) is essential to advancing the state of the art. As benchmarks rapidly evolve to incorporate complex reasoning and subjective tasks, they increasingly necessitate open-ended responses from LALMs. We present Open-ended Response Correctness Assessment (ORCA) -- a reliable and lightweight model-based approach for answer correctness and disagreement modeling. We employ a three-stage annotation pipeline combining human judgment, structured feedback, and human-AI correction, yielding 9,663 annotations across 3,699 question-answer pairs from 15 LALMs on three audio understanding and reasoning benchmarks (achieving a Krippendorff's alpha of 0.82). Our experiments employing curriculum learning show that ORCA models achieve a Spearman correlation of 0.91 with average human correctness ratings on seen benchmarks and generalize to unseen benchmarks with a score of 0.
arXiv:2512.02456v2 Announce Type: replace-cross Abstract: Vision-Language Models (VLMs) have achieved remarkable progress in integrating visual perception with language understanding. However, effective multimodal reasoning requires both accurate perception and robust reasoning, and weakness in either limits the performance of VLMs. Prior efforts to enhance reasoning often depend on high-quality chain-of-thought (CoT) data, obtained via labor-intensive human annotations, costly proprietary models, or self-training methods that overlook perception. To address these limitations, we propose a simple yet effective self-training framework called See-Think-Learn (STL). At its core, STL introduces a structured reasoning template that encourages the model to see before thinking, first extracting visual attributes in textual form, then using them to guide reasoning. The framework jointly improves perception and reasoning by having the model generate and learn from its own structured rationales
arXiv:2511.14900v2 Announce Type: replace-cross Abstract: Vision--language models (VLMs) have recently shown promise for assisting clinical reasoning in dermatological diagnosis. However, their trustworthiness and clinical utility remain limited by three key challenges: heterogeneous datasets with inconsistent diagnostic labels and concept annotations, the lack of grounded diagnostic rationales for reliable reasoning supervision, and limited scalability when transferring knowledge from small, densely annotated datasets to large collections with sparse labels. To address these challenges, we propose Skin-R1, a dermatology-oriented VLM that integrates textbook-grounded clinical reasoning supervision with reinforcement learning (RL) to improve the accuracy and robustness of diagnostic prediction. First, we construct a textbook-based reasoning generator that synthesizes hierarchy-aware and differential-diagnosis (DDx) diagnostic trajectories derived from authoritative dermatology knowledge
arXiv:2511.02734v3 Announce Type: replace-cross Abstract: Current evaluations of Large Language Model (LLM) agents primarily emphasize task completion, often overlooking resource efficiency and adaptability. This neglects a crucial capability: agents' ability to devise and adjust cost-optimal plans in response to changing environments. To bridge this gap, we introduce CostBench, a scalable, cost-centric benchmark designed to evaluate agents' economic reasoning and replanning abilities. Situated in the travel-planning domain, CostBench comprises tasks solvable via multiple sequences of atomic and composite tools with diverse, customizable costs. It also supports four types of dynamic blocking events, such as tool failures and cost changes, to simulate real-world unpredictability and necessitate agents to adapt in real time. Evaluating leading open-sourced and proprietary models on CostBench reveals a substantial gap in cost-aware planning: agents frequently fail to identify cost-optimal
arXiv:2510.12784v2 Announce Type: replace-cross Abstract: Recently, remarkable progress has been made in Unified Multimodal Models (UMMs), which integrate vision-language generation and understanding capabilities within a single framework. However, a model's strong visual understanding often fails to transfer to visual generation: it may correctly judge prompt-image alignment while failing to generate a faithful image from the same prompt. This raises a compelling question: Can a model improve itself by using its understanding module to reward its generation module? We introduce SRUM, a self-rewarding post-training framework directly applicable to existing UMMs of various designs. SRUM creates a feedback loop where the model's own understanding module acts as an internal ``evaluator'', providing corrective signals to improve generation without additional human-labeled data or external reward models. To provide comprehensive feedback, SRUM uses a global-local dual reward system: a \text
arXiv:2510.06096v3 Announce Type: replace-cross Abstract: The objectives that Large Language Models (LLMs) implicitly optimize remain dangerously opaque, making trustworthy alignment and auditing a grand challenge. While Inverse Reinforcement Learning (IRL) can infer reward functions from behaviour, existing approaches either produce a single, overconfident reward estimate or fail to address the fundamental ambiguity of the task (non-identifiability). This paper introduces a principled auditing framework that re-frames reward inference from a simple estimation task to a comprehensive process for verification. Our framework leverages Bayesian IRL to not only recover a distribution over objectives but to enable three critical audit capabilities: (i) Quantifying and systematically reducing non-identifiability by demonstrating posterior contraction over sequential rounds of evidence; (ii) Providing actionable, uncertainty-aware diagnostics that expose spurious shortcuts and identify out-of
arXiv:2509.23292v4 Announce Type: replace-cross Abstract: Tool-integrated reasoning (TIR) has become a key approach for improving large reasoning models (LRMs) on complex problems. Prior work has mainly studied when to invoke tools, while overlooking how tools are applied. We identify two common patterns: a calculator pattern that uses code for direct computation, and an algorithmic pattern that encodes problems as programs. Misaligned choices often cause failures even when reasoning is sound. We propose a two-stage framework that first builds code competence from both patterns and then aligns pattern selection with teacher preferences. Across challenging math datasets, our pattern-aware method substantially improves both code usage and accuracy, for instance raising Code@1 on MATH500 from 64.0% to 70.5% and on AIME24 from 26.7% to 50.0%. These gains highlight the effectiveness of a pattern-aware approach for tool-integrated reasoning.
arXiv:2503.00539v2 Announce Type: replace-cross Abstract: Reinforcement learning from human feedback (RLHF) has evolved to be one of the main methods for fine-tuning large language models (LLMs). However, existing RLHF methods are non-robust, and their performance deteriorates if the downstream task differs significantly from the preference dataset used in fine-tuning. In order to mitigate this problem, we introduce a distributionally robust RLHF for fine-tuning LLMs. In particular, our goal is to ensure that a fine-tuned model retains its performance even when the distribution of prompts significantly differs from the distribution encountered during fine-tuning. We formulate distributionally robust optimization (DRO) version of two popular fine-tuning methods -- (1) reward-based RLHF and (2) reward-free DPO (direct preference optimization). We propose a minibatch gradient descent based algorithms for both of them, and theoretically prove convergence guarantees for the algorithms. Subs
arXiv:2412.08108v3 Announce Type: replace-cross Abstract: Large Vision-Language Models (LVLMs) have achieved remarkable performance on multimodal tasks but remain highly vulnerable to small adversarial perturbations in input images. Existing attacks typically target the vision encoder's final output embeddings, implicitly treating the encoder as a uniform attack surface, while a systematic analysis of which internal components are most vulnerable has remained largely unexplored. We show such analysis is essential, as adversarial vulnerability in LVLM vision encoders is structurally concentrated rather than uniformly distributed. Building on this, we propose Vision Encoder Vulnerable-Component-Targeted Universal Adversarial Perturbation (VEV-UAP), a task-agnostic and cost-efficient attack framework. Through a component- and layer-wise analysis of attention mechanisms, we identify the value components in middle layers as critical vulnerabilities that strongly influence downstream languag