To overcome the well-known memory bottleneck of AI chips, 3D stacked architectures that employ advanced packaging technology with high-density through-silicon vias (TSVs) pins have proven to be a promising solution. The 3D-stacked AI chip enables ultra-high memory bandwidth between compute and memory by stacking numerous DRAM banks atop many AI cores in a distributed manner. However, it is not easy to explore the efficiency of the 3D-stacked AI chip, due to its unique distributed nature. And we need to carefully consider multiple intertwined factors that range from upper-level computing paradigm to machine learning (ML) compiler optimizations, and to the underlying hardware architecture. In this paper, we develop Voxel, a fast and compiler-aware end-to-end simulation framework to facilitate exploring the efficiency of 3D-stacked AI chips for large language model (LLM) inference. Voxel enables the software/hardware co-exploration by employing a programming interface that allows ML compilers to customize the model execution plans. After validating the results of Voxel with an emulator on real silicon, we thoroughly examine the impact and correlation of different aspects of 3D-stacked AI chips, including state-of-the-art compute paradigms, tile-to-core mapping, tensor-to-bank mapping, NoC topologies and link bandwidth, DRAM bank bandwidth, per-core SRAM capacity, and energy/thermal constraints. Our findings disclose that the end-to-end efficiency of a 3D stacked AI chip not only is determined by the cooperative function of these factors, but also significantly depends on the mappings from tiles to AI core and DRAM banks. We report our findings throughout the paper, with the expectation that they will shed light on the development of the 3D-stacked AI chip ecosystem. We will open source Voxel and our study results for public research.
Effective mental health counseling is a complex, theory-driven process requiring the simultaneous integration of psychological frameworks, real-time distress signals, and strategic intervention planning. This level of clinical reasoning is critical for safety and therapeutic effectiveness but is often missing in general-purpose Large Language Models (LLMs). We introduce SAGE (Strategy-Aware Graph-Enhanced), a novel framework designed to bridge the gap between structured clinical knowledge and generative AI. SAGE constructs a heterogeneous graph that unifies conversational dynamics with a psychologically grounded layer, explicitly anchoring interactions in a theory-driven lexicon. Our architecture first employs a Next Strategy Classifier to identify the optimal therapeutic intervention. Subsequently, a Graph-Aware Attention mechanism projects graph-derived structural signals into soft prompts, conditioning the LLM to generate responses that maintain clinical depth. Validated through both automated metrics and expert human evaluation, SAGE outperforms baselines in strategy prediction and recommended response quality. By providing actionable intervention recommendations, SAGE serves as a cutting-edge decision-support tool designed to augment human expertise in high-stakes crisis counseling.
Large language models (LLMs) accelerate software development but often exhibit instability, non-determinism, and weak adherence to development discipline in unconstrained workflows. While test-driven development (TDD) provides a structured Red-Green-Refactor process, existing LLM-based approaches typically use tests as auxiliary inputs rather than enforceable process constraints. We present an AI-native TDD framework that operationalizes classical TDD principles as structured prompt-level and workflow-level governance mechanisms. Extracted principles are formalized in a machine-readable manifesto and distributed across planning, generation, repair, and validation stages within a layered architecture that separates model proposal from deterministic engine authority. The system enforces phase ordering, bounded repair loops, validation gates, and atomic mutation control to improve stability and reproducibility. We describe architecture and discuss encoding software engineering discipline directly into prompt orchestration, which we think offers a promising direction for reliable LLM-assisted development.
Deploying a neuro-symbolic task planner on a new domain today requires significant manual effort: a domain expert must author relaxation and complementary rules, and hundreds of training problems must be solved to supervise a Graph Neural Network (GNN) object scorer. We propose LLM-Flax, a three-stage framework that eliminates all three sources of manual effort using a locally hosted LLM given only a PDDL domain file. Stage 1 automatically generates relaxation and complementary rules via structured prompting with format validation and self-correction. Stage 2 introduces LLM-guided failure recovery with a feasibility-gated budget policy that explicitly reserves API latency cost before each LLM call, preventing the downstream relaxation fallback from being starved. Stage 3 replaces the domain-trained GNN entirely with zero-shot LLM object importance scoring, requiring no training data. We evaluate all three stages on the MazeNamo benchmark across 10x10, 12x12, and 15x15 grids (8 benchmarks total). LLM-Flax achieves average SR 0.945 versus the manual baseline's 0.828 (+0.117), matching or outperforming manual rules on every one of the eight benchmarks. On 12x12 Expert, LLM-Flax attains SR 0.733 where the manual planner fails entirely (SR 0.000); on 15x15 Hard, it achieves SR 1.000 versus Manual's 0.900. Stage 3 demonstrates feasibility (SR 0.720 on 12x12 Hard with no training data) but faces a context-window bottleneck at scale, pointing to the primary open challenge for future work.
Large Language Model (LLM)-based agents exhibit systemic failures in compositional generalization, limiting their robustness in interactive environments. This work introduces AGEL-Comp, a neuro-symbolic AI agent architecture designed to address this challenge by grounding actions of the agent. AGEL-Comp integrates three core innovations: (1) a dynamic Causal Program Graph (CPG) as a world model, representing procedural and causal knowledge as a directed hypergraph; (2) an Inductive Logic Programming (ILP) engine that synthesizes new Horn clauses from experiential feedback, grounding symbolic knowledge through interaction; and (3) a hybrid reasoning core where an LLM proposes a set of candidate sub-goals that are verified for logical consistency by a Neural Theorem Prover (NTP). Together, these components operationalize a deduction--abduction learning cycle: enabling the agent to deduce plans and abductively expand its symbolic world model, while a neural adaptation phase keeps its reasoning engine aligned with new knowledge. We propose an evaluation protocol within the \texttt{Retro Quest} simulation environment to probe for compositional generalization scenarios to evaluate our AGEL agent. Our findings clearly indicate the better performance of our AGEL model over pure LLM-based models. Our framework presents a principled path toward agents that build an explicit, interpretable, and compositionally structured understanding of their world.
Generating sports game reports from structured tables is a complex table-to-text task that demands both precise data interpretation and fluent narrative generation. Traditional model-based approaches require large, annotated datasets, while prompt-based methods using large language models (LLMs) often struggle with hallucination due to weak table comprehension. To overcome these challenges, we propose Tree-of-Text, a tree-structured prompting framework that guides LLMs through a three-stage generation process: (1) Content Planning, where relevant operations and arguments are selected from the input tables; (2) Operation Execution, which breaks down large tables into manageable sub-tables; and (3) Content Generation, where short textual outputs are merged and rewritten into a cohesive report. Experiments show that our method outperforms existing methods on ShuttleSet+, leads in RG and CO metrics on RotoWire-FG, and excels in CS and CO on MLB with roughly 40% of the time and cost of Chain-of-Table. These results demonstrate the effectiveness and efficiency of Tree-of-Text and suggest a promising direction for prompt-based table-to-text generation in the sports domain.
The arrival of large language models (LLMs) capable of multi-step reasoning, tool use, and long-horizon planning has produced a qualitative shift in software engineering. Where earlier code-completion tools such as GitHub Copilot operated at the granularity of a line or function, modern agentic systems -- Claude Code, OpenAI Codex CLI, Google Jules, Devin, OpenHands, SWE-agent, MetaGPT, ChatDev, and DeepMind's AlphaEvolve -- operate at the granularity of a repository, a feature, or an algorithm. We synthesize work from Anthropic, OpenAI, Google DeepMind, Microsoft Research, Princeton, Stanford, and the broader academic community to characterize this transition. We propose a six-layer reference architecture for agentic software engineering systems, contrast a traditional Software Development Lifecycle (SDLC) with an emerging Agentic SDLC (A-SDLC), and consolidate empirical evidence on performance (a rise from 1.96% to 78.4% on SWE-bench Verified between October 2023 and April 2026), productivity (13.6%-55.8% time savings across controlled studies), and labor-market impact (49% of jobs sampled by Anthropic in 2026 saw AI used for at least a quarter of their tasks). We argue that the central object of inquiry has shifted from code generation to delegated execution under human supervision, and we identify five open problems -- evaluation, governance, technical debt, skill redistribution, and the economics of attention -- that will determine whether the agentic transition is net-positive for the discipline.
The integration of Large Language Models (LLMs) with Retrieval-Augmented Generation (RAG) has significantly advanced Knowledge Graph Question Answering (KGQA). However, existing LLM-driven KGQA systems act as stateless planners, generating retrieval plans in isolation without exploiting historical query patterns: analogous to a database system that optimizes every query from scratch without a plan cache. This fundamental design flaw leads to schema hallucinations and limited retrieval coverage. We propose CacheRAG, a systematic cache-augmented architecture for LLM-based KGQA that transforms stateless planners into continual learners. Unlike traditional database plan caching (which optimizes for frequency), CacheRAG introduces three novel design principles tailored for LLM contexts: (1) Schema-agnostic user interface: A two-stage semantic parsing framework via Intermediate Semantic Representation (ISR) enables non-expert users to interact purely in natural language, while a Backend Adapter grounds the LLM with local schema context to compile executable physical queries safely. (2) Diversity-optimized cache retrieval: A two-layer hierarchical index (Domain $\rightarrow$ Aspect) coupled with Maximal Marginal Relevance (MMR) maximizes structural variety in cached examples, effectively mitigating reasoning homogeneity. (3) Bounded heuristic expansion: Deterministic depth and breadth subgraph operators with strict complexity guarantees significantly enhance retrieval recall without risking unbounded API execution. Extensive experiments on multiple benchmarks demonstrate that CacheRAG significantly outperforms state-of-the-art baselines (e.g., +13.2% accuracy and +17.5% truthfulness on the CRAG dataset).
Resolution of complex post-production issues in large-scale open-source software (OSS) projects requires significant cognitive effort, as developers need to go through long, unstructured and fragmented issue discussion threads before that. In this paper, we present SWE-MIMIC-Bench, an issue trajectory dataset generated from raw GitHub discussions using an automated multi-LLM pipeline. Unlike simple summarization, this pipeline utilizes a group of closed-source LLMs to perform granular tasks: analyzing individual comments with awareness of externally-linked resources, classifying comment analyses into label-specific fields (e.g., root cause, solution plan, implementation progress), and synthesizing label-aware trajectories which capture a structured and coherent narrative of the entire discussion thread. Our pipeline uses five closed-source LLM configurations for distinct purposes: label classification, inline code block and external link summarization, comment analysis, label-specific field classification and trajectory synthesis. By generating concise and reliable trajectories from complex conversation threads, this system can assist developers and researchers of broader software engineering community to understand the experience-driven collaborative approach for issue diagnosis. Furthermore, the generated trajectories can be used to train modern LLM agents to think and act like an expert developer. We evaluated our system on 800 real-world GitHub issues drawn from the SWE-Bench-Pro, SWE-Bench-Multilingual and SWE-Bench-Verified dataset, achieving a 91.7% success rate in extracting 734 high-fidelity reasoning trajectories.
Security analysts are overwhelmed by the volume of alerts and the low context provided by many detection systems. Early-stage investigations typically require manual correlation across multiple log sources, a task that is usually time-consuming. In this paper, we present an experimental, agentic workflow that leverages large language models (LLMs) augmented with predefined queries and constrained tool access (structured SQL over Suricata logs and grep-based text search) to automate the first stages of alert investigation. The proposed workflow integrates queries to provide an overview of the available data, and LLM components that selects which queries to use based on the overview results, extracts raw evidence from the query results, and delivers a final verdict of the alert. Our results demonstrate that the LLM-powered workflow can investigate log sources, plan an investigation, and produce a final verdict that has a significantly higher accuracy than a verdict produced by the same LLM without the proposed workflow. By recognizing the inherent limitations of directly applying LLMs to high-volume and unstructured data, we propose combining existing investigation practices of real-world analysts with a structured approach to leverage LLMs as virtual security analysts, thereby assisting and reducing the manual workload.
Instructed code editing is a significant challenge for large language models (LLMs). On the EditBench benchmark, 39 of 40 evaluated models obtain a task success rate (TSR) below 60 percent, highlighting a gap between general code generation and the ability to perform instruction-driven editing under executable test constraints. To address this, we propose SAFEdit, a multi-agent framework for instructed code editing that decomposes the editing process into specialized roles to improve reliability and reduce unintended code changes. A Planner Agent produces an explicit, visibility-aware edit plan, an Editor Agent applies minimal, literal code modifications, and a Verifier Agent executes real test runs. When tests fail, SAFEdit uses a Failure Abstraction Layer (FAL) to transform raw test logs into structured diagnostic feedback, which is fed back to the Editor to support iterative refinement. We compare SAFEdit against both prior single-model results reported for EditBench and an implemented ReAct single-agent baseline under the same evaluation conditions. We used EditBench to evaluate SAFEdit on 445 code editing instances in five languages (English, Polish, Spanish, Chinese, and Russian) under varying spatial context variants. SAFEdit achieved 68.6 percent TSR, outperforming the single-model baseline by 3.8 percentage points and the ReAct single-agent baseline by 8.6 percentage points. The iterative refinement loop was found to contribute 17.4 percentage points to SAFEdit's overall success rate. SAFEdit's automated error analysis further indicates a reduction in instruction-level hallucinations compared to single-agent approaches, providing an additional framework component for interpreting failures beyond pass or fail outcomes.
We propose a human in the loop approach for black-box testing of Functional Mock-up Units (FMUs) using Large Language Models (LLMs). The goal is to reduce the manual effort in defining test scenarios for dynamic simulation models and to improve the interpretability of results. The approach takes the functional and interface specifications of an FMU as input, and prompts an LLM to generate structured scenario goals in Given-When-Then format that define the initial input conditions of the simulation, a possible change in those conditions, and the expected output behaviour of the system against those changes. The corresponding scenario plans specify input patterns and add assertion oracles that describe expected output patterns defined in scenario goals. The approach generates a complete input time series for the scenario plans, runs the FMU simulation, and evaluates assertions on the recorded outputs. It produces human-readable logs and plots that show statistics for each scenario with overlays, aggregate pass rates, and per-goal outcomes. The generated scenarios and results are stored for evaluation and later re-execution. We evaluate the approach on a Lube Oil Cooling system and discuss design choices that make the approach practical for everyday use. Results suggest that LLM-assisted scenario generation can facilitate automatic test design and verification of dynamic simulation models.
Deploying production-ready multi-agent systems (MAS) in complex industrial environments remains challenging due to limitations in scalability, observability, and autonomous evolution. We present OxyGent, an open-source framework driven by two core novelties: a unified Oxy abstraction and the OxyBank evolution engine. The unified abstraction encapsulates agents, tools, LLMs, and reasoning flows as pluggable atomic components, enabling Lego-like scalable system composition and non-intrusive monitoring. To enhance observability, OxyGent introduces permission-driven dynamic planning that replaces rigid workflows with execution graphs generated at runtime, providing adaptive visualizations. Furthermore, to support continuous evolution, OxyBank serves as an AI asset management platform that drives automated data backflow, annotation, and joint evolution. Empirical evaluations and real-world case studies show that OxyGent provides a robust and scalable foundation for MAS. OxyGent is fully open-sourced under the Apache License 2.0 at https://github.com/jd-opensource/OxyGent.
Large Language Models (LLMs) have shown strong potential for narrative generation, but their use in complex, multi-layered role-playing game (RPG) worlds is still limited by issues of coherence, controllability, and structural consistency. This paper explores a dependency-aware, multi-stage prompt pipeline for procedural RPG content generation that models narrative dependencies through structured intermediate representations. The approach decomposes generation into sequential stages: world building, non-player character creation, player character creation, campaign-level quest planning, and quest expansion. Each stage conditions on structured JSON outputs from previous stages. By enforcing schemas and explicit data flow, the pipeline reduces narrative drift, limits hallucinations, and supports scalable creation of interconnected narrative elements. The system is evaluated qualitatively through human-centered analysis across multiple independent runs. Outputs are assessed using criteria such as structural completeness, internal consistency, narrative coherence, diversity, and actionability. Results show that the pipeline consistently generates logically sound and structurally valid RPG content, without quality degradation as complexity increases. Separating high-level campaign planning from detailed quest expansion improves both global structure and local storytelling. These findings suggest that dependency-aware prompt pipelines with structured intermediate representations are an effective design pattern for LLM-based procedural content generation. This approach may also generalize to other domains requiring sequential reasoning over evolving contextual states.
We study clinical trial table reasoning, where answers are not directly stored in visible cells but must be reasoned from semantic understanding through normalization, classification, extraction, or lightweight domain reasoning. Motivated by the observation that current LLM approaches often suffer from "bad reasoning" under implicit planning assumptions, we focus on settings in which the model must recover implicit attributes such as therapy type, added agents, endpoint roles, or follow-up status from partially observed clinical-trial tables. We propose SCOPE (Structured Clinical hybrid Planning for Evidence retrieval in clinical trials), a multi-LLM planner-based framework that decomposes the task into row selection, structured planning, and execution. The planner makes the source field, reasoning rules, and output constraints explicit before answer generation, reducing ambiguity relative to direct prompting. We evaluate SCOPE on 1,500 hybrid reasoning questions over oncology clinical-trial tables against zero-shot, few-shot, chain-of-thought, TableGPT2, Blend-SQL, and EHRAgent. Results show that explicit multi-LLM planning improves accuracy for reasoning-based questions while offering a stronger accuracy-efficiency tradeoff than heavier agentic baselines. Our findings position clinical trial reasoning as a distinct table understanding problem and highlight hybrid planner-based decomposition as an effective solution
Existing web agent benchmarks have largely converged on short, single-site tasks that frontier models are approaching saturation on. However, real world web use consists of long-horizon, multi-site workflows. Common web navigation tasks, such as comparing products across different domains, planning trips across multiple services, or summarizing information from multiple search queries, require sustained context and cross-site reasoning over potentially hours of browsing. To capture and evaluate such behaviors, we introduce Odysseys: a benchmark of 200 long-horizon web tasks derived from real world browsing sessions evaluated on the live Internet. We find that binary pass/fail evaluation is inadequate for long-horizon settings and introduce a rubric-based evaluation, annotating each Odysseys task with an average of 6.1 graded rubrics. We demonstrate that this yields higher agreement with humans and provides a more fine-grained signal than commonly used trajectory-level LLM-as-a-judge evaluation metrics. We tested several leading frontier models and find that the strongest models achieve a success rate of 44.5%, which leaves substantial room for future improvements. Beyond task success, we argue that efficiency is a first-class concern for long-horizon agents. We introduce a Trajectory Efficiency metric (rubric score per step) and find that even frontier agents achieve only 1.15%, marking an evident need for agents that can succeed efficiently and not simply eventually. Odysseys isolates the critical evaluation of long-horizon proficiency in open-web environments, providing a realistic benchmark to measure progress towards computer-use agents that can potentially productively operate for hours. We release our tasks, evaluation scripts, and other results at https://odysseys-website.pages.dev
Masked diffusion language models (MDMs) have recently emerged as a promising alternative to standard autoregressive large language models (AR-LLMs), yet their optimization can be substantially less stable. We study blockwise MDMs and compare them with AR-LLMs on three controlled tasks that stress different aspects of structured generation: in-context linear regression, graph path-finding, and Sudoku solving. We find that standard random-masking MDMs fail to reliably learn linear regression, exhibit high variance training dynamics on graph path-finding, while outperforming AR-LLMs on Sudoku. To mitigate these instabilities, we propose two locality aware blockwise models, namely Jigsaw and Scatter, that inject left-to-right inductive bias by enforcing autoregressive locality within blocks while preserving iterative refinement at the block level. Empirically, Jigsaw matches AR-LLM stability on linear regression and remains strong on Sudoku, while Scatter retains diffusion's planning advantage on path-finding. Our results indicate that standard random-masking MDMs, even with blockwise variants, may be a suboptimal instantiation of diffusion LMs for ordered generation, motivating models beyond random masking.
Cloud computing platforms offer elastic scaling, managed infrastructure, and pay-per-use pricing, but moving existing monolithic backends to them remains a difficult software engineering task. In practice, the migration requires coordinated changes to program structure, source code, infrastructure configuration, and cloud-specific design decisions, and these changes are still largely carried out by hand. In this paper, we present Mono2Sls, an automated pipeline that converts monolithic web backends into deployable AWS SAM applications. The pipeline combines lightweight static analysis of entry points, call graphs, and asynchronous behavior with four sequential tool-using LLM agents: Architect, Code Developer, SAM Engineer, and Consistency Validator. These agents communicate through explicit intermediate artifacts and consult a curated SAM knowledge base. Evaluated on six benchmark applications totaling more than 10K lines of code and 76 business endpoints, Mono2Sls achieves 100% deployment success without manual fixes. It also reaches 66.1% end-to-end correctness and 98.7% API-coverage F1, whereas the commercial baselines achieve 53.7--61.2% and 88.4%, respectively. The migrated systems show more consistent use of AWS-native authentication and asynchronous patterns, and an ablation study indicates that static-analysis-guided architecture planning contributes 23.4 percentage points to end-to-end correctness.
As LLM agents transition to autonomous digital coworkers, maintaining deterministic goal-directedness in non-linear multi-turn conversations emerged as an architectural bottleneck. We identify and formalize a systemic failure mode termed the Attention Latch in decoder-only autoregressive Transformers. This phenomenon, a behavioral manifestation of Information Over-squashing, occurs when the cumulative probabilistic weight of historical context overrides mid-task updates, causing agents to remain anchored to obsolete constraints despite explicit contradictory instructions. We propose Self-Synthesizing Reasoning Protocols (SSRP), a metacognitive framework that implements a discrete separation between high-level architectural planning (Architect) and turn-by-turn procedural execution (Executive). We evaluate SSRP across 9K trajectories using the MultiWOZ 2.2 dataset and the Aggregate Pivot Accuracy (APA), a novel metric we validate by mapping its scores to the U-shaped 'Lost in the Middle' curve. We present 3 experimental tiers: a shallow recency-based retrieval pilot, a high-entropy SOP, and a semantic hijacked 3-hop Multi-Fact Synthesis task. Our results empirically locate the Attention Stability Boundary, where stateless Vanilla ReAct baselines for GPT 5.4 collapse to 0.1% success while SSRP achieves a 715X Resilience Lift. We demonstrate statistically significant gains across Gemini 3.1 Pro, Claude Sonnet 4.6 and DeepSeek V3.2. Audits confirm SSRP necessity by proving attentional lapse via a recursive reflexion baseline (100% success); decoupling the latch from positional bias through equidistant stress testing (90% accuracy); and formalizing SSRP via the Information Bottleneck principle and granularity ablations. Procedural Integrity audit (98.8% adherence) reveals a Grounding Paradox where high-stability models fail by refusing to hallucinate under retrieval-reasoning contamination.
Quantum computing instructors face a compounding problem: the concepts are counterintuitive, the mathematical formalism is dense, and qualified faculty are scarce outside a small number of well-resourced institutions. Our prior work introduced a knowledge-graph-augmented tutoring prototype with two specialized LLM agents: a Teaching Agent for dynamic interaction and a Lesson Planning Agent for lesson generation. Validated on simulated runs rather than in a real course, that prototype left open whether more aggressive agent specialization would be needed to handle the full range of quantum education tasks under real student load. This paper answers the three questions that the prototype could not answer. Can agent specialization solve the reliability problem in a domain as technically demanding as quantum information science? Can the system run in a real course, not a demonstration? Does the instructor gain actionable intelligence from the deployment? We present ITAS (Intelligent Teaching Assistant System), a multi-agent tutoring system built around four contributions: a five-module QIS curriculum grounded in Watrous's information-first framework, a Spoke-and-Wheel teaching architecture with quantum-specialized agents, a cloud infrastructure designed for production use and regulatory compliance, and a conversational analytics layer for instructors and content developers. Piloted in a quantum computing course at Old Dominion University, the system supports all three answers: deployment evidence is consistent with specialization addressing the task-boundary failures observed in the prototype, cloud infrastructure supports classroom-scale concurrency at sub-textbook cost, and the analytics agent surfaces curriculum gaps the instructor could not otherwise see.
Embodied AI agents increasingly rely on large language models (LLMs) for planning, yet per-step LLM calls impose severe latency and cost. In this paper, we show that embodied tasks exhibit strong plan locality, where the next plan is largely predictable from the current one. Building on this, we introduce AgenticCache, a planning framework that reuses cached plans to avoid per-step LLM calls. In AgenticCache, each agent queries a runtime cache of frequent plan transitions, while a background Cache Updater asynchronously calls the LLM to validate and refine cached entries. Across four multi-agent embodied benchmarks, AgenticCache improves task success rate by 22% on average across 12 configurations (4 benchmarks x 3 models), reduces simulation latency by 65%, and lowers token usage by 50%. Cache-based plan reuse thus offers a practical path to low-latency, low-cost embodied agents. Code is available at https://github.com/hojoonleokim/MLSys26_AgenticCache.
We explore a central question in AI for mathematics: can AI systems produce original, nontrivial proofs for open research problems? Despite strong benchmark performance, producing genuinely novel proofs remains an outstanding challenge for LLMs. Through systematic experiments with frontier LLMs on research-level proof tasks, we identify seven failure modes that prevent reliable proof generation, including context contamination, citation hallucination, hand-waving on key steps and misallocation of proof effort, unstable proof plans, unfocused verification, problem modification and single-model bottleneck. We argue that the gap between benchmark success and research-level proving is primarily one of system design, due to those failure modes. We present QED, an open-source multi-agent proof system in which each architectural decision directly addresses a specific failure mode. Evaluated on five open problems in applied analysis and PDEs contributed by domain experts, QED produces correct proofs for three problems, each verified by the contributing experts as original and nontrivial. QED is released as open-source software at https://github.com/proofQED/QED.
Indoor navigation remains a critical accessibility challenge for the blind and low-vision (BLV) individuals, as existing solutions rely on costly per-building infrastructure. We present an agentic framework that converts a single floor plan image into a structured, retrievable knowledge base to generate safe, accessible navigation instructions with lightweight infrastructure. The system has two phases: a multi-agent module that parses the floor plan into a spatial knowledge graph through a self-correcting pipeline with iterative retry loops and corrective feedback; and a Path Planner that generates accessible navigation instructions, with a Safety Evaluator agent assessing potential hazards along each route. We evaluate the system on the real-world UMBC Math and Psychology building (floors MP-1 and MP-3) and on the CVC-FP benchmark. On MP-1, we achieve success rates of 92.31%, 76.92%, and 61.54% for short, medium, and long routes, outperforming the strongest single-call baseline (Claude 3.7 Sonnet) at 84.62%, 69.23%, and 53.85%. On MP-3, we reach 76.92%, 61.54%, and 38.46%, compared to the best baseline at 61.54%, 46.15%, and 23.08%. These results show consistent gains over single-call LLM baselines and demonstrate that our workflow is a scalable solution for accessible indoor navigation for BLV individuals.
Autonomous neutron spectroscopy must solve three distinct tasks: detection (where is the signal?), inference (which Hamiltonian governs it?), and refinement (what are the parameters?). No single controller solves all three equally well. We present TAS-AI, a hybrid agnostic-to-physics-informed framework for autonomous triple-axis spin-wave spectroscopy that separates these tasks explicitly. In blind reconstruction benchmarks, model-agnostic methods such as random sampling, coarse grids, and Gaussian-process mappers reach a global error threshold more reliably and with fewer measurements than physics-informed planning, supporting the claim that discovery and inference are distinct tasks requiring distinct controllers. Once signal structure is localized, the physics-informed stage performs in-loop Hamiltonian discrimination and parameter refinement: in a controlled square-lattice test between nearest-neighbor-only and J1-J2 Hamiltonians, TAS-AI reaches a decisive AIC-derived evidence ratio (>100) in fewer than 10 measurements, while motion-aware scheduling cuts wall-clock time by 32% at a fixed measurement budget. We also identify a failure mode of posterior-weighted design, algorithmic myopia, in which the planner over-refines the current leading model while under-sampling low-intensity falsification probes. A constrained falsification channel sharply reduces time spent committed to the wrong model and accelerates correct model selection without modifying the Bayesian inference engine. In controlled two-model ablations, both a deterministic top-two max-disagreement rule and an LLM-based audit committee achieve this gain under identical constraints. We demonstrate the full workflow in silico using a high-fidelity digital twin and provide an open-source Python implementation.
LLM routing has achieved promising results in integrating the strengths of diverse models while balancing efficiency and performance. However, to support more realistic and challenging applications, routing must extend into agentic LLM settings, where task planning, multi-round cooperation among heterogeneous agents, and memory utilization are indispensable. To address this gap, we propose GraphPlanner, a heterogeneous graph memory-augmented agentic router for multi-agent LLMs that generates routing workflows for each query and supports both inductive and transductive inference. GraphPlanner formulates workflow generation as a Markov Decision Process (MDP), where at each step it selects both the LLM backbone and the agent role, including Planner, Executor, and Summarizer. By leveraging a heterogeneous graph, denoted as GARNet, to capture interaction memories among queries, agents, and responses, GraphPlanner integrates historical memory and workflow memory into richer state representations. The entire pipeline is optimized with reinforcement learning, jointly improving task-specific performance and computational efficiency. We evaluate GraphPlanner across 14 diverse LLM tasks and demonstrate that: (1) GraphPlanner outperforms strong single-round and multi-round routers, improving accuracy by up to 9.3% while reducing GPU cost from 186.26 GiB to 1.04 GiB; (2) GraphPlanner generalizes robustly to unseen tasks and LLMs, exhibiting strong zero-shot capabilities; and (3) GraphPlanner effectively leverages historical memories, supporting both inductive and transductive inference for more adaptive routing. Our code for GraphPlanner is released at https://github.com/ulab-uiuc/GraphPlanner.
The application of large language models (LLMs) in clinical decision support faces significant challenges of "tunnel vision" and diagnostic hallucinations present in their processing unstructured electronic health records (EHRs). To address these challenges, we propose a novel chain-based clinical reasoning framework, called DxChain, which transforms the diagnostic workflow into an iterative process by mirroring a clinician's cognitive trajectory that consists of "Memory Anchoring", "Navigation" and "Verification" phases. DxChain introduces three key methodological innovations to elicit the potential of LLM: (i) a Profile-Then-Plan paradigm to mitigate cold-start hallucinations by establishing a panoramic patient baseline, (ii) a Medical Tree-of-Thoughts (Med-ToT) algorithm for strategic look ahead planning and resource aware navigation, and (iii) a Dialectical Diagnostic Verification procedure utilizing "Angel-Devil" adversarial debates to resolve complex evidence conflicts. Evaluated on two real world benchmarks, MIMIC-IV-Ext Cardiac Disease and MIMIC-IV-Ext CDM, DxChain achieves state-of-the-art performances in both diagnostic accuracy and logical consistency, offering a modular and reliable architecture for next-generation clinical AI. The code is at https://anonymous.4open.science/r/Dx-Chain.
Large Language Models (LLMs) are increasingly deployed in persona-driven applications such as education, customer service, and social platforms, where models are prompted to adopt specific personas when interacting with users. While persona conditioning can improve user experience and engagement, it also raises concerns about how personality cues may interact with gender biases and stereotypes. In this work, we present a controlled study of persona-conditioned story generation in English and Hindi, where each story portrays a working professional in India producing context-specific artifacts (e.g., lesson plans, reports, letters) under systematically varied persona gender, occupational role, and personality traits from the HEXACO and Dark Triad frameworks. Across 23,400 generated stories from six state-of-the-art LLMs, we find that personality traits are significantly associated with both the magnitude and direction of gender bias. In particular, Dark Triad personality traits are consistently associated with higher gender-stereotypical representations compared to socially desirable HEXACO traits, though these associations vary across models and languages. Our findings demonstrate that gender bias in LLMs is not static but context-dependent. This suggests that persona-conditioned systems used in real-world applications may introduce uneven representational harms, reinforcing gender stereotypes in generated educational, professional, or social content.
Accurately characterizing wind power uncertainty under icing and post-disaster conditions remains a critical challenge for resilient power system operation. To address this issue, this paper proposes a physics-aware large language model (LLM) framework for probabilistic wind power scenario generation under extreme icing conditions. The proposed framework integrates supervisory control and data acquisition (SCADA)-based physical modeling, multimodal tokenization, and a causal Transformer architecture trained in an autoregressive manner. A physics-aware decoding scheme effectively enforces rated power limits and ramping constraints on the generated trajectories while preserving stochastic diversity. Case studies using real wind turbine data show that the proposed method reproduces icing-induced power degradation and temporal variability observed during extreme weather. The resulting scenarios are physically consistent and high-fidelity, thereby significantly enhancing resilience assessment and recovery planning in renewable-integrated power systems.
Chain-of-Thought (CoT) reasoning has emerged as a key technique for eliciting complex reasoning in Large Language Models (LLMs). Although interpretable, its dependence on natural language limits the model's expressive bandwidth. Continuous thought models address this bottleneck by reasoning in latent space rather than human-readable tokens. While they enable richer representations and faster inference, they raise a critical safety question: how can we detect misaligned reasoning in an uninterpretable latent space? To study this, we introduce MoralChain, a benchmark of 12,000 social scenarios with parallel moral/immoral reasoning paths. We train a continuous thought model with backdoor behavior using a novel dual-trigger paradigm - one trigger that arms misaligned latent reasoning ([T]) and another that releases harmful outputs ([O]). We demonstrate three findings: (1) continuous thought models can exhibit misaligned latent reasoning while producing aligned outputs, with aligned and misaligned reasoning occupying geometrically distinct regions of latent space; (2) linear probes trained on behaviorally-distinguishable conditions ([T][O] vs [O]) transfer to detecting armed-but-benign states ([T] vs baseline) with high accuracy; and (3) misalignment is encoded in early latent thinking tokens, suggesting safety monitoring for continuous thought models should target the "planning" phase of latent reasoning.
Cross-domain task-oriented dialogue requires reasoning over implicit and explicit feasibility constraints while planning long-horizon, multi-turn actions. Large language models (LLMs) can infer such constraints but are unreliable over long horizons, while Reinforcement learning (RL) optimizes long-horizon behavior yet cannot recover constraints from raw dialogue. Naively coupling LLMs with RL is therefore brittle: unverified or unstructured LLM outputs can corrupt state representations and misguide policy learning. Motivated by this, we propose Verified LLM-Knowledge empowered RL (VLK-RL), a hybrid framework that makes LLM-derived constraint reasoning usable for RL. VLK-RL first elicits candidate constraints with an LLM and then verifies them via a dual-role cross-examination procedure to suppress hallucinations and cross-turn inconsistencies. The verified constraints are mapped into ontology-aligned slot-value representations, yielding a structured, constraint-aware state for RL policy optimization. Experiments across multiple benchmarks demonstrate that VLK-RL significantly improves generalization and robustness, outperforming strong single-model baselines on long-horizon tasks.