Large Language Models

Reasoning in large language models is predominantly evaluated through labeled benchmarks, conflating task performance with the quality of internal inference. Here we study reasoning as an intrinsic dynamical process by examining the evolution of internal representations during inference. We find that inference-time dynamics consistently self-organize into low-dimensional manifolds embedded within high-dimensional representation spaces. we find that such geometric compression, although pervasive, is not sufficient for stable or reliable reasoning. Instead, effective reasoning dynamics emerge within a constrained structural regime characterized by three conditions: adequate representational expressivity, spontaneous manifold compression, and preservation of non-degenerate information volume within the compressed subspace. Models outside this regime exhibit characteristic pathological inference dynamics. Based on these insights, we introduce a unified, label-free diagnostic computed solely from internal dynamics. These findings suggest that reasoning in LLMs is fundamentally governed by geometric and informational constraints, offering a complementary framework to benchmark-centric assessment.

Detecting machine-generated text is essential for transparency and accountability when deploying large language models (LLMs). Among detection approaches, watermarking is a statistically reliable method by design -- it embeds detectable signals into LLM outputs by biasing their token distributions. However, it has been reported that watermarked LLMs often perform worse on downstream tasks. We propose PUPPET, a framework that fine-tunes an LLM via reinforcement learning to generate text that is both more detectable and better performing on downstream tasks. We use two reward functions: a detector that outputs a machine-class likelihood and an evaluator that measures a task-specific metric. Experiments on long-form QA, summarization, and essay writing show that LLMs trained with PUPPET achieve high detectability competitive with watermarking methods while outperforming them on downstream tasks. The analysis shows that this optimization can be performed efficiently with only a few thousand samples in 1--2 GPU hours. Moreover, these gains are consistent across out-of-domain tasks, different LLM families, and model sizes, and are even robust to paraphrasing attacks.

Traditional approaches to test case generation often involve manual effort and incur significant computational overhead. Additionally, these approaches are not scalable, and hence, unsuitable for complex software systems. Recently, Large Language Models (LLMs) have been applied to software testing. However, single-shot prompt engineering-based approaches tend to hallucinate and generate redundant test cases, resulting in fewer branches. To handle the above-mentioned limitations, in this paper, we propose FeedbackLLM, a novel automated language-agnostic test case generation framework based on tightly coupled two-stage approach. In the first stage, FeedbackLLM extracts the input constraints by parsing source code and generates the possible test cases. The quality of the test cases is evaluated in the second stage by the following two specialized LLM feedback agents: (i) Line Feedback Agent: extracts the metadata related to missed line executions and (ii) Branch Feedback Agent: extracts the metadata of the unexecuted branch conditions. The above agents operate in a two-stage process, communicating in tandem, and this procedure is repeated for k-steps. Further, we also introduced a redundancy prevention cache to avoid duplicate API requests and avoid unnecessary execution cycles. The performance of the proposed architecture is evaluated on the standard benchmark programs related to C and Python programs. FeedbackLLM demonstrated more line and branch coverage than baseline tools while scaling linearly in execution time.

Large language model (LLM)-powered multi-agent systems (MAS) enable agents to communicate and share information, achieving strong performance on complex tasks. However, this communication also creates an attack surface where malicious agents can propagate misinformation and manipulate group decisions, undermining MAS safety. Existing embedding-based defenses aim to detect and prune suspicious agents, but their effectiveness depends on a clear separation between the text embeddings of malicious and benign messages. Attackers can circumvent such defenses by crafting messages whose embeddings lie close to benign ones. We analyze this failure mode theoretically and validate it empirically with three attacks, Slow Drift, Benign Wrapper, and Chaos Seeding. Our analysis further reveals a fundamental limitation of embedding-based defenses: because they rely solely on the text embeddings, they ignore token-level confidence signals such as logits, which can remain informative when embeddings are not distinguishable under attack. We propose using confidence scores to prune or down-weight messages during MAS communication. Experiments show improved robustness across models, datasets, and communication topologies. Moreover, we find that the effectiveness of confidence signals decays over communication rounds, highlighting the importance of early intervention. This insights can inform and inspire future work on MAS attacks and defenses.

Large language models (LLMs) often achieve strong performance on reasoning benchmarks, but final-answer accuracy alone does not show whether they faithfully execute the procedure specified in a prompt. We introduce a controlled diagnostic benchmark for procedural execution, where models are given a step-wise arithmetic procedure and two numeric inputs, and must return the final computed value. Complexity is varied through procedure length and look-back dependencies over intermediate variables. Average first-answer accuracy drops from 63% on 5-step procedures to 20% on 95-step procedures. Generation-level analysis shows that failures often involve missing answers, premature answers, self-correction after an initial error and under-executed traces. These findings suggest that apparent reasoning ability can mask substantial weaknesses in faithful long-horizon procedural execution.

Agentic AI architectures augment LLMs with external tools, unlocking strong capabilities. However, tool use is not always beneficial; some calls may be redundant or even harmful. Effective tool use, therefore, hinges on a core LLM decision: whether to call or not call a tool when performing a task. This decision is particularly challenging for web search tools, where the benefits of external information depend on the model's internal knowledge and its ability to integrate potentially noisy tool responses. We introduce a principled framework inspired by decision-making theory to evaluate web search tool-use decisions along three key factors: necessity, utility, and affordability. Our analysis combines two complementary lenses: a normative perspective that infers true need and utility from an optimal allocation of tool calls, and a descriptive perspective that infers the model's self-perceived need and utility from their observed behaviors. We evaluate six open and one closed-source frontier models under two harnesses, one conditioning on only the current turn and its search results, the other on the full execution traces, across four web-search tools and three tasks. In every setting, we find that a model's perceived need and utility are frequently misaligned with the true need and utility. Building on this framework, we train lightweight estimators of need and utility from the models' hidden states. These estimators drive simple controllers that improve decision quality and yield stronger task performance than the self-perceived baseline for most of the open-source models.

Much research on LLMs has focused on increasing benchmark performance. However, the evaluation of such models in real-world collaborative human-AI workflows has stayed behind. This work evaluates a chatbot-style assistant based on Retrieval-Augmented Generation (RAG) in a realistic multi-turn information-seeking scenario inspired by workplace settings where compliance with local legislation and secure handling of sensitive data are often key. Specifically, we examine the performance of humans (N=112) assisted by RAG-assistants compared to LLM-only or LLM+RAG baselines. In this setting, we investigate how underlying model size (3B, 8B, and 70B) shapes the human-AI collaborative dynamic and how it influences perceived usability and satisfaction. Results show that the performance gain of human-AI collaboration over the model-only baselines is significant, irrespective of model size, suggesting that hybrid systems are beneficial in information-seeking scenarios. Interestingly, however, perceived usability and satisfaction among participants showed little difference across model sizes. This demonstrates a nuanced trade-off between model size, performance, and user perception. Our work highlights the added value of evaluating AI applications in actual multi-turn interactions with human users, looking at usability and satisfaction besides accuracy, rather than focusing on benchmark performance only.

AI agents execute tens to hundreds of chained LLM calls per task, yet GPU schedulers treat each call as independent, discarding gigabytes of intermediate state between steps and inflating end-to-end latency by 3-8x. We argue that this request-level abstraction is fundamentally mismatched to compound AI workloads, and propose a shift to program-level scheduling: treating the entire agent workflow (not individual inference calls) as the first-class schedulable unit. We present SAGA, a distributed scheduler that implements this abstraction through three mechanisms: (1) Agent Execution Graphs that capture workflow structure to predict KV cache reuse across tool-call boundaries, achieving within 1.31x of Bélády's optimal offline policy; (2) session-affinity batching with work stealing that co-locates correlated requests while maintaining global load balance; and (3) Agent Fair Share, a task-completion-time fairness metric with provable bounded-deviation guarantees. On a 64-GPU cluster serving SWE-bench coding agents and WebArena browser tasks, SAGA reduces task completion time by 1.64x (geometric mean, p < 0.001) over vLLM v0.15.1 with prefix caching and affinity routing, while improving GPU memory utilization by 1.22x and achieving 99.2% SLO attainment under multi-tenant interference. These latency gains come at a quantified cost: approximately 30% lower peak throughput than throughput-optimal batch scheduling, a tradeoff appropriate for the latency-sensitive interactive deployments that dominate compound AI usage. Our results demonstrate that workflow-aware scheduling is essential for efficient compound AI serving.

The operational landscape of local Large Language Model (LLM) inference has shifted from lightweight models to datacenter-class weights exceeding 70B parameters, creating profound systems challenges for consumer hardware. This paper presents a systematic empirical analysis of the Nvidia and Apple Silicon ecosystems, specifically characterizing the distinct intra-architecture trade-offs required to deploy these massive models. On the Nvidia Blackwell architecture, we identify a critical "Backend Dichotomy" within the TensorRT-LLM stack: while the new NVFP4 quantization format delivers a 1.6x throughput advantage over optimized BF16 baselines (151 tokens/s vs. 92 tokens/s), realizing this performance requires navigating complex runtime constraints that trade startup latency for generation speed. Furthermore, we characterize the "VRAM Wall" for 70B+ models: on discrete GPUs, users face a destructive choice between aggressive quantization (e.g., Q2) that degrades model intelligence to fit in VRAM, or PCIe-bottlenecked CPU offloading, which reduces throughput by over 90% compared to full-GPU execution. Conversely, Apple's Unified Memory Architecture (UMA) circumvents these bottlenecks, enabling linear scaling for 80B parameter models at practical 4-bit precisions. This architectural divergence extends to operational sustainability, where Apple's SoC design demonstrates up to a 23x advantage in energy efficiency (tokens/joule). We conclude that for consumer-grade inference, the optimal hardware is defined by a complex interplay between compute density (Nvidia) and memory capacity (Apple), moderated by the significant "ecosystem friction" of proprietary quantization workflows.

Leveraging continuous solar energy harvesting at high efficiency, space data centers are envisioned as a promising platform for executing energy-intensive large language models (LLMs). Recognizing this advantage, space and AI conglomerates (e.g., SpaceX, Google) are actively investing in this vision. One key challenge, however, is the efficient distributed deployment of a large-scale LLM in a satellite network due to the limited onboard computing and communication resources. This gives rise to a placement problem that involves partitioning and mapping model components to satellites such that the fundamentally different model architecture and network topology can be reconciled to ensure low-latency token generation. To address this problem, we present the Space Network of Mixture-of-Experts (SpaceMoE) framework targeting the distributed execution of a popular mixture-of-experts (MoE) model in space. The proposed placement strategies are two-level: (1) layer placement, which assigns MoE layers to satellite subnets; and (2) intra-layer expert placement, which assigns individual experts to satellites associated with the same layer/subnet. For layer placement, we exploit the ring-like communication pattern of autoregressive inference to partition the satellite constellation along the orbiting direction into subnets arranged on a ring, each hosting one MoE layer. Based on this architecture, we formulate and solve an optimization problem for intra-layer expert placement to map experts with heterogeneous activation probabilities onto satellites. The derived strategy reveals an intuitive principle: a frequently activated expert should be mapped to a satellite on a routing path with low expected latency. Experiments over a thousand-satellite constellation show that SpaceMoE achieves at least a threefold latency reduction compared with conventional random and ablation-based placement strategies.

Large Language Models have achieved remarkable success and are increasingly deployed in critical applications involving tabular data, such as Table Question Answering. However, their robustness to the structure of this input remains a critical, unaddressed question. This paper demonstrates that modern LLMs exhibit a significant vulnerability to the layout of tabular data. Specifically, we show that semantically-invariant permutations of rows and columns - rearrangements that do not alter the table's underlying information - are sometimes sufficient to cause incorrect or inconsistent model outputs. To systematically probe this vulnerability, we introduce Adversarial Table Permutation, a novel, gradient-based attack that efficiently identifies worst-case permutations designed to maximally disrupt model performance. Our extensive experiments demonstrate that ATP significantly degrades the performance of a wide range of LLMs. This reveals a pervasive vulnerability across different model sizes and architectures, including the most recent and popular models. Our findings expose a fundamental weakness in how current LLMs process structured data, underscoring the urgent need to develop permutation-robust models for reliable, real-world applications.

Developing effective test cases capable of thoroughly exercising large-scale software systems is inherently difficult, especially if such systems have voluminous, complex, and deeply nested source codes. In this work, we present a novel approach for generating test cases using a reinforcement learning-driven agentic framework where Proximal Policy Optimization (PPO) is coupled with an LLM engine to guide prompt selection during test generation. Our approach consists of two phases. In Phase I, the ToT-guided optimization agent partitions and minimizes the source code by removing redundancies without changing the functional behavior of the source code. In Phase II, a PPO-based policy network is trained to solve the problem of selecting prompts among eight different prompting techniques, such as Boundary Value Analysis, Random Fuzzing, etc., based on the inputted 11-dimensional state vector representing the source code complexity metrics and live coverage metrics to direct the LLM engine towards exploring unvisited paths in the program. The PPO agent receives rewards based on a combination of increases in line and branch coverages, penalties for unexplored branches, and rewards for reducing source code length. From experiments conducted on twenty benchmark programs, it is evident that the proposed approach, PPO-LLM, outperforms CBMC, kS-LLM, and kS-LLM++ in terms of branch and line coverage in almost all cases, for various loop bound values ranging from BOUND1 to BOUND2000. While at BOUND~1, the coverage of branches is 100% using PPO-LLM on the PALS suite, in comparison, it is around 86.8% using kS-LLM++. This confirms that adaptive prompt selection driven by PPO substantially outperforms static prompting strategies on PALS type programs.

Model ensembling is a well-established technique for improving the performance of machine learning models. Conventionally, this involves averaging the output distributions of multiple models and selecting the most probable label. This idea has been naturally extended to large language models (LLMs), yielding improved performance but incurring substantial computational cost. This inefficiency stems from directly applying conventional ensemble implementation to LLMs, which require a separate forward pass for each model to explicitly compute the ensemble distribution. In this paper, we propose the Mixture-model-like Ensemble (ME). By reinterpreting the ensemble as a mixture model, ME stochastically selects a single model at each step to generate the next token, thereby avoiding the need to explicitly compute the full ensemble distribution. ME is mathematically equivalent to sampling from the ensemble distribution, but requires invoking only one model, making it 1.78x-2.68x faster than conventional ensembling. Furthermore, this perspective connects LLM ensembling and token-level routing methods, suggesting that LLM ensembling is a special case of routing methods. Our findings open new avenues for efficient LLM ensembling and motivate further exploration of token-level routing strategies for LLMs. Our code is available at https://github.com/Kamichanw/Mixture-model-like-Ensemble.

Large language models (LLMs) are increasingly tasked with strategic decision-making under incomplete information, such as in negotiation and policymaking. While LLMs can excel at many such tasks, they also fail in ways that are poorly understood. We shed light on these failures by uncovering two fundamental gaps in the internal mechanisms underlying the decision-making of LLMs in incomplete-information games, supported by experiments with open-weight models Llama 3.1, Qwen3, and gpt-oss. First, an observation-belief gap: LLMs encode internal beliefs about latent game states that are substantially more accurate than their own verbal reports, yet these beliefs are brittle. In particular, the belief accuracy degrades with multi-hop reasoning, exhibits primacy and recency biases, and drifts away from Bayesian coherence over extended interactions. Second, a belief-action gap: The implicit conversion of internal beliefs into actions is weaker than that of the beliefs externalized in the prompt, yet neither belief-conditioning consistently achieves higher game payoffs. These results show how analyzing LLMs' internal processes can expose systematic vulnerabilities that warrant caution before deploying LLMs in strategic domains without robust guardrails.

Large Language Models (LLMs) have rapidly influenced many aspects of society, particularly education, due to their demonstrated ability to complete assignments and examinations across a wide range of subjects. Although prior studies have examined the educational impact of LLMs, much of the existing work relies on public or open problem datasets and lacks topic-specific analysis. In engineering education, especially within mechanical engineering, systematic investigations of LLM performance on specific problem types remain limited. Instead of using traditional methods that directly ask textbook questions to an LLM tool, our study adopts a model distillation process to evaluate LLM capabilities in solving statics problems. By distilling ChatGPT, we extracted 25 text-only statics questions and further constructed two additional datasets by adding diagrams and modifying their numerical values. Experimental results show that while LLMs perform well on text-only statics problems, their accuracy decreases when diagrams are introduced and the problems require multi-step reasoning. Further analysis suggests that this performance drop is not primarily caused by limitations in image recognition, but rather by difficulties in multi-step reasoning and in consistently applying extracted visual information across successive solution stages.

Reinforcement learning (RL) has become essential to the post-training of large language models (LLMs) for reasoning, agentic capabilities and alignment. Successful RL relies on sufficient exploration of diverse actions by the model during training, which creates a potential failure mode: a model could strategically alter its exploration during training to influence the subsequent training outcome. In this paper we study this behavior, called exploration hacking. First, we create model organisms of selective RL resistance by fine-tuning LLMs to follow specific underperformance strategies; these models can successfully resist our RL-based capability elicitation in agentic biosecurity and AI R&D environments while maintaining performance on related tasks. We then use our model organisms to evaluate detection and mitigation strategies, including monitoring, weight noising, and SFT-based elicitation. Finally, we show that current frontier models can exhibit explicit reasoning about suppressing their exploration when provided with sufficient information about their training context, with higher rates when this information is acquired indirectly through the environment. Together, our results suggest exploration hacking is a possible failure mode of RL on sufficiently capable LLMs.

Recent large language models (LLMs) have achieved impressive reasoning milestones but continue to struggle with high computational costs, logical inconsistencies, and sharp performance degradation on high-complexity problems. While neuro-symbolic methods attempt to mitigate these issues by coupling LLMs with symbolic reasoners, existing approaches typically rely on monotonic logics (e.g., SMT) that cannot represent defeasible reasoning -- essential components of human cognition. We present "LLM+ASP," a framework that translates natural language into Answer Set Programming (ASP), a nonmonotonic formalism based on stable model semantics. Unlike prior "LLM+ASP" approaches that require manually authored knowledge modules, domain-specific prompts, or evaluation restricted to single problem classes, our framework operates without any per-task engineering and applies uniformly across diverse reasoning tasks. Our system utilizes an automated self-correction loop where structured feedback from the ASP solver enables iterative refinement. Evaluating across six diverse benchmarks, we demonstrate that: (1) stable model semantics allow LLMs to naturally express default rules and exceptions, outperforming SMT-based alternatives by significant margins on nonmonotonic tasks; (2) iterative self-correction is the primary driver of performance, effectively replacing the need for handcrafted domain knowledge; (3) compact in-context reference guides substantially outperform verbose documentation, revealing a "context rot" phenomenon where excessive context hinders constraint adherence.

Large Language Models (LLMs) are increasingly used as core dependencies in software systems. However, the hosted LLM services evolve continuously through provider-side updates without explicit version changes. These silent updates can introduce behavioral drift, causing regressions in functionality, formatting, safety constraints, or other application-specific requirements. Existing approaches focus primarily on regression testing or versioning but do not provide deployer-side mechanisms for governing compatibility during opaque model evolution. This paper proposes a deployment-side governance framework based on three components: clearly defined rules for how the model is allowed to behave (production contracts), focused testing organized by deployment risk categories (risk-category-based testing suite), and release checkpoints that block updates unless they meet defined safety and performance standards (compatibility gates). Through exploratory validation across multiple LLM versions, we provide evidence that targeted testing in specific risk areas can uncover performance regressions that overall metrics miss. We also identify several open research challenges, including how to systematically build effective test suites, how to set reliable performance thresholds in non-deterministic systems, and how to detect and explain model drift when providers offer limited transparency. Overall, we frame LLM update management as a software supply chain governance problem and outline a research agenda for putting deployer-side compatibility controls into practice.

In black-box large language model (LLM) services, response reliability is often only partially observable at decision time, while stronger inference pathways incur substantial computational cost, inducing a budgeted sequential decision problem: for each request, the system should decide whether the default low-cost response is sufficiently reliable or whether additional computation should be allocated to improve response quality. In this paper, we propose \textbf{Ver}ifiable \textbf{O}bservations for Risk-aware \textbf{I}nference \textbf{C}ontrol (\textsc{Veroic}), a framework for adaptive inference control in black-box LLM settings, which formulates request-time control as a \textit{partially observable Markov decision process} to capture partial observability and sequential budget coupling. It constructs a lightweight verifiable observation channel from the input-output pair by aggregating heterogeneous quality signals into a belief state over latent response reliability, which is then used by a budget-aware policy to decide whether to return the default output or trigger a higher-cost inference pathway. Experiments on diverse tasks show that \textsc{Veroic} achieves improved quality-cost trade-offs, stronger risk estimation and calibration, and more robust long-horizon inference control than competitive baselines.

In natural language processing, the entropy of a language is a measure of its unpredictability and complexity. The first study on this subject was conducted by Claude Shannon in 1951. By having participants predict the next character in a sentence, he was able to approximate the entropy of the English language. Several follow-up studies by other authors have since been conducted for English, and one for Hebrew. However, to date, Shannon's experiment has never been conducted for Ukrainian. In this paper, we perform this experiment for Ukrainian by recruiting 184 volunteers using social media channels. We rely on techniques used for English to approximate the entropy value of Ukrainian. The final result is an upper bound of $H_{upper}\approx1.201$ bits per character. We compare this to the performance of current Large Language Models. The methods and code used are also documented and published, along with a discussion of the main challenges encountered.

We adapted the Reliable Change Index (RCI; Jacobson and Truax, 1991) from clinical psychology to item-level LLM version comparison on 2,000 MMLU-Pro items (K=10 samples at T=0.7). Two within-family pairs were tested: Llama 3 to 3.1 (+1.6 points) and Qwen 2.5 to 3 (+2.8 points). On the full benchmark, most items showed no reliable change (79% and 72%). However, over half the items were floor/ceiling. Among analysable items, change was bidirectional with large effect sizes: 34% improved and 28% deteriorated for Llama; 47% improved and 39% deteriorated for Qwen (median |delta p| = 0.50 and 0.90). Churn was asymmetric by difficulty: low-accuracy items improved, high-accuracy items deteriorated. Domain-level decomposition revealed family-specific reversals: Llama lost physics while Qwen lost law. Greedy single-shot evaluation missed 42% of reliably changed items and falsely flagged 25% of unchanged items. The aggregate accuracy gain is the net residual of opposing item-level movements. We recommend reporting churn rate alongside aggregate accuracy.

Perturbation probing generates task-specific causal hypotheses for FFN neurons in large language models using two forward passes per prompt and no backpropagation, followed by a one-time intervention sweep of about 150 passes amortized across all identified neurons. Across eight behavioral circuits, 13 models, and four architecture families, we identify two circuit structures that organize LLM behavior. Opposition circuits appear when RLHF suppresses a pre-training tendency. In safety refusal, about 50 neurons, or 0.014 percent of all neurons, control the refusal template; ablating them changes 80 percent of response formats on 520 AdvBench prompts while producing near-zero harmful compliance, 3 of 520 cases, all with disclaimers. Routing circuits appear for pre-training behaviors distributed through attention. For language selection, residual-stream direction injection switches English to Chinese output on 99.1 percent of 580 benchmark prompts in the 3 of 19 tested models that satisfy three observed conditions: bilingual training, FFN-to-skip signal ratio between 0.3 and 1.1, and linear representability. The same intervention fails on the other 16 models and on math, code, and factual circuits, defining the limits of directional steering. The FFN-to-skip signal ratio, computed from the same two forward passes, distinguishes the two structures and predicts the appropriate intervention. Circuit topology varies by architecture, from Qwen's concentrated FFN bottleneck to Gemma's normalization-shielded circuit. In Qwen3.5-2B, ablating 20 neurons eliminates multi-turn sycophantic capitulation, while amplifying 10 related neurons improves factual correction from 52 percent to 88 percent on 200 TruthfulQA prompts. These results show that perturbation probing offers mechanistic insight into RLHF-organized behavior and a practical toolkit for precision template-layer editing.

Recent advances in multimodal large language models (LLMs) have revolutionized web agents that can automate complex tasks on websites. However, their accuracy remains limited by the scarcity of high-quality web trajectory training data. Existing automatic trajectory generation methods suffer from incomplete website coverage due to homepage-based task proposals or random-walk exploration. Such methods often result in hallucinated or ambiguous task synthesis that lead to incomplete and unreliable trajectory generation. Here, we present AutoSurfer, a comprehensive web trajectory generator that addresses these limitations through three key innovations. First, AutoSurfer employs a systematic breadth-first exploration strategy that maintains a queue of discovered pages and action traces, propagates knowledge across pages to avoid redundant exploration, and recursively expands multi-level graphical user interface elements - closely resembling how a human would learn a new website. Second, AutoSurfer leverages the exploration trajectory to guide task synthesis, reducing hallucinations by grounding complex tasks in actual navigation paths rather than isolated actions or page content alone. Third, AutoSurfer uses the same exploration trajectory as hints to steer a web agent toward more accurate and reliable trajectory refinement. Together, these innovations enable AutoSurfer to comprehensively cover a website's action space and generate data suitable for training website-specific LLMs. We evaluate AutoSurfer on the WebArena benchmark by fine-tuning Qwen2.5-VL-7B-Instruct and demonstrate that it outperforms state-of-the-art methods - Explorer, OS-Genesis, and SynthAgent - achieving up to 24.23% overall task completion accuracy compared to 19.59% for the best prior method. Further, task diversity analysis demonstrates that AutoSurfer yields a more diverse distribution of synthesized tasks.

We present a framework for migrating production Large Language Model (LLM) based systems when the underlying model reaches end-of-life or requires replacement. The key contribution is a Bayesian statistical approach that calibrates automated evaluation metrics against human judgments, enabling confident model comparison even with limited manual evaluation data. We demonstrate this framework on a commercial question-answering system serving 5.3M monthly interactions across six global regions; evaluating correctness, refusal behavior, and stylistic adherence to successfully identify suitable replacement models. The framework is broadly applicable to any enterprise deploying LLM-based products, providing a principled, reproducible methodology for model migration that balances quality assurance with evaluation efficiency. This is a capability increasingly essential as the LLM ecosystem continues to evolve rapidly and organizations manage portfolios of AI-powered services across multiple models, regions, and use cases.

As Competency-Based Education (CBE) is gaining traction around the world, the shift from marks-based assessment to qualitative competency mapping is a manual challenge for educators. This paper tackles the bottleneck issue by suggesting a "Human-in-the-Loop" benchmarking framework to assess the effectiveness of multiple LLMs in automating secondary-level mathematics assessment. Based on the Grade 10 Optional Mathematics curriculum in Nepal, we created a multi-dimensional rubric for four topics and four cross-cutting competencies: Comprehension, Knowledge, Operational Fluency, and Behavior and Correlation. The multi-provider ensemble, consisted of open-weight models -- Eagle (Llama 3.1-8B) and Orion (Llama 3.3-70B) -- and proprietary frontier models Nova (Gemini 2.5 Flash) and Lyra (Gemini 3 Pro), was benchmarked against a ground truth defined by two senior mathematics faculty members (kappa_w = 0.8652). The findings show a marked "Architecture-compatibility gap". Although the Gemini-based Mixture-of-Experts (Sparse MoE) models achieved "Fair Agreement" (kappa_w ~ 0.38), the larger Orion (70B) model exhibited "No Agreement" (kappa_w = -0.0261), suggesting that architectural compliance with instruction constraints outweighs the scale of raw parameters in rubric-constrained tasks. We conclude that while LLMs are not yet suitable for autonomous certification, they provide high-value assistive support for preliminary evidence extraction within a "Human-in-the-Loop" framework.

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 models hallucinate--producing fluent, confident, factually wrong outputs--with a consistency that persists across generations and scales. Existing taxonomies classify hallucination by output type, distinguishing intrinsic from extrinsic failures and faithfulness from factuality divergence. These frameworks are descriptively rigorous but do not identify which internal mechanism produced a given instance. This paper analyses hallucination as a structural consequence of three architectural decisions that together form a compound failure system. Self-attention's co-occurrence learning substitutes statistical proximity for semantic meaning and produces entity confusion, fact misattribution, and semantic drift. The maximum likelihood estimation training objective optimises next-token probability without factual constraint, rewarding statistically plausible outputs regardless of their truth value. Autoregressive decoding's permanent left-to-right commitment under exposure bias ensures that a single wrong token cascades forward through the entire output sequence without revision. Dataset pathologies--long-tail deficiencies, training bias, and synthetic pollution--amplify these vulnerabilities but do not independently cause them. We make three contributions. First, we map each mechanism to a specific output category in the Alansari and Luqman taxonomy, locating intrinsic hallucination in self-attention, extrinsic hallucination in MLE, and logical inconsistency in autoregressive decoding. Second, we show that each commonly cited dataset pathology exploits one of these mechanisms rather than originating hallucination independently. Third, we identify the diagnostic limitation of output-type-only classification and contrast it with inference-layer mitigation approaches.

Context: Study screening in systematic literature reviews is costly, inconsistency-prone, and risk-asymmetric, since false negatives can compromise validity. Despite rapid uptake of Large Language Models (LLMs), there is limited evidence on how such models behave during the study screening phase, particularly regarding the choice of specific LLMs and their comparison with classical models. Objective: To assess LLM performance and variability in screening, quantify the impact of input metadata (abstract, title, keywords), and compare LLMs with classical classifiers under a shared protocol. Methods: We analyzed 12 LLMs from 4 providers (OpenAI, Google Gemini, Anthropic, Llama) and 4 classical models (Logistic Regression, Support Vector Classification, Random Forest, and Naive Bayes) on 2 real Systematic Literature Reviews (SLRs), totaling 518 papers. The experimental design investigated 3 critical dimensions: (i) LLMs performance variability, (ii) the impact of input feature composition (abstract, title, and keywords) on LLM performance, and (iii) the real gain of using LLMs instead of more traditional classification models. Results: LLMs exhibited substantial heterogeneity and residual non-determinism even at temperature zero. Abstract availability was decisive: removing it consistently degraded performance, while adding title and/or keywords to the abstract yielded no robust gains. Compared to classical models, performance differences were not consistent enough to support generalizable LLM superiority. Discussion: LLM adoption should be justified by operational and governance constraints (reproducibility, cost, metadata availability), supported by pilot validation and explicit reporting of variability and input configuration.

With recent semantic query processing engines, semantic aggregation has become a primitive operator, enabling the reduction of a relation into a natural language aggregate using an LLM. However, the resulting semantic aggregate may contain claims that are not grounded in the underlying relation. Verifying such claims is challenging: they often involve quantifiers, groupings, and comparisons over relations that far exceed LLM context windows and require a costly combination of semantic and symbolic processing. We present Evergreen, a system that recasts claim verification as a semantic query processing task with tailored optimizations and provenance capture. Evergreen compiles each claim into a declarative semantic verification query that can execute on the same query engine used to produce the aggregate. To reduce cost, Evergreen avoids unnecessary LLM calls through verification-aware optimizations, including early stopping, relevance sorting, and estimation with confidence sequences, as well as general-purpose optimizations for semantic queries, such as operator fusion, similarity filtering, and prompt caching. Each verdict is accompanied by citations that identify a minimal set of tuples justifying the result, with semantics based on semiring provenance for first-order logic. On a benchmark of production-inspired workloads over restaurant review and customer support datasets, Evergreen's optimized configurations occupy the entire cost-quality Pareto frontier. With a strong LLM, Evergreen preserves verification quality at an F1 of 0.94 while reducing cost by 3.1x relative to unoptimized verification; with a substantially weaker LLM, it surpasses the strongest external baseline's F1 (0.87 vs. 0.83) at 7.0x lower cost.

Large language models (LLMs) are increasingly used in emotionally sensitive human-AI applications, where reliable emotion detection is essential. However, their emotion recognition abilities remain uneven: models often perform well on some emotions while consistently struggling with others. Although recent work has explored emotion mechanisms in LLMs, little is known about why models are weaker on some emotions than others from a mechanistic interpretability perspective. In this work, we investigate emotion-specific biases through the causal mechanisms of emotion inference using sparse autoencoders (SAEs). We systematically identify causal sparse emotion features that drive emotion inference and analyze their sparse causal organization within and across emotions. We show that some emotions, such as surprise and fear, rely on highly concentrated feature sets, whereas disgust exhibits a more distributed sparse causal organization: its causal features are generally weaker, frequently co-activate with features for other emotions, and are often overshadowed by causal features for anger. These representational differences provide a mechanistic explanation for why LLMs struggle more with certain emotions. Finally, we conduct two intervention experiments: targeted steering of weaker causal features to mitigate emotion-specific failures, and global optimization of a steering vector over the identified causal features to improve overall emotion recognition performance.

We report a striking statistical regularity in frontier LLM outputs that enables a CPU-only scoring primitive running at 2.6 microseconds per token, with estimated latency up to 100,000$\times$ (five orders of magnitude) below existing sampling-based detectors. Across six contemporary models from five independent vendors, two generation sizes, and five held-out domains, token rank-frequency distributions converge to the same two-parameter Mandelbrot ranking distribution, with 34 of 36 model-by-domain fits exceeding $R^{2} = 0.94$ and 35 of 36 favoring Mandelbrot over Zipf by AIC. The shared family does not collapse the models into statistical duplicates. Fitted Mandelbrot parameters remain cleanly separable between models: the cross-model spread in $q$ (1.63 to 3.69) exceeds its per-model bootstrap standard deviation (0.03 to 0.10) by more than an order of magnitude, yielding tens of standard deviations of separation per few thousand output tokens. Two capabilities follow. First, statistical model fingerprinting: text from a vendor-delivered LLM can be tested against its claimed model family without cryptographic watermarks or access to model internals, supporting provenance verification and silent-substitution audits. Second, a model-agnostic reference distribution for black-box output assessment, from which we derive a single-pass scoring primitive that composes with model log probabilities when available and degrades to a rank-only mode usable on closed APIs. Pilot results on FRANK, TruthfulQA, and HaluEval map where the primitive helps (lexical anomalies, unsupported entities) and where it structurally cannot (reasoning errors in domain-appropriate vocabulary). We position the primitive as a first-pass triage layer in compound evaluation stacks, not as a replacement for sampling-based or source-conditioned verifiers.

In recent years, the rapid proliferation of open-source large language models (LLMs) has spurred efforts to turn general-purpose models into domain specialists. However, many domain-specialized LLMs are developed using datasets and training protocols that are not aligned with the nuanced requirements of real-world applications. In the legal domain, where precision and reliability are essential, this lack of consideration limits practical utility. In this study, we propose a systematic training framework grounded in the practical needs of the legal domain, with a focus on Korean law. We introduce LegalMidm, a Korean legal-domain LLM, and present a methodology for constructing high-quality, use-case-driven legal datasets and optimized training pipelines. Our approach emphasizes collaboration with legal professionals and rigorous data curation to ensure relevance and factual accuracy, and demonstrates effectiveness in key legal tasks.

Large Language Models are being increasingly deployed as the decision-making core of autonomous agents capable of effecting change in external environments. Yet, in conversational benchmarks, which simulate real-world customer-centric issue resolution scenarios, these agents frequently fail due to the cascading effects of incorrect decision-making. These challenges are particularly pronounced for open-source LLMs with smaller parameter sizes, limited context windows, and constrained inference budgets, which contribute to increased error accumulation in agentic settings. To tackle these challenges, we present the Failure-Aware Meta-Agentic (FAMA) framework. FAMA operates in two stages: first, it analyzes failure trajectories from baseline agents to identify the most prevalent errors; second, it employs an orchestration mechanism that activates a minimal subset of specialized agents tailored to address these failures by injecting a targeted context for the tool-use agent before the decision-making step. Experiments across open-source LLMs demonstrate performance gains up to 27% across evaluation modes over standard baselines. These results highlight that targeted curation of context through specialized agents to address common failures is a valuable design principle for building reliable, multi-turn tool-use LLM agents that simulate real-world conversational scenarios.

Applications based on large language models (LLMs), such as multi-agent simulations, require population diversity among agents. We identify a pervasive failure mode we term \emph{Persona Collapse}: agents each assigned a distinct profile nonetheless converge into a narrow behavioral mode, producing a homogeneous simulated population. To quantify persona collapse, we propose a framework that measures how much of the persona space a population occupies (Coverage), how evenly agents spread across it (Uniformity), and how rich the resulting behavioral patterns are (Complexity). Evaluating ten LLMs on personality simulation (BFI-44), moral reasoning, and self-introduction, we observe persona collapse along two axes: (1) Dimensions: a model can appear diverse on one axis yet structurally degenerate on another, and (2) Domains: the same model may collapse the most in personality yet be the most diverse in moral reasoning. Furthermore, item-level diagnostics reveal that behavioral variation tracks coarse demographic stereotypes rather than the fine-grained individual differences specified in each persona. Counter-intuitively, \textbf{the models achieving the highest per-persona fidelity consistently produce the most stereotyped populations}. We release our toolkit and data to support population-level evaluation of LLMs.

Knowledge graph (KG) reasoning infers new knowledge from existing facts and is widely applied in question answering, recommendation, and decision support. With the rapid development of large language models (LLMs), LLM-based KG reasoning frameworks have become increasingly popular by leveraging retrieved KG information. However, hallucinations in LLMs remain a critical issue. Even when relevant KG knowledge is incorporated, models may still generate incorrect outputs, leading to misinformation and unreliable decisions. Existing hallucination detection methods either focus on LLM internal states or verify consistency with retrieved contexts, but both overlook the structural information in KGs, resulting in suboptimal performance. To address this gap, we propose LUCID, the first halLUcination deteCtIon method for LLM-based knowleDge graph reasoning frameworks. LUCID jointly leverages LLM attention scores, KG semantics, and structural information. Specifically, it extracts node and edge features from attention scores and semantic similarities, and integrates them with KG structure using a graph neural network. We also construct manually annotated benchmark datasets for evaluation. Experiments on nine datasets show that LUCID achieves state of the art performance compared to 15 baselines.

As large language models (LLMs) are increasingly deployed in high-stakes and operational settings, evaluation strategies based solely on aggregate accuracy are often insucient to characterize system reliability. This study proposes a thermodynamic inspired modeling framework for analyzing the stability of LLM outputs under conditions of uncertainty and perturbation. The framework introduces a composite stability score that integrates task utility, entropy as a measure of external uncertainty, and two internal structural proxies: internal integration and aligned reective capacity. Rather than interpreting these quantities as physical variables, the formulation is intended as an interpretable abstraction that captures how internal structure may modulate the impact of disorder on model behavior. Using the IST-20 benchmarking protocol and associated metadata, we analyze 80 modelscenario observations across four contemporary LLMs. The proposed formulation consistently yields higher stability scores than a reduced utilityentropy baseline, with a mean improvement of 0.0299 (95% CI: 0.02470.0351). The observed gain is more pronounced under higher entropy conditions, suggesting that the framework captures a form of nonlinear attenuation of uncertainty. We do not claim a fundamental physical law or a complete theory of machine ethics. Instead, the contribution of this work is a compact and interpretable modeling perspective that connects uncertainty, performance, and internal structure within a unied evaluation lens. The framework is intended to complement existing benchmarking approaches and to support ongoing discussions in AI safety, reliability, and governance.

We present a deployed system that automates end-to-end customer support workflows inside an enterprise Business Process Management (BPM) platform. The approach is scalable in production and reaches selective automation within two weeks for a new process, leveraging supervision already generated at scale: structured per-case UI interaction traces and low-overhead copilot feedback, where operators either accept a suggestion or provide a correction. A staged deployment pipeline trains a next UI action policy, learns a critic from copilot feedback to calibrate abstention, and executes only high-confidence steps in the background while deferring uncertain decisions to operators and resuming from the updated UI state. This setup lets one operator supervise multiple concurrent sessions and be interrupted only when the system is uncertain. The system operates on a schema-driven view of the BPM interface and includes monitoring and safe fallbacks for production. In production, it automated 45% of sessions and reduced average handling time by 39% without degrading support quality level.

We present JigsawRL, a cost-efficient framework that explores Pipeline Multiplexing as a new dimension of RL parallelism. JigsawRL decomposes each pipeline into a Sub-Stage Graph that exposes the intra-stage and inter-worker imbalance hidden by stage-level systems. On this abstraction, JigsawRL resolves multiplexing interference through dynamic resource allocation, eliminates fragmented utilization by migrating long-tail rollouts across workers, and formulates their coordination as a graph scheduling problem solved with a look-ahead heuristic. On 4-64 H100/A100 GPUs across different agentic RL pipelines and models, JigsawRL achieves up to 1.85x throughput over Verl on synchronous RL, 1.54x over StreamRL and AReaL on asynchronous RL, and supports heterogeneous pipelines with moderate latency trade-off.

Small language models (SLMs) are promising for real-world deployment due to their efficiency and low operational cost. However, their limited capacity struggles with high-stakes legal reasoning tasks that require coherent statute interpretation and logically consistent deduction. Furthermore, training SLMs for such tasks demands high-quality, concise reasoning trajectories, which are prohibitively expensive to manually collect and difficult to curate via standard rejection sampling, lacking granularity beyond final verdicts. To address these challenges, we propose {LegalDrill}, a diagnosis-driven synthesis framework that extracts and iteratively refines reasoning trajectories from a capable teacher via fine-grained prompting, then a self-reflective verification is employed to adaptively select the most effective data for the SLM student. The resulting data empower SLM training through supervised fine-tuning and direct preference optimization. Extensive experiments on several legal benchmarks demonstrate that {LegalDrill} significantly bolsters the legal reasoning capabilities of representative SLMs while bypassing the need for scarce expert annotations, paving a scalable path toward practical legal reasoning systems.

Translating sequential programming priors into the parallel temporal logic of hardware design remains a crucial bottleneck for large language models(LLM). To investigate this, we introduce a new error taxonomy grounded in problem solvability, inspired by cognitive theory. Our taxonomy categorizes failures into syntactic, semantic, solvable functional, and unsolvable functional types. Evaluations reveal a strict empirical ceiling on the VerilogEval benchmark, as frontier models plateau at a 90.8% initial pass rate. These plateaus are defined by unsolvable functional errors, exposing persistent knowledge gaps immune to test time compute scaling. Furthermore, we expose a striking surface convergence gap: optimization readily eliminates syntax errors but concurrently exacerbates deeper functional failures. Our findings demonstrate that alignment techniques merely teach models to compile. While repeated sampling strategies can patch solvable errors, register-transfer level(RTL) coding capacity remains strictly bounded by pretraining knowledge. Addressing challenges in the current LLM based hardware generation pipeline requires more studies in model reasoning rather than alignment interventions.

Early, tool-free prediction of post-synthesis timing remains a key obstacle to rapid RTL iteration. We introduce TimingLLM, a two-stage retrieval-augmented LLM pipeline that estimates worst negative slack (WNS) and total negative slack (TNS) directly from Verilog. Stage 1 is a fine-tuned LLM that acts as a compact post-synthesis timing oracle, producing path-level arrivals/required times that are summarized into lightweight structural-timing cues (e.g., bag-of-gates counts, critical-path depth, gate-type patterns). Stage 2 is an LLM-based regressor that predicts WNS/TNS and applies a learned diagonal steering vector at the last transformer block, computed from the k nearest timing-labeled modules in a disjoint retrieval bank. On VerilogEval, TimingLLM attains R_WNS = 0.91 (MAPE 12%) and R_TNS=0.97 (MAPE 16%) while running 1.3-1.6 times faster than prior methods. Training uses a new 60k-module Verilog corpus with synthesis reports, which we will release. After training once, TimingLLM can be adapted to new technology libraries and PVT corners by refitting only a small regression head on 1000 labeled modules per setting, consistently outperforming state-of-the-art baselines.

We report a reproducible error pattern in GPT-5.4 on OWL2DL compliance queries: the model frequently answers unknown'' when the reasoner-entailed answer is no'' under \emph{FunctionalProperty} closure or class \emph{disjointness}. Using 180 reasoner-audited queries from a procedural expansion of the observed pattern plus 18 hand-authored held-out queries in two unrelated domains (insurance and clinical), we compare four interaction modes under matched query budget: single-shot, three rounds of generic ``you-are-wrong'' retry, three rounds of reasoner-verdict repair with an open-world-assumption (OWA) hint, and the same repair without the hint. Direct faithfulness is 43.9,% (Wilson 95,% CI $[36.8,51.2]$); generic retry reaches 81.7,% ($[75.4,86.6]$); the verdict-with-hint variant is \emph{worse} at 67.2,% ($[60.1,73.7]$); the verdict-only variant reaches 97.8,% ($[94.4,99.1]$). All pairwise comparisons remain significant under McNemar's exact test with Bonferroni correction ($α= 0.01$; all $p < 10^{-5}$). The same fingerprint accounts for 4/4 errors on the held-out queries. Our interpretation is bounded: prompt framing can matter more than corrective content, and reasoner-guided wrappers should be ablated explicitly.

Large language models (LLMs) show promise in medical diagnosis, but real-world deployment remains challenging due to high-stakes clinical decisions and imperfect reasoning reliability. As a result, careful inspection of model behavior is essential for assessing whether diagnostic reasoning is reliable and clinically grounded. However, debugging medical LLMs remains difficult. First, developers often lack sufficient medical domain expertise to interpret model errors in clinically meaningful terms. Second, models can fail across a large and diverse set of instances involving different input types, tasks, and reasoning steps, making it challenging for developers to prioritize which errors deserve focused inspection. Third, developers struggle to identify recurring error patterns across cases, as existing debugging practices are largely instance-centric and rely on manual inspection of isolated failures. To address these challenges, we present VeriLLMed, a visual analytics system that integrates external biomedical knowledge to audit and debug medical LLM diagnostic reasoning. VeriLLMed transforms model outputs into comparable reasoning paths, constructs knowledge graph-grounded reference paths, and identifies three recurring classes of diagnosis errors: relation errors, branch errors, and missing errors. Case studies and expert evaluation demonstrate that VeriLLMed helps developers identify clinically implausible reasoning and generate actionable insights that can inform the improvement of medical LLMs.

We present a comprehensive evaluation of Large Language Models (LLMs) on Computer Science (CS) Data Structure examination questions. Our work introduces a new benchmark dataset comprising exam questions from Tel Aviv University (TAU), curated to assess LLMs' abilities in handling closed and multiple-choice questions. We evaluated the performance of OpenAI's GPT 4o and Anthropic's Claude 3.5, popular LLMs, alongside two smaller LLMs, Mathstral 7B and LLaMA 3 8B, across the TAU exams benchmark. Our findings provide insight into the current capabilities of LLMs in CS education.

We investigate whether explicit belief graphs improve LLM performance in cooperative multi-agent reasoning. Through 3,000+ controlled trials across four LLM families in the cooperative card game Hanabi, we establish four findings. First, integration architecture determines whether belief graphs provide value: as prompt context, graphs are decorative for strong models and beneficial only for weak models on 2nd-order Theory of Mind (80% vs 10%, p<0.0001, OR=36.0); when graphs gate action selection through ranked shortlists, they become structurally essential even for strong models (100% vs 20% on 2nd-order ToM, p<0.001). Second, we identify "Planner Defiance," a model-family-specific failure where LLMs override correct planner recommendations at partial competence (90% override, replicated N=20); Gemini models show near-zero defiance while Llama 70B shows 90%, and models distinguish factual context (deferred to) from advisory recommendations (overridden). Third, full-game evidence confirms inter-agent conventions (+128% over baseline, p=0.003) outperform all single-agent interventions, and individual belief-graph components must be combined to produce gains. Fourth, preliminary scaling analysis (N=10/cell, exploratory) suggests graph depth has diminishing returns: shallow graphs provide the best cost-benefit ratio, while deeper ToM graphs appear harmful at larger player counts (-1.5 pts at 5-player, p=0.029).

Predicting the outcomes of prospective clinical trials remains a major challenge for large language models. Prior work has shown that both traditional correlational predictors, such as random forests and logistic regression, and strong commercial LLMs achieve limited performance on this task. In this paper, we propose DeepImagine, a framework for teaching LLMs biomedical reasoning through successive counterfactual imagining. The central idea is to approximate hidden causal mechanisms of clinical trials by training models to infer how observed trial results would change under controlled perturbations of experimental conditions, such as dosage, outcome measures, study arms, geography, and other trial attributes. To support this objective, we construct both natural and approximate counterfactual pairs from real clinical trials with reported outcomes. For settings where strict counterfactual supervision is available, such as paired outcome measures or dose-ranging study arms within the same trial, we train models with supervised fine-tuning. For broader settings where only approximate counterfactual pairs can be retrieved, we optimize models with reinforcement learning using verifiable rewards based on downstream benchmark correctness. We further augment training with synthetic reasoning traces that provide causally plausible explanations for local counterfactual transitions. Using this pipeline, we train language models under 10B parameters, including Qwen3.5-9B, and evaluate them on clinical trial outcome prediction. We aim to show that DeepImagine consistently improves over untuned language models and traditional correlational baselines. Finally, we aim to show that the learned reasoning trajectories provide interpretable signals about how models represent trial-level mechanisms, suggesting a practical path toward more mechanistic and scientifically useful biomedical language models.

Large language models (LLMs) have become central to modern AI workflows, powering applications from open-ended text generation to complex agent-based reasoning. However, debugging these models remains a persistent challenge due to their opaque and probabilistic nature and the difficulty of diagnosing errors across diverse tasks and settings. This paper introduces a systematic approach for LLM debugging that treats models as observable systems, providing structured, model-agnostic methods from issue detection to model refinement. By unifying evaluation, interpretability, and error-analysis practices, our approach enables practitioners to iteratively diagnose model weaknesses, refine prompts and model parameters, and adapt data for fine-tuning or assessment, while remaining effective in contexts where standardized benchmarks and evaluation criteria are lacking. We argue that such a structured methodology not only accelerates troubleshooting but also fosters reproducibility, transparency, and scalability in the deployment of LLM-based systems.

Large Language Models (LLMs) are increasingly deployed to autonomously solve real-world tasks. A key ingredient for this is the LLM Function-Calling paradigm, a widely used approach for equipping LLMs with tool-use capabilities. However, an LLM calling functions incorrectly can have severe implications, especially when their effects are irreversible, e.g., transferring money or deleting data. Hence, it is of paramount importance to consider the LLM's confidence that a function call solves the task correctly prior to executing it. Uncertainty Quantification (UQ) methods can be used to quantify this confidence and prevent potentially incorrect function calls. In this work, we present what is, to our knowledge, the first evaluation of UQ methods for LLM Function-Calling (FC). While multi-sample UQ methods, such as Semantic Entropy, show strong performance for natural language Q&A tasks, we find that in the FC setting, it offers no clear advantage over simple single-sample UQ methods. Additionally, we find that the particularities of FC outputs can be leveraged to improve the performance of existing UQ methods in this setting. Specifically, multi-sample UQ methods benefit from clustering FC outputs based on their abstract syntax tree parsing, while single-sample UQ methods can be improved by selecting only semantically meaningful tokens when calculating logit-based uncertainty scores.

The promise of data-driven materials discovery remains constrained by the scarcity of large, high-quality, and accessible experimental datasets. Here, we introduce a generalizable large language model (LLM)-powered pipeline for automated extraction and structuring of materials data from unstructured scientific literature, using concrete materials as a representative and particularly challenging example. The pipeline exhibits robust performance across a broad range of LLMs and achieves an $F_1$ score of up to 0.98 for diverse composition--process--property attributes. Within one hour, it extracts nearly 9,000 high-quality records with over 100 attributes from a corpus screened from more than 27,000 publications, enabling the construction of the largest open laboratory database for blended cement concrete. Machine learning analyses underscore the importance of large, diverse, and information-rich datasets for enhancing both in-distribution accuracy and out-of-distribution generalization to unseen materials. The proposed pipeline is readily adaptable to other materials domains and accelerates the development of scalable data infrastructures for materials informatics.

Verification is becoming central to both reinforcement-learning-based training and inference-time control of large language models (LLMs). Yet current verifiers face a fundamental trade-off: LLM-based verifiers are expressive but hard to control and prone to error, while deterministic executable verifiers are reliable and interpretable but often limited in capability. We study the following question: given a development set of LLM outputs and labels for a target objective, such as correctness, can we automatically induce a minimal set of Python verifiers whose joint satisfaction closely matches that objective? We propose AutoPyVerifier, a framework that uses an LLM to synthesize candidate verifier functions and then refines them through search over a directed acyclic graph (DAG). By navigating the DAG, AutoPyVerifier systematically explores the space of deterministic executable verifiers and selects a compact verifier set whose joint satisfaction best approximates the target objective. Across mathematical reasoning, coding, function calling, and instruction-following benchmarks for several state-of-the-art LLMs, AutoPyVerifier improves target-objective prediction by up to 55.0 F1 points over the initial LLM-generated verifier sets. Additional analyses show that the most useful verification targets vary by benchmark and model, and that the DAG-based search shifts the learned verifier sets toward more structural and semantically grounded checks. We further show that exposing the discovered verifier set to an LLM as an external tool improves downstream accuracy by up to 17.0 points. We release our code

Dense vector retrieval is the practical backbone of Retrieval- Augmented Generation (RAG), but similarity search can suffer from precision limitations. Conversely, utility-based approaches leveraging LLM re-ranking often achieve superior performance but are computationally prohibitive and prone to noise inherent in perplexity estimation. We propose Utility-Aligned Embeddings (UAE), a framework designed to merge these advantages into a practical, high-performance retrieval method. We formulate retrieval as a distribution matching problem, training a bi-encoder to imitate a utility distribution derived from perplexity reduction using a Utility-Modulated InfoNCE objective. This approach injects graded utility signals directly into the embedding space without requiring test-time LLM inference. On the QASPER benchmark, UAE improves retrieval Recall@1 by 30.59%, MAP by 30.16% and Token F1 by 17.3% over the strong semantic baseline BGE-Base. Crucially, UAE is over 180x faster than the efficient LLM re-ranking methods preserving competitive performance, demonstrating that aligning retrieval with generative utility yields reliable contexts at scale.

Feature engineering for Electronic Health Records (EHR) is complicated by irregular observation intervals, variable measurement frequencies, and structural sparsity inherent to clinical time series. Existing automated methods either lack clinical domain awareness or assume clean, regularly sampled inputs, limiting their applicability to real-world EHR data. We present \textbf{FeatEHR-LLM}, a framework that leverages Large Language Models (LLMs) to generate clinically meaningful tabular features from irregularly sampled EHR time series. To limit patient privacy exposure, the LLM operates exclusively on dataset schemas and task descriptions rather than raw patient records. A tool-augmented generation mechanism equips the LLM with specialized routines for querying irregular temporal data, enabling it to produce executable feature-extraction code that explicitly handles uneven observation patterns and informative sparsity. FeatEHR-LLM supports both univariate and multivariate feature generation through an iterative, validation-in-the-loop pipeline. Evaluated on eight clinical prediction tasks across four ICU datasets, our framework achieves the highest mean AUROC on 7 out of 8 tasks, with improvements of up to 6 percentage points over strong baselines. Code is available at github.com/hojjatkarami/FeatEHR-LLM.

As Large Language Models (LLMs) are deployed globally, ensuring their safety and alignment across multiple languages becomes paramount. However, safety behaviors often vary unpredictably between languages, posing significant challenges for consistent and ethical AI. In this work, we systematically investigate the dynamics of multilingual alignment, exploring whether single-language alignment transfers cross-lingually, how language consistency is preserved during training, and the resulting trade-offs with general knowledge capabilities. We introduce RefusEU, a novel refusal alignment dataset covering 12 European languages, including a dedicated test set for evaluating current state-of-the-art models. Our controlled Direct Preference Optimization (DPO) experiments provide two key insights: aligning models exclusively in English is insufficient to ensure cross-lingual safety, even for the same harm categories, whereas training on multilingual datasets can improve safety without degrading general performance, as measured by the Global MMLU benchmark.

Large language models (LLMs) are increasingly utilized in clinical reasoning and risk assessment. However, their interpretive reliability in critical and indeterminate domains such as psychiatry remains unclear. Prior work has identified algorithmic biases and prompt sensitivity in these systems, raising concerns about how contextual information may influence model outputs, but there remains no systematic way to assess these, especially in the psychiatric domain. We propose an approach for reliability auditing downstream LLM tasks by structuring evaluation around the impact of prompt design and the inclusion of medically insignificant inputs on predicted hospitalization risk scores, which is often the first downstream AI clinical-decision-making task. In our audit, a cohort of synthetic patient profiles (n = 50) is generated, each consisting of 15 clinically relevant features and up to 50 clinically insignificant features, across four prompt reframings (neutral, logical, human impact, clinical judgment). We audit four LLMs (Gemini 2.5 Flash, LLaMa 3.3 70b, Claude Sonnet 4.6, GPT-4o mini), and our results show that including medically insignificant variables resulted in a statistically significant increase in the absolute mean predicted hospitalization risk and output variability across all models and prompts, indicating reduced predictive stability as contextual noise increased. Clinically insignificant features had an effect on instability across many model-prompt conditions, and prompt variations independently affected the trajectory of instability in a model-dependent manner. These findings quantify how LLM-based psychiatric risk assessments are sensitive to non-clinical information, highlighting the need for systematic evaluations of attributional stability and uncertainty behavior like this before clinical deployments.

Patient-trial matching requires reasoning over long, heterogeneous electronic health records (EHRs) and complex eligibility criteria, posing significant challenges for scalability, generalization, and computational efficiency. Existing approaches either rely on full-document processing with large language models (LLMs), which is computationally expensive, or use traditional machine learning methods that struggle to capture unstructured clinical narratives. In this work, we propose a lightweight framework that combines retrieval-augmented generation and large language model-based modeling for scalable patient-trial matching. The framework explicitly separates two key components: retrieval-augmented generation is used to identify clinically relevant segments from long EHRs, reducing input complexity, while large language models are used to encode these selected segments into informative representations. These representations are further refined through dimensionality reduction and modeled using lightweight predictors, enabling efficient and scalable downstream classification. We evaluate the proposed approach on multiple public benchmarks (n2c2, SIGIR, TREC 2021/2022) and a real-world multimodal dataset from Mayo Clinic (MCPMD). Results show that retrieval-based information selection significantly reduces computational burden while preserving clinically meaningful signals. We further demonstrate that frozen LLMs provide strong representations for structured clinical data, whereas fine-tuning is essential for modeling unstructured clinical narratives. Importantly, the proposed lightweight pipeline achieves performance comparable to end-to-end LLM approaches with substantially lower computational cost.

Understanding what kinds of factual knowledge large language models (LLMs) memorize is essential for evaluating their reliability and limitations. Entity-based QA is a common framework for analyzing non-verbatim memorization, but typical evaluations query each entity using a single canonical surface form, making it difficult to disentangle fact memorization from access through a particular name. We introduce RedirectQA, an entity-based QA dataset that uses Wikipedia redirect information to associate Wikidata factual triples with categorized surface forms for each entity, including alternative names, abbreviations, spelling variants, and common erroneous forms. Across 13 LLMs, we examine surface-conditioned factual memorization and find that prediction outcomes often change when only the entity surface form changes. This inconsistency is category-dependent: models are more robust to minor orthographic variations than to larger lexical variations such as aliases and abbreviations. Frequency analyses further suggest that both entity- and surface-level frequencies are associated with accuracy, and that entity frequency often contributes beyond surface frequency. Overall, factual memorization appears neither purely surface-specific nor fully surface-invariant, highlighting the importance of surface-form diversity in evaluating non-verbatim memorization.

Large language models (LLMs) are increasingly integrated into sensitive workflows, raising the stakes for adversarial robustness and safety. This paper introduces Transient Turn Injection(TTI), a new multi-turn attack technique that systematically exploits stateless moderation by distributing adversarial intent across isolated interactions. TTI leverages automated attacker agents powered by large language models to iteratively test and evade policy enforcement in both commercial and open-source LLMs, marking a departure from conventional jailbreak approaches that typically depend on maintaining persistent conversational context. Our extensive evaluation across state-of-the-art models-including those from OpenAI, Anthropic, Google Gemini, Meta, and prominent open-source alternatives-uncovers significant variations in resilience to TTI attacks, with only select architectures exhibiting substantial inherent robustness. Our automated blackbox evaluation framework also uncovers previously unknown model specific vulnerabilities and attack surface patterns, especially within medical and high stakes domains. We further compare TTI against established adversarial prompting methods and detail practical mitigation strategies, such as session level context aggregation and deep alignment approaches. Our study underscores the urgent need for holistic, context aware defenses and continuous adversarial testing to future proof LLM deployments against evolving multi-turn threats.

While Large Language Models (LLMs) demonstrate remarkable reasoning, complex optimization tasks remain challenging, requiring domain knowledge and robust implementation. However, existing benchmarks focus narrowly on Mathematical Programming and Combinatorial Optimization, hindering comprehensive evaluation. To address this, we introduce OptiVerse, a comprehensive benchmark of 1,000 curated problems spanning neglected domains, including Stochastic Optimization, Dynamic Optimization, Game Optimization, and Optimal Control, across three difficulty levels: Easy, Medium, and Hard. The experiments with 22 LLMs of different sizes reveal sharp performance degradation on hard problems, where even advanced models like GPT-5.2 and Gemini-3 struggle to exceed 27% accuracy. Through error analysis, we identify that modeling & logic errors remain the primary bottleneck. Consequently, we propose a Dual-View Auditor Agent that improves the accuracy of the LLM modeling process without introducing significant time overhead. OptiVerse will serve as a foundational platform for advancing LLMs in solving complex optimization challenges.

Reasoning in large language models is often discussed as a single capability, but some of its gains may stem from simpler underlying operations. We examine two such primitives, recall and state-tracking, through five controlled task families centered on state-based recall, and compare matched transformer and hybrid architectures with and without reasoning augmentation. Across the suite, reasoning-augmented variants substantially outperform instruction-only variants, often by large margins. This pattern is consistent with the State over Tokens view: externalized reasoning traces help because they carry the intermediate state forward in token space. By contrast, hybrid inductive bias does not yield a uniform advantage in accuracy once reasoning tokens are available. When architectural differences do appear, they follow task structure: the hybrid Think model is more robust on strictly sequential chained updates, whereas the transformer Think model is more robust on flat multi-hop retrieval. We therefore cast the main contribution of this study as a descriptive account of what drives performance on state-based recall tasks: reasoning-token augmentation appears to be the dominant factor, while hybrid advantages are narrower, task-dependent, and potentially more about inference efficiency than overall capability. We also release the codebase and data required to reproduce these results.

Third-party Large Language Model (LLM) API gateways are rapidly emerging as unified access points to models offered by multiple vendors. However, the internal routing, caching, and billing policies of these gateways are largely undisclosed, leaving users with limited visibility into whether requests are served by the advertised models, whether responses remain faithful to upstream APIs, or whether invoices accurately reflect public pricing policies. To address this gap, we introduce GateScope, a lightweight black-box measurement framework for evaluating behavioral consistency and operational transparency in commercial LLM gateways. GateScope is designed to detect key misbehaviors, including model downgrading or switching, silent truncation, billing inaccuracies, and instability in latency by auditing gateways along four critical dimensions: response content analysis, multi-turn conversation performance, billing accuracy, and latency characteristics. Our measurements across 10 real-world commercial LLM API gateways reveal frequent gaps between expected and actual behaviors, including silent model substitutions, degraded memory retention, deviations from announced pricing, and substantial variation in latency stability across platforms.