A long-horizon agent's trace outgrows both of its consumers: the human observer monitoring the run, and the agent itself, whose bounded context the trace must be folded back into. We present a live trace model, an append-only event ledger folded incrementally into typed run state and compiled into per-consumer views, and evaluate it for both consumers against deterministic ground truth. For the observer side, evaluated with an LLM reader as proxy, the compiled view answers monitoring questions using approximately 14x and 15x fewer input tokens (by reader) and at 5-7x lower cost than a budget-capped single-call reading of the raw trace, with higher accuracy (0.85-0.87 versus 0.48). Because the questions were co-designed with the view schema, we treat the token and cost reduction, conditional on schema coverage, as the transferable result. For the agent, on 120-link sequential-dependency tasks, mechanisms that maintain the task's running statistic in per-step state succeed where full-context prompting fails (30/30 versus 8/30 under a clean protocol, n=30, labeled descriptive owing to benchmark-system co-development); a prompt-level scratchpad matches the fold's accuracy at lower cost, and a two-arm decomposition attributes the fold's accuracy to its deterministic aggregate and its cost advantage to its compactness. The fold's remaining value over cheaper alternatives is deterministic auditability and serving the observer from the same state. We derive eleven candidate requirements for trace folding from observed failures and delimit them with an order-sensitive task family on which the fold ceases to help. Code, benchmarks, a regenerable synthetic corpus, and all workbench traces are released.
Multi-agent systems built from large language models are deployed widely, yet how much performance is lost when two LLMs must coordinate rather than act alone remains unclear. We formulate the collaboration tax as the team-decentralisation loss of a two-player cooperative game with private information, with two propositions characterising its sign and its equivalence to a max-superadditivity violation. We operationalise this definition on 32 solo-tractable tasks grouped by source of grounding friction and measure it on 11 models from 7 providers. The tax is structured along two no-exception axes: a category ordering across every model and a monotonic decrease with capability. The proximate mechanism is not a reasoning deficit but a four-stage conversational cascade in which agents make ungrounded claims, fail to query the partner, skip integrating both views, and accept the answer without re-derivation. The tax is mechanically predictable from conversation features and partly tractable: a prompt intervention targeting all four stages closes a substantial fraction of the gap, with the dominant bottleneck differing across categories. In heterogeneous pairs the tax is pulled toward the stronger partner rather than the additive midpoint, empirically realising the max-superadditivity violation predicted by our framework. Together these results recast collaboration in LLM systems as a measurable, predictable, and partly tractable cost.
LLM evaluation pipelines often have many candidate judges: general LLM-as-a-judge prompts, reward models, safety classifiers, confidence variants, and task-specific verifiers. The deployment question is not only which judge is best, but which judges should be called, on which examples, and when panel construction should stop. We formulate judge-panel design as a role-conditioned allocation problem. From a small labeled audit set, declared slices, and judge costs, the method estimates target-relative roles: copies add no conditional information, complements improve the global panel, and specialists help only on slices. These roles induce a policy: drop copies, add complements globally, route specialists conditionally, and stop when validation gain falls below a threshold. Across reasoning, code, safety, preference, reward-model, summarization, and math audits, the method is compared with single judges, flat panels, matched diversity heuristics, full-call stacking, reliability juries, and frugal cascades. The result is a regime map for judge calls: route specialists on deployable slices, stop in saturated verifier regimes, keep broad ensembles when their risk benefit is worth the cost, and ignore conditional copies. The output is a reusable, auditable call plan for the next evaluation batch.
Theory of Mind (ToM) is essential for agent interactions, yet existing evaluations either rely on static scenarios that oversimplify mental-state reasoning or interactive settings that provide limited diagnostic insight. We present Avalon-ToM-Bench, a fine-grained benchmark that operationalizes ToM through the asymmetric-information mechanics of The Resistance: Avalon. Rather than evaluating end-to-end gameplay, it decomposes ToM into a 2$\times$2 taxonomy -- epistemic versus motivational reasoning crossed with inference versus action -- using human-crafted, perspective-constrained queries. Benchmarking 28 LLMs reveals three insights: 1) Reasoning, not knowledge. Models show strong game-rule comprehension but markedly weaker ToM abilities, isolating failures to social reasoning rather than missing domain knowledge. 2) Expression, not representation. Mechanistic analyses via linear probing and activation steering show that models frequently represent correct mental-state inferences in their hidden states but fail to express them during generation -- linear probes recover 77-82% accuracy versus 62-70% from the models' own chain-of-thought. 3) Policy, not deliberation. Dedicated reasoning training yields substantial improvements whereas test-time chain-of-thought provides only marginal gains (+11.0 versus +1.1 points on average), suggesting that robust ToM depends on a learned reasoning policy rather than increased inference-time deliberation.
LLM analytics agents are evaluated on SQL syntax accuracy, but production failures look different: questions with two valid business definitions, questions the warehouse cannot answer, deprecated columns after a schema change, and queries that execute successfully while returning the wrong business number. No execution-match metric can score them. This paper introduces WarehouseReliabilityBench, 400 frozen tasks over two synthetic warehouses in which roughly half the correct responses are a clarification, an abstention or a refusal, with pinned denominators and a pre-registered paired bootstrap fixing each claim verb before the numbers existed. QueryProof, a 7B agent, uses rules derived from a semantic layer and physical catalog to determine its behaviour, and gates every answer on deterministic post-execution checks. On an 80-task synthetic test split evaluated once, QueryProof outperforms a direct-prompted 32B baseline by +0.237 [+0.112, +0.375] Business Truth Rate at 71.0% lower cost per correct answer; against a cost-matched few-shot baseline the accuracy gain holds but the cost difference does not resolve. This compares systems rather than model sizes: the 32B baseline receives none of the scaffolding. False success falls from 0.754 to 0.351 of returned answers, and no wrong number was returned on an answerable task (0 of 24), though 13 answers went to questions requiring clarification or abstention. Removing the routing layer changes little (0.562 against 0.537), so the result does not depend on escalation. Routing tuned on validation over-abstains on test, and the fitted confidence model loses to the heuristic it replaced. Resampling template families rather than tasks widens both accuracy intervals to include zero, so the effect's direction is better supported than its magnitude. The gain tracks the deterministic layer, though no component ablation was run.
Multi-agent LLM systems route among model-backed advisors, yet a deployer rarely knows before shipping whether routing will help at all. Prevailing routers optimize a gate's AUC and presume that advisor complementarity suffices. We show that neither determines the deployable gain. We introduce RouteGuard, a deployment-certification framework. Routing gain decomposes as $G = πΔ_E$, and the achievable gain is governed by a conditional-regret functional $Φ$, not by AUC. A finite-sample certification bracket comes with a matching Le Cam lower bound, constant-sharp over the fixed-activity class, and a robustness phase transition. On two benchmarks the framework acts as a guardrail. On RouterBench (11 cross-family models) the verdict depends on the sampling unit: the protocol certifies a gain over GPT-4 under prompt-level sampling and withholds it under workload-cluster resampling, because the gain rests on 3 of 86 workload cells. On OpenRCA (three Gemini advisors) the advisors are statistically redundant: the realized oracle sits at or below the independence baseline in all pools we tested (221 RouterBench pools and three OpenRCA distributions), so the protocol correctly refuses to certify. A pre-registered semi-synthetic control confirms calibration: the protocol certifies a genuine gain once $m \ge m^\star$ and does not certify a true null. Code and frozen artifacts will be released with the published version.
Large language models are increasingly used as synthetic research participants and are often validated by whether their marginal responses resemble human data. We study a fixed panel of sixteen lightweight persona-conditioned GPT-4.1 configurations in repeated strategic games. The panel met preregistered broad-reference condition-mean criteria in three of four repeated-game cells; the sole miss was 0.011 below the lower reference bound. Variation was strongly prompt-indexed, but its share depended on uncertainty assumptions: fixed-panel symmetric-Dirichlet sensitivities produced median between-prompt shares of 63%-71% under Jeffreys alpha=0.5 and 47%-53% under alpha=1, while finite-opportunity plug-in estimates were 85%-96%. Aggregate continuation-probability contrasts were +0.083 and +0.078, with conservative simultaneous 95% intervals [-0.171, +0.330] and [-0.181, +0.330]. The treatment jointly changed the continuation process and its textual representation. A separate wording-and-position operation shifted cooperation from 0/40 to 37/40 in the bare configuration, and a label conflict also revealed representation control. The original persona-level p13 result was not prospectively family-controlled, while a post-adjudication exact gate was structurally underpowered; p13 is therefore a replication target rather than a finding. External review exposed family-error, dependence, construct, and boundary-uncertainty defects, and zero-call reanalysis changed the interpretation without rewriting the historical record. The registered marginal criteria could be passed without precisely estimating the treatment-response object. A public capsule verifies 4,916 confirmatory Phase 3-5 runs with no live model calls. The results concern one fixed model-prompt panel and do not establish human substitutability.
Agentic commerce is moving from concept to deployed infrastructure: payment networks, retailers, and AI platforms are setting the stage for agents to transact on behalf of merchants and consumers. Yet whether the LLMs behind these agents can price competently in real markets, where customer preferences are hidden, competitors adapt in real time, and demand can shift without warning, has not been systematically tested. We introduce Bazaar, a dynamic sealed-bid benchmark for multi-attribute auction under these conditions. Despite its dynamics, the benchmark is grounded in closed-form customer utilities, enabling exact evaluation. Across 11 frontier LLMs from four providers, the leading agents on customer acquisition (e.g. Gemini 3.1 Pro) are often not the leading agents on profit (e.g. Opus 4.6). The ranking shifts again under demand shocks: agents that learned fastest pre-shock are typically the slowest to revise their beliefs afterwards, while Gemini 3.1 Pro recovers fastest despite not leading on profit. However, even the strongest agent captures less than a third of hindsight-optimal profit, suggesting current LLMs are progressing in agentic commerce but leave substantial headroom.
Agentic systems have widened the gap between producing candidate outputs and reviewing them. This paper asks a practical architectural question: should domain specialization be built into an evaluator's weights, or into the rule that decides when its judgment can be trusted? We study 99,952 public, rubric-conditioned examples. Supplying the correct rubric improves locked-test accuracy by 2.11 points over a response-only control; replacing it with an unrelated rubric costs 2.66 points. Dividing the same training corpus among eight criterion-family LoRA judges, however, loses 10.05 points and cuts audited coverage at a 5% risk target from 24.44% to 5.43%. Matching the bank's stored capacity with one rank-64 adapter does not reproduce this loss. Nor is the result explained by learning rate or optimizer steps. Initializing the family adapters from a shared, trained judge recovers test accuracy to 76.85%, 19.94 points above scratch training at the same learning rate (95% interval 18.88-21.02). The result changes when specialization governs deferral rather than judgment. On RewardBench 2, learned correctness heads route examples through a 0.6B-4B-8B cascade without changing any reward score. Across 20 locked repartitions, the cascade attains 89.40% accuracy, compared with 84.75% for 8B alone, at 0.415 normalized parameter compute. Every run passes an exact one-sided 95% risk audit; margin-based rules remain near 84.8% accuracy while using at least 0.94 compute. These results suggest a qualified design rule: share the learning of judgment until there is enough data to justify a split, and place domain-specific adaptation in an audited release boundary.
A self-evolving agentic loop repeatedly proposes a tweaked version of an agent (its prompt template or program) and accepts or rejects the change based on a per-iteration quality signal. Designing that signal is often the costly part of the project: a reliable scalar reward requires domain expertise and labeled examples that are themselves as expensive to assemble as the agent's underlying task. We propose replacing the scalar at the accept/reject gate with a pairwise validator: a frozen LLM that, given the parent and child candidate, returns a binary verdict on which is better. Pairwise judgment is generally easier and more stable than absolute scoring, due to its contrastive nature, which mitigates the need for strict scale calibration. The validator also requires no training of its own. We integrate the validator into three published self-evolving engines (GEPA, ADRS, ShinkaEvolve) and report two flavors: Adaptive Focus, which retains the engine's existing val-set parent selection, and Soft Elo, which lets the validator's verdicts drive parent selection so that val-set rewards drop as well. Across multiple agents and two artifact substrates (prompt and code), our method matches or exceeds the full-reward baseline on the majority of settings we evaluate, and the pattern survives a cross-family validator swap. The pairwise gate is thus a drop-in replacement for per-step reward design at competitive task accuracy without the labeling cost.
Deep research agents are increasingly used to produce long-form financial reports, yet large-scale evaluation remains bottlenecked by the need for human experts to define and execute high-quality rubrics. We address this problem by proposing a scalable pipeline for generating high-quality rubrics without human experts in the final loop. We build a financial deep research benchmark from 104 real-world user queries and automatically synthesize 14,450 query-specific candidate rubrics from model-generated reports. To justify removing human experts from rubric execution, we compare rubric judgments from three human experts with those from a three-LLM judge panel on a sampled subset, and show that LLM-based evaluation is sufficiently consistent with human evaluation to replace it for large-scale rubric screening, including 98.67\% label-level agreement on jointly unanimous items. We then derive consensus-derived gold rubrics through two filters: a strict consistency filter, which keeps a rubric only if the three LLM judges unanimously agree on every report under the same query, and a distinguishability filter, which keeps a rubric only if it assigns at least one majority-yes and at least one majority-no label across the evaluated systems. This process retains 3,687 consistency-passed rubrics, of which 2,600 remain distinguishable and form the final set of consensus-derived gold rubrics. Using this final rubric set, we obtain clearly differentiated rankings across 10 deep research systems, with item-level pass rates ranging from 58.58\% to 22.23\%. More broadly, because the pipeline removes human-expert execution from rubric generation and evaluation, it is naturally scalable for benchmark evaluation, automatic system comparison, and future studies of evaluation-driven system improvement.
Matthew Aitchison, Scott Jeen, Toby Shevlane +1cs.AI
Top AI forecasting systems are approaching superforecaster-level accuracy on future world events, but still rely primarily on off-the-shelf LLMs combined with forecasting-specific context gathering and scaffolding. We study how to improve this recipe through ensembling: given a fixed number of samples, which off-the-shelf model forecasts should be combined to maximize accuracy? On binary questions from the Metaculus AI Benchmark, we find that individual accuracy is not enough: many frontier LLMs make highly correlated predictions, limiting the value of additional forecasts from the same or similar models. Instead, the strongest ensembles combine accurate but diverse forecasters, with models such as \model{Grok 4} contributing disproportionately because their predictions are less correlated with other frontier LLMs. These results suggest that the strength of the AI crowd comes not from sampling more forecasts indiscriminately, but from combining forecasts across models with complementary errors, motivating forecasting systems that explicitly optimize for both model quality and diversity.
We identify intervention bias as a previously unquantified failure mode of zero-shot large-language-model (LLM) educational advisory agents: without task-specific training, they recommend action when a hindsight-optimal oracle policy mandates inaction. In a six-arm ablation on the Open University Learning Analytics Dataset (N=800 students, four temporal cutoffs), at day 56 -- when the oracle designates 70.1% of students as needing no intervention -- zero-shot GPT-4o recommends action for 73%, a 43 percentage-point false-positive rate. Commercial RAG and SQL-augmented retrieval are comparably miscalibrated; at 10,000 students this implies about 4,300 unnecessary advisor contacts per cycle. Supervised policy learning eliminates this bias: a trajectory-conditioned ONNX Decision Transformer (DT) and a snapshot XGBoost classifier, trained on the same oracle-labelled trajectories under strict prefix-only features, both achieve near-zero calibration error. The DT reaches macro-F1 0.79 (macro-recall 0.85) across all five action classes, predicting even the rare load-reduction action without collapsing, at a 0% action flip rate and sub-5 ms CPU decision latency. The two supervised arms are on par; the DT's edge over XGBoost at the final cutoff is indicative only (unpaired across cohorts). Scope: we validate Stage-2 decision-making (EAV state vector to supervised policy) under controlled oracle input from structured OULAD data; high fidelity reflects feature-oracle alignment, not general high-stakes-AI capability. The most robust finding is the intervention-bias contrast, not the absolute accuracies. We also show an Evaluation Gap: LLM-as-judge scoring (DeepEval G-Eval) is blind to intervention bias, rewarding fluent over-prescription rather than decision quality.
Interactive travel planning has become a popular use case for language models. Agents are deployed to manage evolving preferences and unexpected disruptions over multiple turns. Such settings require models to make complex, profile-conditioned planning decisions. However, existing benchmarks often evaluate feasibility, personalization, or interaction in relatively isolated settings. We therefore introduce Trip+ to measure the ability of agents to plan travel holistically. In Trip+, given traveler profiles and dynamic interactions, agents must generate and revise minute-level itineraries. End-to-end traveler experiences are evaluated via an LLM-based simulator, enabling the assessment of subjective metrics like fatigue. Our scenarios range from simple request resolutions to complex environment-driven replanning. We evaluate 18 LMs and find a consistent gap in experiential quality. Models favor technically feasible but exhausting itineraries that diverge sharply from profiled traveler preferences.
Fixed benchmarks are costly to renew and cannot adapt their questions to model-specific failures. We ask whether LLMs can instead discover one another's weaknesses and turn those observations into an evaluation process. To study this question, we introduce \textbf{LivingArena}, an automated peer-probing framework in which models take turns testing one another. Using the interaction history, each questioner identifies potential weaknesses of its opponent and constructs targeted, verifiable questions to probe them. A 3,600-round tournament of ten models reveals a clear role asymmetry: strong answerers are not always reliable questioners, because they may generate internally inconsistent tests or fail to verify their own reference answers. After a questioner exposes an answerer's failure, it is more likely to pursue the same capability domain, while the answerer's weakness recurs on independently generated questions, including questions written by different models. These findings show that peer probing can reveal persistent model-specific weaknesses while separately evaluating answering and reliable test construction. By automating this process and allowing test difficulty to evolve with model capabilities, LivingArena provides a "living" benchmark for model development, red-teaming, and capability-aware multi-agent coordination. We publicly release our code: https://github.com/galaxyChen/LivingArena
When large language models serve as evaluators in multi-agent systems, their strategy preferences -- whether induced by explicit prompts or by shared architectural priors -- propagate through the agent network. We introduce Contagion Networks, a formal framework for measuring how evaluator preferences spread across interacting LLM agents. In a controlled 3-agent experiment using DeepSeek-chat with three distinct evaluator preference profiles (structured, balanced, evidence-based), we measure the Cross-Agent Contagion Matrix Gamma_3 and find that preferences consistently propagate between agents (gamma in [0.157, 0.352]). A neutral-prompt control experiment reveals a counter-intuitive result: shared architectural priors dominate explicit preference prompts as the driver of contagion (rho_neutral = 1.498 vs. rho_mixed = 1.299; prompt contribution: -63.5%). We identify three propagation regimes governed by the spectral radius rho(Gamma_N) and demonstrate that the same agents suppress preference contagion in chain topology (beta_3 = 0.0126 +/- 0.0038, 95% CI [0.0089, 0.0163], n=4 seeds) but cascade in fully-connected topology (Delta H_avg = -0.020) -- a topology-dependent regime transition validated both for homogeneous and cross-model agent pools (rho^cross = 1.296 +/- 0.016, n=4). We show that increasing evaluator committee size from k=1 to k=3 reduces effective contagion by 68.9% +/- 14.1% (n=4 seeds), providing an actionable mitigation strategy. We release the open-source Contagion Network experimental framework.
Large language models (LLMs) are increasingly being used for automated decision-making systems in finance, healthcare, or environmental monitoring. Time series data are ubiquitous in these fields, yet hard to process automatically. Can time series be analyzed by LLM agents? We examine three approaches: providing the agent with raw numerical data, using the LLM as a coding agent, or a combination of both. In the coding agent setup, the model iteratively queries the data using Python code. Using two time series understanding benchmarks, we show that agents with code access can outperform models processing raw data by up to 10%. However, even the best performing agent still answers about 22-34% of the questions incorrectly. To get insights into models' strategies and reasoning gaps, we analyze the model outputs with a strong LLM judge. Our analysis reveals that coding agents can select appropriate statistical tests, but often miss important nuances. Meanwhile, models with access to raw data can reach the right conclusions using back-of-the-envelope calculations.
We present a modular two-agent simulation framework for evaluating conversational shopping assistant architectures. An independent buyer agent, configured with personas, missions, and patience levels, is paired with an interchangeable responder that integrates with a real e-commerce search API. Holding the buyer constant across experiments enables controlled comparison of responder designs on identical scenarios. Using 2011 conversations across 14 persona buckets, we establish four empirical findings. First, rolling-window memory outperforms intent-extraction memory on all quality metrics while being 35% faster per query. Second, illustrating rapid evidence-driven iteration, a systematic failure analysis of a responder version enables targeted fixes that reduce failure and near-failure rates by 62% across the full dataset. Third, swapping the responder LLM backbone from Gemini~2.5 to Llama~3.3~70B costs 0.16--0.45 points despite identical architecture. Finally, we document systematic philosophical disagreement between frontier LLM judges: Gemini rewards process correctness while Claude demands concrete outcomes, despite using the same evaluation prompt.
Automated Planning is a subfield of Artificial Intelligence (AI) where the main objective is generating a sequence of actions, known as a plan, that helps us reach a goal state from an initial state. A planning problem is defined by a set of objects, an initial state and a desired goal state. The objective is to compute a plan that'll lead us from the inital state to the goal state. Programs that generate plans are called planners. In this paper, we did a complementary study to the state-of-the-art LLM called PlanGPT which was released last year. We redid some experiments to verify whether planning with LLMs is \textbf{pertinent} and \textbf{worthwhile}. We also check whether the results obtained in the official PlanGPT paper for plan coverage were correct, and we also performed a more comprehensive study on PlanGPT's performance: in our paper PlanGPT's performance was evaluated using two metrics: Plan Cost and Plan Generation Time. The results of planGPT were compared to those produced by a traditional planner for the same plans and same metrics. We discovered that PlanGPT is no better than a Greedy search strategy.
Benchmarks are fundamental for evaluating and advancing LLMs and MLLMs by providing standardized and explicit measures of performance. However, their construction is labor-intensive and hard to reuse, raising concerns about sustainability and scalability. Moreover, existing benchmarks often quickly reach performance saturation after their release, resulting in insufficient discrimination among state-of-the-art models. To address these challenges, we introduce Benchmark Agent, a fully autonomous agentic system designed for benchmark building. Our framework orchestrates the complete benchmark construction pipeline, from user query analysis and subtask design to data annotation and quality control. To assess Benchmark Agent, we implement it to produce 15 representative benchmarks, spanning diverse evaluation scenarios, including text understanding, multimodal understanding, and domain-specific reasoning. Extensive experiments, including human evaluation, LLM-as-a-judge assessment, and consistency checks, demonstrate Benchmark Agent can generate high-quality benchmark samples with minimal human involvement. More importantly, through continual evaluation, we observe several insightful findings, including that current models struggle with certain domain-specific reasoning tasks. We believe that rapidly evolving benchmarks can contribute significantly to the research community. The preview and code will be publicly available at the demo page and code repository.
AI-Driven Research Systems (ADRS) -- systems coupling LLMs with automated evaluation to discover algorithms, proofs, and designs -- are being optimized and adopted across domains, but the tools to analyze them have not kept pace. ADRS performance depends on component interactions that are poorly understood, expensive to explore, and (as we show) not well captured by standard convergence guarantees. These guarantees rely on structural assumptions that do not hold under the ADRS process we formalize. We introduce GAMBLe, a framework that decomposes ADRS behavior into four parameters (generator $G$, assessor $\mathcal{A}$, discovery mechanism $\mathcal{M}$, budget $B$) and one compositional object, the effective landscape $L_{\text{eff}} = \mathcal{A} \circ G$, which reveals that distinct generator-assessor pairs induce structurally different per-problem optimization landscapes. We exercise the framework on 760+ replicated runs (>46,000 iterations) spanning generators from single LLMs to dynamically-adaptive ensembles, mechanisms from greedy selection to co-evolutionary meta-search, and three NP-hard problems whose assessors range from continuous scoring to cliff functions. The experiments reveal no total ordering of generators or mechanisms: frontier models can underperform open-source alternatives and the simplest mechanism sometimes outperforms state-of-the-art meta-search. Results show that even under limited budgets (60 iterations per run), the right component choices can improve performance by 13-67% and search efficiency by 6-39x.
Fabian Hoppe, Melven Röhrig-Zöllner, Philipp Knechtgescs.AI cs.SE
We consider LLM-based algorithm development through a case study on contractionorder optimisation for tensor networks with OpenEvolve. We pay particular attention to the choice of the LLM as well as design choices such as evaluation metric and test instances. Our results highlight both the promise of verifier-guided evolutionary coding agents for algorithm development/improvement and the continuing importance of evaluation, validation, and interpretation -- and corresponding challenges -- by the human scientist.
Smart homes are evolving toward complex state-dependent living environments, requiring Large Language Models (LLMs) to reason over user intent, preferences, and multi-device interactions. However, existing smart-home benchmarks often focus on static instruction-to-API mapping or limited simulations, failing to evaluate whether LLMs can reason, interact, and act reliably in realistic household scenarios. To address these limitations, we introduce SMH-Bench, a comprehensive benchmark for evaluating LLMs in smart-home environments. Built upon HomeEnv, an executable and verifiable smart-home simulator, SMH-Bench contains 1,100 high-quality tasks spanning 7 categories and 22 fine-grained subcategories. It further stratifies tasks across simple, medium and complex homes, ranging from small apartments to dense multi-room environments with 135 devices. Experiments show that although frontier LLMs achieve strong performance on explicit control and query tasks, they still exhibit significant weaknesses in automation task scheduling, ambiguity handling and personalized reasoning, especially as home complexity increases. We hope SMH-Bench will facilitate the development of more reliable, context-aware, and practically deployable smart-home agents.
Eunsu Kim, Jessica R. Mindel, Kyungjin Kim +1cs.CL
As large language models (LLMs) increasingly shape how users form, refine, and extend their goals, attributing contributions in human-AI collaboration becomes critical for users calibrating their own reliance and for evaluators assessing AI-assisted work. Yet existing methods focus on final artifacts, missing the process through which goals themselves are jointly shaped. We introduce a goal-level attribution framework, CoTrace, that decomposes explicit goals into verifiable requirements and traces both direct contributions and indirect influences across dialogue turns. Applying CoTrace to 638 real-world collaboration logs, we find that while models account for only 11-26% of goal-shaping contribution, they contribute substantially more on introducing lower-level concrete requirements, and make various kinds of indirect contributions. Through controlled simulations, we show that interaction design choices significantly affect model goal-shaping behavior. In a user study, exposing participants to goal-level analyses shifts their perceived contributions by nearly 2 points on a 5-point scale, revealing systematic miscalibration in how users understand their own AI-assisted work.
The risks posed by AI features are increasing as they are rapidly integrated into software applications. In response, regulations and standards for safe and secure AI have been proposed. In this paper, we present an agentic framework that constructs knowledge graphs (KGs) from AI policy documents and retrieves policy-relevant information to answer questions. We build KGs from three AI risk-related polices under two ontology schemas, and then evaluate five LLMs on 42 policy QA tasks spanning six reasoning types, from entity lookup to cross-policy inference, using both heuristic scoring and an LLM-as-judge. KG augmentation improves scores for all five models, and an open, LLM-discovered schema matches or exceeds the formal ontology.