Academic Intelligence · Curated Daily

探索全球前沿学术脉络

AcademicHub 汇聚顶级期刊与预印本平台的实时文献。定制您的专属科研雷达,利用大语言模型自动生成交叉领域文献分析简报。

01.
arXiv (quant-ph) 2026-06-19

On the significance of Wigner's Friend in contexts beyond quantum foundations

arXiv:2402.08727v3 Announce Type: replace Abstract: There has been a surge of recent interest in the Wigner's Friend paradox, sparking several novel thought experiments and no-go theorems. The main narrative has been that Wigner's Friend highlights a counterintuitive feature that is unique to quantum theory, and which is closely related to the quantum measurement problem. Here, we challenge this view. We argue that the gist of the Wigner's Friend paradox can be reproduced without assuming quantum physics, and that it underlies a much broader class of enigmas in the foundations of physics and philosophy. To show this, we first consider several recently proposed Extended Wigner's Friend scenarios, and demonstrate that some of their implications for the absoluteness of observations can be reproduced by classical thought experiments that involve the duplication of agents. Crucially, some of these classical scenarios are technologically much easier to implement than their quantum counterparts. Then, we argue that the essential structural ingredient of all these scenarios is a feature that we call "Restriction A": that a physical theory cannot give us a probabilistic description of the observations of all agents. Finally, we argue that this difficulty is at the core of other puzzles in the foundations of physics and philosophy, and demonstrate this explicitly for cosmology's Boltzmann brain problem. Our analysis suggests that Wigner's Friend should be studied in a larger context, addressing a frontier of human knowledge beyond quantum foundations: to obtain reliable predictions for experiments in which these predictions can be privately but not intersubjectively verified.

02.
arXiv (CS.CL) 2026-06-18

Human-AI Coevolution Dynamics: A Formal Theory of Social Intelligence Emergence Through Long-Term Interaction

Current conversational AI systems have made significant progress in language generation, personalization, and long-context interaction. However, most existing methods model social behavior through isolated components such as emotion modeling, memory retrieval, or persona conditioning, lacking a unified framework to explain the emergence of stable social relationships and social intelligence in long-term human-AI interaction.To address this, we propose the Human-AI Coevolution Dynamics Framework (HACD-H), a formal model of human-AI interaction as a self-organizing social cognitive system. HACD-H integrates emotional adaptation, relational organization, social memory, and personality consistency into a unified dynamical framework and introduces principles including multi-timescale social cognition, relational attractors, trust basins, developmental phase transitions, and social cognitive energy dynamics.We construct a conversational dataset with approximately 14,700 interaction turns and develop a theory-driven empirical evaluation framework. Results reveal a hierarchy of temporal persistence in social cognition, stable relational attractors, phase-transition-like developmental patterns, and a structured social cognitive energy landscape. Social intelligence shows a significant negative correlation with social cognitive energy (r = -0.391, p < 0.001), and interaction trajectories exhibit progressive energy reduction over time.These findings suggest that social intelligence emerges from long-term social cognitive coevolution rather than isolated conversational capabilities. HACD-H provides a unified theoretical foundation for modeling adaptive human-AI social interaction and developing socially intelligent AI systems.

03.
arXiv (CS.CV) 2026-06-15

FLaRA: Predicting Future Latent Representations for Accident Anticipation

Anticipating traffic accidents from dashcam videos is a critical challenge in intelligent transportation systems. Existing methods typically map visual context directly to a collision probability without explicitly modeling the future evolution of the driving scene. In this paper we propose FLaRA (Predicting Future Latent Representations for Accident Anticipation), a novel predictive architecture that shifts this paradigm by forecasting future latent representations for accident anticipation. Building upon the Video Joint-Embedding Predictive Architecture (V-JEPA2), our model conditions a predictor network on observed context frames to predict the forthcoming latent features of the scene. A classifier then operates on these predicted future representations rather than only on past observations. To ensure these forecasts remain grounded in realistic future dynamics, we introduce a joint training objective that simultaneously optimizes an auxiliary feature-level reconstruction loss and a cross-entropy classification loss. Extensive evaluations on the Nexar dataset, alongside cross-domain validations on the DAD, DADA-2000, and DoTA benchmarks, demonstrate that our approach achieves state-of-the-art performance while maintaining realistic early warning capabilities.

04.
arXiv (CS.CL) 2026-06-16

Measuring Whether LLM Tutors Teach or Solve: A Diagnostic for Educational Impact

Large language models are increasingly proposed as educational tutors, yet stronger task-solving ability does not necessarily imply stronger learning support. Motivated by recent calls to measure the social impact of NLP systems in practice, we study whether public LLM tutoring benchmarks distinguish learning-supportive behavior from mere answer production. We propose a lightweight diagnostic based on the gap between solving-oriented and pedagogy-oriented benchmark performance. Using public MathTutorBench leaderboard results, we show that these dimensions are only partially aligned: across eight publicly reported models, the correlation between solving and pedagogy composites is 0.421, and several models shift meaningfully in rank when evaluation moves from solving to pedagogy. We then analyze the public TutorBench sample and show that agency-relevant behaviors are explicitly encoded in benchmark rubrics, especially in active-learning settings that reward guiding questions, calibrated hints, and non-disclosive scaffolding. Together, these findings suggest that educational-impact evaluation should not treat task success as a sufficient proxy for learning support. We argue that public tutoring benchmarks can better support positive-impact evaluation by reporting solving-oriented and pedagogy-oriented scores separately and by making disclosure-sensitive, student-agency-preserving criteria more explicit.

05.
arXiv (CS.CL) 2026-06-18

CoreMem: Riemannian Retrieval and Fisher-Guided Distillation for Long-Term Memory in Dialogue Agents

Personalized dialogue agents require continuous long-term memory to maintain coherent interactions across multiple sessions. However, deploying these capabilities on consumer-grade hardware (e.g., 8 GB VRAM edge devices) introduces severe memory and compute bottlenecks. Existing systems typically rely on isotropic cosine similarity for retrieval and heuristic rules for context compression. These approaches lack a unified theoretical foundation, frequently suffering from the hubness problem in high-dimensional retrieval and syntactic fragmentation during compression. To overcome these limitations, we propose CoreMem, a resource-efficient edge-cloud memory architecture fundamentally unified by information geometry. First, Riemannian retrieval replaces cosine matching with a locally adaptive Fisher-Rao metric, effectively penalizing hub memories via Mahalanobis distance with O(Ndr) Woodbury acceleration for real-time search. Second, Fisher-guided discrete token distillation (FDTD) introduces a hierarchical sentence-to-token compression mechanism. It derives sensitivity scores from Fisher information traces, providing a principled compression-KL tradeoff augmented with explicit structural syntax protection. Evaluated on the LOCOMO and LongMemEval-S benchmarks, CoreMem achieves strong accuracy improvements, yielding substantial gains in Open-domain (+4.51 pp) and Temporal (+4.17 pp) reasoning. Extensive profiling confirms that CoreMem operates seamlessly within a strict 8 GB VRAM budget, successfully bridging the gap between resource-constrained edge devices and the demand for theoretically grounded, lifelong memory agents.

06.
bioRxiv (Bioinfo) 2026-06-20

The recount3 Python package for programmatic access to uniformly processed RNA-seq data

The recount3 online resource provides tens of thousands of uniformly processed RNA-seq samples across human and mouse from major sequencing repositories like the Sequence Read Archive. While access to these datasets has traditionally been centered in the R/Bioconductor ecosystem, the growing prominence of Python in bioinformatics and machine learning necessitates native, efficient tooling for Python users. Therefore, we present the recount3 Python package with robust application programming interface (API) and command-line interface (CLI) for discovering, downloading, and materializing recount3 resources. The software orchestrates uniform resource locator (URL) resolution, persistent on-disk caching, and the automatic parsing of data into analysis-ready data structures, including Pandas DataFrames and BiocPy RangedSummarizedExperiment objects. The recount3 Python package drastically lowers the barrier to entry for large-scale utilization of RNA-seq data in Python-based computational pipelines, bridging the gap between massive public transcriptomic data and modern machine learning ecosystems.

07.
arXiv (quant-ph) 2026-06-11

Implementing Hamiltonian Renormalization Group Flow on Quantum Computers with VAPOR

arXiv:2606.11306v1 Announce Type: cross Abstract: While Hamiltonian Lattice Gauge Theory is gaining traction, today's limited numerical capacity leaves simulations affected by discretization errors. This motivates the implementation of renormalization group (RG) techniques to find discretization-error-free operators. To this end, we introduce VAPOR, a variational quantum algorithm that decomposes operators into Pauli strings, identifies RG flow orbits, and determines fixed points of a naively discretized operator. We illustrate this using a toy model of a kinematic operator in a symmetry-restricted SU(2) Yang-Mills theory.

08.
arXiv (CS.CV) 2026-06-16

Focus When Necessary: Adaptive Routing and Collaborative Grounding for Training-Free Visual Grounding

While Multimodal Large Language Models (MLLMs) excel in cross-modal reasoning, they often struggle to perceive fine-grained details in complex high-resolution images. Recent training-free methods address this through image scaling and localized cropping. However, applying these manipulations indiscriminately introduces computational redundancy for simple queries and can degrade accuracy by truncating essential global context or introducing irrelevant background noise. To this end, we propose LazyMCoT, a dynamic and training-free framework that adaptively allocates visual grounding efforts based on sample difficulty. The framework features an Adaptive Routing mechanism that evaluates predictive uncertainty using first-token statistics from a single forward pass. This efficiently bypasses confident cases while ensuring the recall of difficult samples via conformal calibration. For these challenging cases, a Collaborative Grounding module integrates the inherent cross-modal attention of the model with an external visual expert through a two-stage refinement process. This refinement process generates a precise localized display to recover small or occluded targets. Extensive experiments across diverse benchmarks demonstrate that LazyMCoT rivals training-based approaches by simultaneously improving reasoning accuracy and reducing average inference latency. Our code is availble at https://github.com/TencentBAC/LazyMCoT.

09.
arXiv (CS.CV) 2026-06-15

QualiaNet: An Experience-Before-Inference Network

作者:

Human 3D vision involves two distinct stages: an Experience Module, where stereo depth is extracted relative to fixation, and an Inference Module, where this experience is interpreted to estimate 3D scene properties. Paradoxically, although stereo vision does not provide us with absolute distance information, it nonetheless affects our inferences about distance. We propose the Inference Module exploits a natural scene statistic: near scenes produce vivid disparity gradients, while far scenes appear comparatively flat. QualiaNet implements this two-stage architecture computationally: disparity maps simulating human stereo experience are passed to a CNN trained to estimate distance. The network can recover distance from disparity gradients alone, validating this approach.

10.
arXiv (CS.LG) 2026-06-17

A 3D Isovist World Model – Revealing a City's Unseen Geometry and Its Emergent Cross-City Signature

arXiv:2606.03609v3 Announce Type: replace-cross Abstract: Embodied agents that navigate cities rely on world models that predict how their surroundings will change as they move. But for navigation, what matters is not what the buildings look like; it is where the agent can go. Most world models nonetheless predict appearance, learning how a scene looks rather than the space an agent can move through. Those that do target geometry, such as bird's-eye-view occupancy grids, flatten the three-dimensional environment onto a ground plane, discarding the above-ground and multi-level structure that shapes real navigation. What is missing is a predictive target that captures the navigable geometry an agent actually traverses, without photometric entanglement and without collapsing the third dimension. Our key idea is to model the open volume between buildings, the negative space, encoded as a 3D isovist: a spherical visibility-depth map recording the distance to the nearest surface in every direction. We introduce an embodied world model that predicts the next isovist from a short history of past isovists and a movement action. The prediction is formulated as a depth residual so the decoder inherits sharp building edges, trained with self-rollout scheduled sampling to keep corrupted context on the geometry manifold, and equipped with a persistent latent bird's-eye-view spatial map for cross-path consistency. Our central finding is emergent and unexpected: a single city-blind model trained on Manhattan and Paris develops a cross-city spatial signature, with city identity linearly decodable from its temporal latents far above single-frame baselines, so the signature lives in the learned dynamics rather than in appearance. The representation is lightweight, interpretable, and reproducible, offering a geometric substrate for spatial reasoning in embodied AI, robotics, and urban analysis, released with an open dataset and pipeline.

11.
arXiv (CS.CL) 2026-06-16

G-Loss: Graph-Guided Fine-Tuning of Language Models

Traditional loss functions, including cross-entropy, contrastive, triplet, and su pervised contrastive losses, used for fine-tuning pre-trained language models such as BERT, operate only within local neighborhoods and fail to account for the global semantic structure. We present G-Loss, a graph-guided loss function that incorporates semi-supervised label propagation to use structural relationships within the embedding manifold. G-Loss builds a document-similarity graph that captures global semantic relationships, thereby guiding the model to learn more discriminative and robust embeddings. We evaluate G-Loss on five benchmark datasets covering key downstream classification tasks: MR (sentiment analysis), R8 and R52 (topic categorization), Ohsumed (medical document classification), and 20NG (news categorization). In the majority of experimental setups, G-Loss converges faster and produces semantically coherent embedding spaces, resulting in higher classification accuracy than models fine-tuned with traditional loss functions.

12.
arXiv (CS.LG) 2026-06-19

VIMPO: Value-Implicit Policy Optimization for LLMs

arXiv:2606.20008v1 Announce Type: new Abstract: Reinforcement learning with verifiable rewards has become a central tool for improving the reasoning ability of large language models, but current methods face a trade-off between simplicity and credit assignment. Group-relative methods such as GRPO avoid training a critic, but typically assign a trajectory-level advantage to every token. Actor-critic methods provide denser learning signals, but require a learned value function with its own training instability. We introduce VIMPO, a critic-free policy optimization method that derives a policy-implied value function from the optimality conditions of KL-regularized reinforcement learning. For autoregressive generation, the resulting value recurrence can be written in terms of policy-reference log-ratios and anchored by the terminal condition that no future reward remains at the end of a trajectory. This gives a simple value loss that incorporates outcome-level verifiable rewards without training a critic. The same derivation also yields a critic-free actor advantage, allowing VIMPO to separate reward incorporation through the value loss from policy improvement through a PPO-style actor update. On mathematical RLVR benchmarks, VIMPO improves over GRPO across MATH-500, AIME 2024, AIME 2025, and OlympiadBench, with especially larger gains on competition-style evaluations. Under noisy rewards, VIMPO retains a consistent advantage over GRPO, suggesting that policy-implied value optimization can provide finer credit assignment while preserving the practical simplicity of critic-free training.

13.
medRxiv (Medicine) 2026-06-17

Accounting for Human Movement to Improve Exposure-Health Models

Background. Current exposure-health models rely on averaged, residential-based environmental exposures, failing to account for human movement. This aggregation can lead to exposure misclassification and biased exposure-response estimates, potentially distorting our understanding of the true health effects of environmental conditions. We developed exposure disaggregation regression models that explicitly account for human movement when linking environmental exposures to health outcomes. Methods. By weighting pixel-level exposures according to distance from home as a simple proxy for human movement, our model linked disaggregated environmental exposures to individual-level health outcomes. Weights were either fixed a priori or derived from a latent distance-decay power parameter learned from the data. We additionally evaluated model performance under a nonlinear exposure-response relationship. Model performance was assessed across multiple sample sizes (N = 1,114; 50,000; and 100,000). A simulation study examined parameter recovery using bias, empirical standard error (EmpSE), and credible interval coverage. As a case study, Demographic and Health Surveys (DHS) data from Albania were used to link acute respiratory infection (ARI) outcomes among children under five to pixel-level NDVI within a 3 km buffer around DHS cluster centroids, and the proposed models were applied to these data. Results. Across all models (fixed-weight, learned-weight, and restricted cubic spline models), parameter recovery improved with increasing sample size. At N = 1,114, estimates were biased and imprecise, with incorrect effect direction for exposure-response parameters (e.g., learned-weight {beta}1 bias = - 0.79; EmpSE = 2.61; coverage = 0.88). In contrast, the models accurately recovered parameters at larger sample sizes, including the latent distance-decay parameter (bias = - 0.02; EmpSE = 0.15; coverage = 0.95 at N = 100,000), demonstrating their ability to reliably learn movement-based exposure weights when sufficient data were available. Conclusion. Instead of relying on arbitrarily-sized buffers, this statistical framework provides a novel method for studying environmental exposure-health relationships whilst accounting for human movement. With sufficiently large sample sizes, it can accurately estimate the influence of disaggregated environmental exposures on individual-level health and help address exposure misclassification arising from residential-only metrics. This methodological framework remains scalable, interpretable, and adaptable to other exposures and outcomes, offering a foundation for future work that integrates richer mobility-informed exposure-health research.

14.
medRxiv (Medicine) 2026-06-15

Routine use of oral iron for people with heart failure and iron deficiency in primary care; retrospective cohort study

Aims: Iron deficiency is common among people with heart failure and associated with morbidity and mortality. While intravenous iron improves clinical outcomes, oral iron continues to be prescribed in routine practice despite limited evidence of benefit. Methods: We completed a retrospective primary care cohort study (2016 to 2021) to investigate the proportion of people with an incident diagnosis of heart failure who had iron deficiency identified (defined as ferritin

15.
arXiv (CS.CL) 2026-06-15

C2-Faith: Benchmarking LLM Judges for Causal and Coverage Faithfulness in Chain-of-Thought Reasoning

Large language models (LLMs) are increasingly used as judges of chain-of-thought (CoT) reasoning, yet it remains unclear whether they can reliably assess process faithfulness rather than merely answer plausibility. We introduce C2-Faith, a benchmark built from PRM800K that explicitly decomposes faithfulness into two complementary dimensions: causality (whether each step logically follows from prior context) and coverage (whether essential intermediate inferences are present). Using controlled perturbations, we construct examples with known causal error positions by replacing a single step with a logically inconsistent variant, and with controlled coverage deletions at varying rates, enabling direct measurement against reference labels. We evaluate three frontier LLM judges across three tasks: binary causal detection, causal step localization, and coverage scoring. Our results reveal that judge reliability is highly task-dependent, with no single model dominating across settings. While models often detect that an error exists, they struggle to accurately localize it, indicating a substantial gap between detection and attribution. Moreover, all judges systematically overestimate reasoning completeness, assigning high coverage scores even when substantial portions of intermediate reasoning are missing. These findings expose fundamental limitations of LLM judges in process-level evaluation and highlight the need for more reliable and calibrated methods when using LLMs to assess reasoning quality.

16.
arXiv (CS.CV) 2026-06-16

Scribby: A Multi-Level LLM Framework for Semantic Video Analysis

As video content continues to expand across educational platforms, recorded lectures, and live-streamed entertainment, the need for efficient and structured analysis of long-form footage has increased [1]. Although many existing AI programs provide high-level video summaries based on AI-generated transcripts [2,3,4,5], these approaches are often limited to coarse overviews and lack detailed analysis of a video's structure, thematic progression, and semantic relationships, all of which are required for comprehensive video analysis. This paper proposes an LLM-based video summarization framework that balances macro-level comprehension with micro-level semantic analysis [6,12,13]. The first stage of the process indexes the video at a micro level by (1) analyzing the full transcript, (2) analyzing individual transcript sentences, and (3) grouping these sentences by semantic similarity using an LLM as a judge [6,13]. Contextual continuity is retained during sentence-level processing by incorporating both the global transcript analysis and adjacent sentence information into each evaluation prompt. This framework establishes a foundation for video analysis tools that visualize semantic chunking and semantic matching through relevance-based heatmaps. Limitations and future expansions of the framework are also discussed.

17.
arXiv (CS.LG) 2026-06-18

Estimating carbon pools in the European Shelf sea environment: replacing reanalysis by model-informed machine learning?

arXiv:2508.10178v3 Announce Type: replace-cross Abstract: Shelf seas are important for the economy and the carbon cycle, but shelf sea observations for carbon pools are often sparse, or highly uncertain. An alternative can be provided by carbon reanalyses (whether assimilating proxy variables, such as chlorophyll-$a$, or directly carbon), but these are often expensive to run. We propose to use a computationally cheap ensemble of neural networks (i.e. deep ensemble) to learn the relationship between the directly observable (atmospheric, riverine and ocean) variables and marine carbon pools from a coupled physics-biogeochemistry model. The deep ensemble was trained on a North-West European Shelf (NWES) physical-biogeochemistry model free run simulation. After training, the deep ensemble was run using inputs from the NWES reanalysis instead of the free run, demonstrating that it can efficiently predict several NWES carbon pools (e.g., detritus, zooplankton, heterotrophic bacteria) in much better agreement with the reanalysis than the free run, while also providing uncertainty information. We further show that the deep ensemble performs similarly well when it is driven directly by the observations assimilated into the reanalysis, with the limitation that carbon pools can then be predicted only at the observed locations and times. We focus on explainability of the results and demonstrate potential use of the deep ensembles for future climate what-if scenarios. We suggest that model-informed machine learning presents a viable alternative to expensive reanalyses and could complement observations, wherever they are missing and/or highly uncertain.

18.
bioRxiv (Bioinfo) 2026-06-13

MoE-Bind: Guiding De Novo Protein Binder Generation with Sparse Experts

作者:

De novo protein binder design has been dominated by structure-based pipelines that require known three-dimensional target conformations and consume substantial compute and generation time per design, limiting their throughput and accessibility for routine large-scale binder exploration. Sequence-only generative models promise a faster and lighter alternative, yet existing systems remain uniformly dense and frequently reintroduce structural computation at inference, undermining the core advantages they were intended to deliver. Across the broader language modelling community, transformers have meanwhile transitioned from fully dense designs to sparse Mixture-of-Experts architectures that decouple capacity from per-token compute, a shift that has yet to reach sequence-only protein binder generation. We present MoE-Bind, an autoregressive protein binder generator that, for the first time in this domain, combines Multi-head Latent Attention with a sparse Mixture-of-Experts feed-forward network and is evaluated under two independent structure predictors, Boltz-2 and AlphaFold2-Multimer. Despite activating less than half the per-token parameters of compute-matched dense baselines, MoE-Bind matches or exceeds them on full-length receptor-conditioned binder generation on a leakage-free Docking Benchmark 5.0 evaluation, transfers without peptide-specific training to short-peptide design, and reduces training and inference compute by a large margin. Routing analysis on generated binders reveals interpretable expert specialization at both the individual amino acid and biochemical group level, a structured expert-token alignment not previously reported for natural-language MoE models. These results show that sparse architectural design, rather than scale, can deliver fast, structure-free, and interpretable protein binder generation.

19.
arXiv (CS.AI) 2026-06-11

Grounding Computer Use Agents on Human Demonstrations

arXiv:2511.07332v2 Announce Type: replace-cross Abstract: Building reliable computer-use agents requires grounding: accurately connecting natural language instructions to the correct on-screen elements. While large datasets exist for web and mobile interactions, high-quality resources for desktop environments are limited. To address this gap, we introduce GroundCUA, a large-scale desktop grounding dataset built from expert human demonstrations. It covers 87 applications across 12 categories and includes 56K screenshots, with every on-screen element carefully annotated for a total of over 3.56M human-verified annotations. From these demonstrations, we generate diverse instructions that capture a wide range of real-world tasks, providing high-quality data for model training. Using GroundCUA, we develop the GroundNext family of models that map instructions to their target UI elements. At both 3B and 7B scales, GroundNext achieves state-of-the-art results across five benchmarks using supervised fine-tuning, while requiring less than one-tenth the training data of prior work. Reinforcement learning post-training further improves performance, and when evaluated in an agentic setting on the OSWorld benchmark using o3 as planner, GroundNext attains comparable or superior results to models trained with substantially more data,. These results demonstrate the critical role of high-quality, expert-driven datasets in advancing general-purpose computer-use agents.

20.
arXiv (CS.CV) 2026-06-16

Landmark-free Assessment of Lower-limb Alignment with Implicit Neural Shape Functions from Knee Radiographs

Radiographic assessment of lower-limb alignment (LLA) is important for predicting joint health and surgical outcomes in total knee arthroplasty. Traditional measurement methods are manual and time-consuming, while recent machine learning approaches typically rely on locating a fixed set of anatomical landmarks. This dependence limits flexibility and may require re-annotation when clinical definitions change. To address this, we propose an automated workflow using Implicit Neural Shape Functions (INSF). Rather than relying on explicit landmark coordinates, we encode the anatomy into a compact latent space and regress clinical alignment measurements directly from these latent codes. This architecture allows for rapid extendability to new tasks without altering the backbone representation. We trained our method on an internal dataset of 566 knee radiographs, each annotated with the outline of the femur and tibia. We evaluated it on both an internal test dataset of 50 patients and a separate external set of 402 preoperative cases from the MRKR dataset. Manual clinical measurements are available for these data, and the MRKR measurements will be made publicly accessible. Performance was comparable to state-of-the-art landmark-based methods and manual agreement, while offering a flexible shape representation that can be extended to additional measurement tasks.

22.
arXiv (CS.AI) 2026-06-16

Large Language Models as Optimizers: A Survey of Direct vs. Tool-Augmented Approaches and Their Performance Frontiers

arXiv:2606.15577v1 Announce Type: new Abstract: Large Language Models (LLMs) are increasingly involved in complex mathematical optimization, even if the pragmatic user who triggers them is unaware of it. After all, many real-world problems reduce to the search for better or the best solutions. The field of LLM-as-optimizer has three paradigms: direct optimization, tool-augmented optimization, and tool-creating optimization. Direct optimization uses iterative prompting and heuristic generation to navigate solution spaces. Tool-augmented optimization translates natural language problems into formal specifications and orchestrates external solvers. Tool-creating optimization goes further, using LLMs to discover reusable algorithms or heuristics that can be deployed at zero marginal LLM cost. We describe current performance frontiers based on the benchmarks from the literature. We identify the critical reasoning gap in current architectures and argue for trade-offs between the future potential of direct optimization and the auditability of tool-augmented optimization. Even future, more powerful models might opt for tool-making to improve operational efficiency for repetitive families of problems.

23.
arXiv (CS.AI) 2026-06-16

Edit Knowledge, Not Just Facts via Multi-Step Reasoning over Background Stories

arXiv:2602.02028v2 Announce Type: replace Abstract: Enabling artificial intelligence systems, particularly large language models, to update knowledge and flexibly apply it during reasoning remains a central challenge. Existing knowledge editing approaches emphasize atomic facts, improving factual recall but often failing to integrate updated information into a coherent framework usable across contexts. In this work, we argue that knowledge update is fundamentally a reasoning problem rather than a memorization problem. Consequently, a model should be trained in situations where the new information is instrumental to solving a task, combined with pre-existing knowledge, and exercised through multi-step reasoning. Based on this insight, we propose a training strategy based on three principles. First, new knowledge is introduced as a coherent background story that contextualizes novel facts and explains their relation to existing knowledge. Second, models are trained using self-generated multi-hop questions that require multi-step reasoning involving the new information. Third, training is done using knowledge distillation, forcing a student model to internalize the teacher's reasoning behavior without access to the novel information. Experiments show that models trained with this strategy effectively leverage newly acquired knowledge during reasoning and achieve remarkable performance on challenging questions that require combining multiple new facts.

24.
arXiv (CS.CL) 2026-06-11

Breaking Entropy Bounds: Accelerating RL Training via MTP with Rejection Sampling

Reinforcement learning (RL) has become a key component in modern large language models, yet the rollout stage remains the key bottleneck in RL training pipelines. Although Multi-Token Prediction (MTP) offers a natural solution to accelerate rollouts through speculative decoding, many studies have observed that MTP acceptance rates degrade significantly during RL training, leading to limited speedup performance. To address this bottleneck, we present Bebop, a systematic study of MTP in LLM post-training, and offer practical recipes to integrate MTP into large-scale RL pipelines. First, we reveal that the MTP acceptance rate is fundamentally bounded by the fluctuation of model entropy, which demonstrates a clear negative linear relationship with the rise of entropy in the RL stage. Second, we show that probabilistic rejection sampling largely alleviates the disturbance introduced by entropy in RL compared to greedy draft sampling. We further identify that the conventional MTP training objectives (cross-entropy or KL) are suboptimal in such settings, and therefore we propose a novel end-to-end TV loss that directly optimizes multi-step rejection sampling acceptance rate, yielding ~10% acceptance rate improvements, achieving up to 95% acceptance rates and up to 25% extra inference throughput gains across mathematical reasoning, code generation, and agentic tasks. Third, we test various online MTP training strategies during RL and show that pre-RL MTP training with e2e TV loss and rejection sampling achieves a consistent acceptance rate and speedup throughout the entire RL, eliminating the need for costly online MTP updating. We provide extensive experiments and analysis that validate our findings. Experimental results show our method achieves up to 1.8x end-to-end acceleration in async RL training of Qwen3.5, Qwen3.6, and Qwen3.7 models.

25.
arXiv (CS.CL) 2026-06-16

SpatialWorld: Benchmarking Interactive Spatial Reasoning of Multimodal Agents in Real-World Tasks

Spatial reasoning is a foundational capability for multimodal large language models (MLLMs) to perceive and operate within the physical world. However, existing benchmarks predominantly rely on passive evaluation (e.g., static VQA) or simulator-specific pipelines, failing to assess general interactive spatial understanding. We introduce SpatialWorld, a unified benchmark designed specifically for evaluating the interactive spatial understanding of multimodal agents in complex real-world tasks. Integrating eight heterogeneous simulation backends under a shared, simulator-agnostic protocol, SpatialWorld features 760 human-annotated tasks across diverse domains (e.g., household routines, travel, social collaboration). Agents must solve tasks under vision-only partial observability, actively gathering egocentric visual evidence and expressing decisions via a unified, text-based action interface native to MLLMs. For reliable evaluation, each task includes a human-validated initial state, a reference trajectory, and a terminal-state verifier. Evaluating 15 advanced agents reveals that robust spatial task solving remains challenging: the strongest model, GPT-5, achieves an average task success rate (TSR) of only 17.4%, while the leading open-source model, Qwen-3.5, reaches 14.1%. Further analysis exposes a clear mismatch between task success and execution efficiency, alongside substantial domain-specific performance variations. These bottlenecks in active exploration and long-horizon planning position SpatialWorld as a rigorous testbed for future spatial agents.