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.
Nature (Science) 2026-06-09

A unicellular relative links aggregative multicellularity to animal origins

作者:

How animals evolved complex multicellularity from their unicellular ancestors remains unanswered. Unicellular relatives of animals exhibit simple multicellularity through clonal division, formation of multinucleate coenocytes, or aggregation. 1 Therefore, animal multicellularity may have evolved from one (or a combination) of these behaviours. Aggregation has classically been dismissed as a means to complex multicellularity. 2 However, aggregation occurs in many extant animal cells and has also been recently described in three close unicellular relatives of animals (the choanoflagellates Salpingoeca rosetta and Choanoeca flexa, and the filasterean Capsaspora owczarzaki). 3-5 It is unclear whether aggregation in these species is derived or ancestral, and its relevance for animal origins remains unknown. To fill this gap, we investigated whether an additional close unicellular relative of animals can undergo aggregation. We discovered that the marine free-living bacterivorous filasterean Ministeria vibrans 6 forms homogeneous aggregates with reproducible kinetics that have long-term stability, and that improved feeding and mating may be evolutionary drivers of this aggregation. Notably, we found that homologs of many animal multicellularity genes involved in cell adhesion, signalling, and transcriptional regulation were deployed during the aggregation process, indicating that they may have been used for aggregation in the unicellular ancestors of animals before being co-opted into animal multicellular development. Thus, our results imply that aggregative multicellularity was key to the development of the multicellular animal genetic toolkit.

02.
arXiv (CS.AI) 2026-06-19

Neural Additive and Basis Models with Feature Selection and Interactions

arXiv:2606.19850v1 Announce Type: cross Abstract: Deep neural networks (DNNs) exhibit attractive performance in various fields but often suffer from low interpretability. The neural additive model (NAM) and its variant called the neural basis model (NBM) use neural networks (NNs) as nonlinear shape functions in generalized additive models (GAMs). Both models are highly interpretable and exhibit good performance and flexibility for NN training. NAM and NBM can provide and visualize the contribution of each feature to the prediction owing to GAM-based architectures. However, when using two-input NNs to consider feature interactions or when applying them to high-dimensional datasets, training NAM and NBM becomes intractable due to the increase in the computational resources required. This paper proposes incorporating the feature selection mechanism into NAM and NBM to resolve computational bottlenecks. We introduce the feature selection layer in both models and update the selection weights during training. Our method is simple and can reduce computational costs and model sizes compared to vanilla NAM and NBM. In addition, it enables us to use two-input NNs even in high-dimensional datasets and capture feature interactions. We demonstrate that the proposed models are computationally efficient compared to vanilla NAM and NBM, and they exhibit better or comparable performance with state-of-the-art GAMs.

03.
arXiv (CS.AI) 2026-06-19

Emergent Alignment

arXiv:2606.19527v1 Announce Type: new Abstract: Can Large Language Models (LLMs) discern when their own outputs are misaligned with human ethics? And can they self-correct? We endow an LLM with a conscience step that reviews its own reasoning and outputs, and we extend the training loss with an alignment component using Direct Preference Optimization (DPO) to steer the model away from non-ethical outputs. The result is an online technique to align models in a wide range of applications: training, fine-tuning, adversarial prompting, and zero-shot learning. It does not require a weaker or stronger judge, relying instead on a frozen copy of itself. In previous work, the Emergent Misalignment scenario showed a range of emergent unethical behaviors from fine-tuning the model to hack code. Instead, we empirically show how to achieve Emergent Alignment: a single high-level introspective question steers training toward an ethical model under the same code hacking scenario.

04.
arXiv (CS.AI) 2026-06-18

User as Engram: Internalizing Per-User Memory as Local Parametric Edits

作者:

arXiv:2606.19172v1 Announce Type: new Abstract: Personal memory in a language model is two problems: content and reasoning skill. The brain keeps the two apart (a sparse, local engram in the hippocampus for each episode, a slow neocortex for the shared skills that interpret it), so a new fact need not overwrite everything else. Most personalization today keeps a user's facts outside the weights, in a natural-language memory file or a retrieval index. When facts are written into the model instead, the standard recipe is the per-user LoRA adapter, which does the opposite of the brain, folding content and skill into one global weight delta. Writing a user's facts as a LoRA contaminates text unrelated to them; writing the same facts as local Engram rows leaves it mathematically untouched, resulting in a roughly 33,000x smaller memory footprint. We therefore propose User as Engram: store a user's content as surgical edits to the hash-keyed memory table of an Engram model, and carry the reasoning skill in one shared adapter. This layered design matches per-user LoRA's direct recall while delivering 5.6x higher indirect-reasoning accuracy on average, and never makes a single user worse at reasoning than the untouched base. The edit is a glass box: writing a fact switches on its lookup at exactly the trigger, adds the value the answer needs, leaves every other position unchanged to the last bit, and fails if written into the wrong layer. Because different users' facts land in disjoint hash slots, their edits compose: many users live in one shared table at once, stacking additively and losslessly, where a per-user LoRA, a single global weight delta, admits only one. Upon retrieval, a per-user Engram table does not grow with the population the retriever must search, so past ~100 facts it overtakes a retrieval pipeline on a 2.5x larger model.

05.
arXiv (CS.LG) 2026-06-15

Equivariant Representation Learning via Class-Pose Decomposition

arXiv:2207.03116v4 Announce Type: replace Abstract: We introduce a general method for learning representations that are equivariant to symmetries of data. Our central idea is to decompose the latent space into an invariant factor and the symmetry group itself. The components semantically correspond to intrinsic data classes and poses respectively. The learner is trained on a loss encouraging equivariance based on supervision from relative symmetry information. The approach is motivated by theoretical results from group theory and guarantees representations that are lossless, interpretable and disentangled. We provide an empirical investigation via experiments involving datasets with a variety of symmetries. Results show that our representations capture the geometry of data and outperform other equivariant representation learning frameworks.

06.
arXiv (CS.CL) 2026-06-19

How Linear Is a Transformer Feed-Forward Block? Per-Block Linear Recoverability Is Learned, Not Architectural

作者:

Transformer feed-forward networks (FFNs) are often treated as nonlinear stores of computation, yet how nonlinear a trained FFN block actually is has rarely been measured. We treat each FFN as a position-wise input-to-output map and split it into the exact least-squares linear approximation plus a residual. The held-out variance the closed-form linear map explains defines a block's linear recoverability (R^2_lin), an optimiser-free measure of its linearity. Across all twelve blocks of GPT-2, Pythia-160m, and llama-160m, R^2_lin is highly heterogeneous and non-monotone with depth, ranging from near-linear (>0.99) to strongly nonlinear (

07.
arXiv (CS.CL) 2026-06-15

TA-RAG: Tone-Aware Retrieval-Augmented Generation for Peer-Support Health Communication

Retrieval-augmented generation (RAG) successfully grounds large language model (LLM) outputs in trusted documents, but factual grounding alone is insufficient for sensitive peer-support health communication. In domains such as HIV peer support, responses must also be accessible, stigma-free, empathetic, and tailored to the recipient. This paper presents TA-RAG, a lightweight, prompt-based tone-aware RAG framework that embeds explicit tone control into a RAG pipeline without requiring model fine-tuning. We operationalise tone across four core components: stigma-free rewriting, readability adjustment, recipient adaptation, and empathy rephrasing. We evaluate TA-RAG through component-level tests using questions derived from HIV Online Learning Australia (HOLA), UNAIDS terminology guidance, readability metrics, peer-support standards from National Association of People with HIV Australia (NAPWHA), and a public empathy dataset. Results show that the TA-RAG's components improve their targeted communication quality while preserving key content. These findings emphasise that prompt-based tone control is a potential direction for making RAG outputs suitable for sensitive peer-support health communication.

08.
medRxiv (Medicine) 2026-06-15

The clinical utility of functional testing in fibroblasts to diagnose primary mitochondrial disease

Genome sequencing of the heterogeneous primary mitochondrial disorders (PMD) frequently reveals variants of uncertain significance that require functional tests for diagnosis, and does not identify variants in all patients. We analyzed mitochondrial enzyme assays, blue native polyacrylamide gel electrophoresis (BN-PAGE) with in-gel activity staining, complex I assembly blot, and select protein abundances in fibroblasts of a case series of 204 PMD patients divided into functional classes, in comparison to 51 controls and 53 differential diagnostic conditions. Overall, sensitivity and specificity for respiratory chain enzyme assays were 46% and 93% respectively, for BN-PAGE 40% and 98%, for complex I assembly assay 49% and 99%. The overall sensitivity of all tests was 76%, specificity 93%, with positive predictive value 96% and negative predictive value 67%. Categories with high sensitivity were isolated complex deficiencies, nuclear DNA-encoded mitochondrial protein synthesis defects, co-factor defects, and mitochondrial amino-acyl-tRNA synthetase conditions when aided by protein abundance. Mitochondrial DNA mutations and maintenance disorders showed poor sensitivities. Secondary dysfunctions were rare. A complete battery of functional tests showed strong diagnostic clinical utility in fibroblasts.

09.
arXiv (CS.AI) 2026-06-16

ToolMenuBench: Benchmarking Tool-Menu Filtering Strategies for Reliable and Efficient LLM Agents

arXiv:2606.15508v1 Announce Type: new Abstract: Tool-augmented large language model agents increasingly operate over large tool libraries, but existing evaluations often focus on whether a model can call a tool correctly rather than how the visible tool menu shapes reliability, efficiency, and safety-relevant risk exposure. We introduce ToolMenuBench, a benchmark for evaluating tool-menu construction in multi-step LLM agents. ToolMenuBench varies tool-menu size, distractor type, state-dependent task structure, and risk exposure, and reports both filter-level and downstream agent metrics, including visible-tool count, risky-tool exposure, task success, wrong-tool calls, premature actions, and token usage. In a controlled evaluation across seven model backends, three tool-menu sizes, six filtering methods, and seven evaluation settings, CMTF improves task success from 32.1% under all-tools exposure to 85.7%, while reducing average token usage by roughly 98%. Causal minimal tool filtering achieves the strongest overall tradeoff, reducing visible tools, wrong-tool calls, premature actions, and risky-tool exposure relative to unfiltered exposure, lexical filtering, state-aware filtering, and broader causal-path baselines. ToolMenuBench provides a reusable evaluation framework for studying the agent-interface problem: which tools should be visible, when they should be visible, and under what cost or risk constraints.

10.
arXiv (CS.AI) 2026-06-16

From Detection to Recovery: Operational Analysis on LLM Pre-training with 504 GPUs

arXiv:2605.09370v5 Announce Type: replace-cross Abstract: Large-scale AI training is fundamentally a distributed systems problem, where hardware failures are routine operating conditions rather than rare exceptions, yet public operational evidence from production training clusters remains limited. This report presents an empirical analysis of a 63-node NVIDIA B200 production cluster (504 GPUs), using 55 days of Prometheus time-series data and 73 days of operational logs covering 224 multi-node training sessions. The environment is cross-organizational: five parties (SKT, Upstage, Lablup, NVIDIA Korea, VAST Data) share a unified monitoring pipeline. This enabled joint diagnosis of a 60-node-scale storage I/O bottleneck absent in 2-4-node tests, a production-scale phenomenon no single team could isolate alone. We perform three quantitative analyses yielding four findings. First, over 751 Prometheus metrics and 10 XID-identified GPU failures, no single metric is consistently dominant across failure types, motivating multi-signal detection. Second, 523 checkpoint events trace the save/load path from GPU VRAM to the NFS server: restart loading reaches 21.5% of maximum read bandwidth (700 GB/s) and save bursts 16.0% of maximum write bandwidth (250 GB/s), with NFS/RPC queueing and transport-layer backlog rising together. Third, across 224 sessions over 73 days, node exclusions concentrate so the top 3 of 63 nodes account for over 50%. Fourth, auto-retry chain analysis shows a 33.3% success rate over 12 chains (73 attempts), 2.7x the 12.5% manual rate, with a median retry interval of 11 minutes (IQR 10-11). All analyses are grounded in production infrastructure providing session-level workload management, GPU-centric scheduling, and unified observability.

11.
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.

12.
arXiv (quant-ph) 2026-06-12

From 2D Yang-Mills to Calogero-Sutherland via a colored particle

arXiv:2606.13388v1 Announce Type: cross Abstract: We study Yang-Mills theory coupled to a particle on a cylinder, where gauge invariance and compactness reduce the dynamics to a finite dimensional quantum system. In the Abelian case, this yields a model equivalent to the Landau problem on a torus, with a degenerate ground state structure. We generalize this construction to non-Abelian gauge groups and show that, for SU(N), the system reduces to a one dimensional quantum many body problem with a singular Calogero-Sutherland-type interaction.

13.
arXiv (CS.AI) 2026-06-12

Boosting Direct Preference Optimization with Penalization

作者:

arXiv:2606.12505v1 Announce Type: cross Abstract: Offline preference optimization has become a practical substitute for reinforcement learning from human feedback, but pairwise objectives such as Direct Preference Optimization (DPO) and its variants use only the chosen and rejected responses stored in a static dataset. This leaves a useful signal unused: the response that the reference model itself would generate for the same prompt. We propose Direct Preference Optimization with Penalization (DPOP), a simple extension of DPO that augments the base preference loss with a gated penalty on reference-greedy responses. DPOP activates this penalty only when the current policy still assigns a lower likelihood to the preferred response than to the rejected response. On AlpacaEval 2.0, DPOP improves length-controlled win rate over DPO, SimPO, and AlphaDPO on both Llama-3-8b-it and Gemma-2-9b-it, achieving relative gains of 5.3\% and 4.4\% over baselines on the two models, respectively. Ablations further show that a SimNPO-style length-normalized penalty is stronger than NPO and token-level unlikelihood in this setting.

14.
arXiv (math.PR) 2026-06-12

Mixing times of one-sided $k$-transposition shuffles

arXiv:2112.05085v2 Announce Type: replace Abstract: We study mixing times of the one-sided $k$-transposition shuffle. We prove that this shuffle mixes relatively slowly, even for $k$ big. Using the recent ``lifting eigenvectors'' technique of Dieker and Saliola and applying the $\ell^2$ bound, we prove different mixing behaviors and explore the occurrence of cutoff depending on $k$.

15.
medRxiv (Medicine) 2026-06-16

Higher Population Coverage with Typhoid Conjugate Vaccine is Needed to Induce Herd Protection: Evidence from a Cluster-Randomized Trial in Urban Bangladesh

Introduction: A cluster randomized trial (CRT) in Bangladesh found that Vi-tetanus toxoid (Vi-TT) vaccine conferred 85% protection to vaccinees at 18 months of follow-up; however, it failed to confer significant herd protection to non-vaccinees. Methods: In the CRT, children aged 9 months to

16.
arXiv (quant-ph) 2026-06-15

Extending Covariant Fluctuation Theorems into Quantum Regime through Quasiprobability Approach

arXiv:2606.14519v1 Announce Type: cross Abstract: The covariant formulation of stochastic thermodynamics requires treating the stochastic work as a 4-vector, posing significant challenges for quantum systems due to the non-commutativity. We introduce a new quasiprobability distribution for the work 4-vector, which combines the Wigner and Margenau-Hill quasiprobabilities. This extends the covariant fluctuation theorems from classical to quantum regime. We illustrate our findings with a scalar field driven by classical particles with a generalized version of trace formula. Our work establishes a quasiprobability approach to studying relativistic quantum thermodynamics in a covariant way.

17.
arXiv (CS.CV) 2026-06-16

Faithful Action-unit Causal Reasoning for Counterfactually Faithful Emotion Explanations

Multimodal models can name the action units (AUs) behind a facial emotion, but their AU->emotion rationales are typically plausible rather than faithful: nothing forces the AUs a model invokes to be the AUs that actually drive its prediction. We cast AU->emotion reasoning as a counterfactual-consistency problem between the rationale, the label, and a structural AU->emotion causal graph G, and propose FACR, which grounds the reasoner in an independently induced, polarity-aware G and trains a counterfactual-faithfulness objective: a do-intervention on an AU that G marks causal for a class must move the prediction, while one it marks irrelevant must leave it unchanged. Faithfulness is thereby both trainable and measurable through a matching interventional metric, which we evaluate against a known causal structure, the PSPI pain-AU composition, as no existing affective-reasoning benchmark allows. We are explicit that this metric tests fidelity to the supplied structure rather than its rediscovery: it asks whether the trained reasoner invokes the AUs the structure marks causal, on held-out subjects and a second dataset. Under subject-independent evaluation on UNBC-PAIN, the objective raises the agreement between the invoked AUs and the PSPI composition from a no-objective baseline of 0.08 to 0.57, at a small detection cost; an unfaithfulness control attributes the gain to the objective. On a cross-dataset emotion transfer, the objective likewise raises fidelity to G on a seven-class task (0.50 to 0.84). Finally, we attach a language verbalizer and extend the audit to the generated text: biasing each action unit's emission by its latent activation makes the rationale faithful by construction, so that ablating an AU removes it from the explanation, a property that transfers to a second language-model backbone, whereas a freely generated rationale is unfaithful.

18.
arXiv (CS.CV) 2026-06-15

Pix2Fact: When Vision Is Not Enough – Benchmarking Fine-Grained VQA with Web Verification on High-Resolution Real-World Scenes

Despite progress on general tasks, vision-language models (VLMs) still struggle with challenges that demand both fine-grained visual grounding and external knowledge, a synergy overlooked by existing benchmarks that evaluate these abilities in isolation. To fill this void, we introduce Pix2Fact, a visual question-answering benchmark designed to assess expert-level visual perception and knowledge search. Pix2Fact comprises 1,000 high-resolution (4K+) images spanning eight scenarios. Its questions and answers are meticulously crafted by PhD-holding annotators from top global universities across diverse disciplines. Each question requires detailed visual grounding and the integration of external knowledge. Evaluating ten state-of-the-art VLMs, including proprietary models such as Gemini-3.1-Pro and GPT-5.4, we find that Pix2Fact poses a formidable challenge: the most advanced model (Gemini-3.1-Pro) achieves only 51.7% average accuracy, even with access to visual ground truth and search tools. Our analysis attributes this low accuracy to three factors, frequent visual grounding errors even with visual ground truth, shallow search harnessing, and VLM's inability to retrieve long-tail, unstructured local information. This striking gap exposes the limitations of current models in assisting humans with real-world scenarios that demand overwhelming visual comprehension. We believe Pix2Fact will serve as a critical benchmark to drive the next generation of language-vision agents that seamlessly integrate fine-grained perception with robust knowledge search.

19.
arXiv (CS.LG) 2026-06-17

Non-negative Matrix Factorisation with Topological Regularisation

arXiv:2606.17531v1 Announce Type: new Abstract: We investigate the learning of interpretable bases in non-negative matrix factorisation (NMF) by regularising the topology of the learned basis functions. Our approach is motivated by the observation that many data modalities can be viewed as non-negative functions on a structured domain, where the quality of a basis is intrinsically linked to its topology. However, naive methods for incorporating the topology of the support are often hindered by discreteness and threshold dependence, rendering them unsuitable for continuous optimisation. We address these challenges by employing persistent homology as a stable, threshold-free topological quantifier and by designing topological scores that integrate into the NMF objective as regularisers. The resulting framework encompasses spatially coherent image components, periodic time-series structures, and clique-like graph signals within a unified modelling language.

20.
bioRxiv (Bioinfo) 2026-06-18

Deciphering shared and divergent tissue architectures from cross-species spatial transcriptomics

作者:

The integration of spatial transcriptomics (ST) data across species is essential for cross-species and translational studies, but remains challenging due to molecular divergence and anatomical differences between organisms. We present STACAME, a graph attention autoencoder-based framework to decipher shared and divergent tissue architectures from cross-species ST data by explicitly modeling both orthologous and species-specific genes. STACAME aligns ST slices in a spatially aware manner, identifies homologous and species-specific domains, and enables a suite of downstream comparative analyses. We demonstrate its utility by integrating ST datasets from diverse tissues, including hippocampus, isocortex, embryo, breast, liver, and cerebellum, across multiple species such as human, macaque, marmoset, mouse, and zebrafish. STACAME supports cross-species spatial domain alignment, the detection of shared and divergent spatially variable genes, development alignment and comparison, and the 3D integration of tissue architecture. This flexible approach facilitates the translation of findings from model organisms to humans, providing a unified computational platform for cross-species spatial transcriptomics.

21.
arXiv (CS.AI) 2026-06-16

HAMON: Passive Optical Sequence Mixing for Long-Horizon Forecasting

arXiv:2606.17028v1 Announce Type: cross Abstract: Simple linear and frequency-domain models remain surprisingly competitive in long-horizon time-series forecasting, and recent mechanistic evidence suggests that standard forecasting benchmarks may not require the dense superposed representations that make transformers powerful in other domains. This raises a substrate-level question: if the core forecasting operator is often low-complexity and approximately linear, does it need to be implemented as learned digital temporal mixing? We introduce HAMON, a passive diffractive optical forecasting core in which historical values are encoded onto an optical aperture, future positions are left dark, and cascaded trainable phase masks with free-space diffraction shape the forecast directly in the output field. At inference, prediction is performed by a single passive optical propagation pass with no trainable digital sequence-mixing layer. Across standard benchmarks, HAMON outperforms the strongest digital baselines considered on ETTm2 at all horizons and on ETTh2 at all but the longest horizon, improving MSE by up to 14\% and doing so consistently across horizons rather than at isolated points. It is competitive on Weather and trails the strongest baselines on the remaining ETT settings and on the high-channel-count Traffic and Electricity datasets. Phase encoding, intensity-compatible readout, and phase-scrambling ablations, together with a TorchOptics cross-simulator check, indicate that the forecasts arise from the data-bearing optical field rather than from a digital forecasting head. Because the passive core uses standard Fourier optics, HAMON defines a concrete target for optical hardware and for passive physical sequence mixing.

22.
arXiv (CS.LG) 2026-06-18

Context-Aware Optimization of Follow-Up Intervals for Type 2 Diabetes Care Using Markov Decision Processes

arXiv:2606.19092v1 Announce Type: cross Abstract: Chronic disease management relies on regular patient-provider interactions to follow-up on disease progression and control. For Type 2 Diabetes (T2D), current guidelines prescribe fixed time intervals between subsequent primary care visits for all patients, overlooking heterogeneity in clinical trajectories and patient characteristics. This study introduces a Contextual Markov Decision Process (CMDP) model to optimize subpopulation-specific follow-up interval decisions using Electronic Health Record (EHR) data from 22,154 T2D patients across 10 primary care clinics. Contexts are identified by: i) dimensionality reduction of variables representing the individual health trajectories utilizing Principal Component Analysis, and ii) assigning patients to contexts via principal components and additional patient-level features using clustering. Two distinct contexts emerged, representing a lower- and a higher-risk subpopulation. CMDP-derived policies recommend: (i) follow-up within 1 month if lab value at current visit is unmeasured; (ii) up to 3 months for elevated lab values or recent hospitalizations; and (iii) 6 to 12 months for sustained glycemic control, with shorter follow-up intervals for patients in high-risk context. The optimal policies achieved lower expected cumulative cost than benchmarks (e.g., in the higher-comorbidity context, the CMDP policy reduced cost by about 34.8%, and in the lower-comorbidity context by about 6.4%, relative to an American Diabetes Association-like fixed interval follow-up policy. These findings demonstrate how context-aware approaches can inform adaptive follow-up strategies, and have the potential to advance chronic care management in primary care by synthesizing machine learning and probabilistic decision models.

23.
arXiv (CS.CL) 2026-06-18

Sumi: Open Uniform Diffusion Language Model from Scratch

Diffusion models have become a promising alternative to autoregressive models. Among these, uniform diffusion language models (UDLMs) permit any token to be updated at any step, in principle enabling more flexible generation. However, no UDLM has yet been pretrained from scratch at both large parameter scale and large token budget. Both autoregressive modeling and masked diffusion modeling already have capable models at scale that the community can study and build on; uniform diffusion has none. A scratch-pretrained UDLM at scale would provide a clean reference point for studying scaling behavior, generation dynamics, controllability, and trade-offs against established autoregressive and masked diffusion models. To this end, we introduce Sumi ("ink" in Japanese), a fully open 7B uniform diffusion language model pretrained from scratch on 1.5T tokens. Sumi performs competitively with autoregressive models trained at comparable token budgets on knowledge, reasoning, and coding benchmarks, while under-performing on commonsense benchmarks, where our education-heavy data mixture is a likely contributor. We release our model weights, checkpoints, and full training recipe, including a complete specification of the data mixture over publicly available corpora. We hope this release enables the community to study native uniform diffusion at scale and catalyzes work on its as-yet poorly understood aspects.

24.
arXiv (CS.AI) 2026-06-16

Where Did It Go Wrong? Process-Level Evaluation of Web Agents with Semantic State Tracking

arXiv:2606.15673v1 Announce Type: new Abstract: Web agents act through long interaction sequences, yet existing benchmarks evaluate only terminal success, discarding all process information and offering little guidance on improvement. In this work, we conduct a process-level analysis of web agents. We introduce WebStep, a benchmark of 1,800 task instances with controlled difficulty and automatic semantic state tracking. Each website exposes a deterministic semantic MDP alongside the GUI: the agent operates on the interface, while the environment records high-level states and transitions in the background, enabling fine-grained analysis without manual annotation. Based on the semantic trajectory, we first show that process metrics reveal differences invisible to outcome evaluation: three agents whose success rates cluster within 31-33% diverge in exploration reach versus execution accuracy. Then, decomposing by skill characterizes the nature of these differences, exposing opposite per-skill rankings hidden within the same website: e.g., on Housing, OpenAI CUA outperforms Qwen3.5 by 23.7% on commit actions yet underperforms it by 15.6% on filtering, pinpointing a concrete skill to improve even within a domain. Bifurcation analysis further localizes the decisive error that loses the task and shows that this error is agent-specific rather than shared. Finally, these differences widen as tasks grow harder: success rate is similar on easy tasks but separates sharply as exploration becomes more demanding. Our process-level analysis opens a new avenue in web agent evaluation, providing fine-grained and actionable insight into where and how each agent should be improved.

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

MVEB: Massive Video Embedding Benchmark

We introduce the Massive Video Embedding Benchmark (MVEB), a 23-task benchmark for video embeddings spanning classification, zero-shot classification, clustering, pair classification, retrieval, and video-centric question answering. We evaluate 33 models and find that no single model dominates: MLLM-based embeddings lead on classification, clustering, pair classification, and QA; multimodal binding leads on retrieval and zero-shot classification; generative MLLMs without contrastive adaptation collapse on cross-modal tasks. Paired video-only vs. audio+video evaluations show that audio's contribution depends on dataset annotation provenance: audio helps when labels were produced from both modalities and hurts when they were produced from visuals alone, a six-point gap consistent across model families. MVEB is derived from MVEB+, a 184-task pool, and is designed to maintain task diversity while reducing evaluation cost. It integrates into the MTEB ecosystem for unified evaluation across text, image, audio, and video. We release MVEB and all 184 tasks along with code and a leaderboard at https://github.com/embeddings-benchmark/mteb.