new

Get trending papers in your email inbox!

Subscribe

Daily Papers

byAK and the research community

Nov 20

Can We Trust In-Distribution Success? Locked Evaluation Reveals Transfer Failure and Sampling-Depth Entanglement in CRISPRi Perturbation Prediction

AI evaluation can support the wrong inference when an in-domain benchmark success does not survive distribution shift, or when the benchmark endpoint is entangled with a design factor. We study this problem in CRISPRi perturbation-effect prediction, evaluating a frozen Geneformer representation under a locked, pre-registered protocol: heads and model selection were frozen before test evaluation; the protocol required external outcome labels to remain withheld until final unblinding; and analysis-governing decisions were fixed before the evaluations they govern. In-distribution on the Virtual Cell Challenge (VCC), the frozen representation carries measurable predictive information beyond a dimension-matched random-feature control (Delta R^2 = +0.1645, 95% CI [+0.1375, +0.1920]), satisfying the pre-registered informativeness gate required before interpreting transfer. It then fails zero-shot transfer on both external screens (Spearman rho = -0.139 and -0.267), lying below that control on each. Adding a predefined magnitude block improves the representation externally (Delta rho = +0.032 and +0.143) but, under the frozen primary head, does not rescue transfer: both remain negative. A pre-registered, count-adjusted max-response secondary is positively associated with the outcome on both screens; we report it as correlational and secondary, not as a recovered magnitude signal. Finally, the VCC endpoint is strongly sample-size associated: a count-only linear model reaches R^2 = +0.4325, versus +0.2589 for the four magnitude scalars; adding those scalars to cell count improves R^2 by only +0.0017, so much of the aggregate-magnitude signal overlaps with cell count. This case study shows how locking the evaluation, harmonizing the measured endpoint, and separating primary from secondary evidence can change the inference supported by an AI benchmark.

  • 2 authors
·
Aug 30

Humanized but Still True? Auditing Facts and Detection in AI-Text Humanizers

AI "humanizers" rewrite machine-generated text so that detectors classify it as human-written. Public leaderboards score evasion and embedding similarity, but not whether checkable details survive. We check whether numbers, dates, person and organization names, and quoted spans survive each rewrite, using a rule-based checker validated by two AI annotators (precision 0.84-1.00, recall 0.93-0.95; no human annotation). In the October 2026 cycle of a public humanizer leaderboard, all 11 commercial tools changed or dropped at least one of these units in 35-65% of the documents that contained one. Most changes were to quoted spans; counting numbers and names alone, 10 of the 11 tools still altered at least one in 20% or more of the documents. The September 2026 cycle was similar (19-50%). An open 12B humanizer kept at least 97% of numbers and names in fact-dense U.S. federal agency text, but reworded 67-78% of the quoted spans, against its own prompt. Unit retention and detector evasion were not clearly related across tools (Spearman rho = -0.46, 95% CI -0.65 to 0.08), and the open model's four released files (two of them quantization-aware low-bit files) did not show the pre-registered monotone trade-off. On federal texts with matched human, AI-styled and humanized versions, a general-purpose decision model, Jev, asked a single zero-shot question, separated human originals from humanized rewrites with AUROC 0.864 (0.827-0.902), 0.18-0.22 higher than Binoculars, Fast-DetectGPT and the adversarially trained RADAR, while calling 24% of the human originals AI at a 0.5 threshold. On the AI-styled rewrites by 2026 models, both perplexity-based detectors scored below chance (AUROC 0.37 and 0.39), also with line breaks removed. Hypotheses, data rules and analyses were frozen before the main runs; two earlier data contacts are disclosed. We release the checker and the per-item results.

  • 3 authors
·
Oct 29 3

Finding Duplicates in 1.1M BDD Steps: cukereuse, a Paraphrase-Robust Static Detector for Cucumber and Gherkin

Behaviour-Driven Development (BDD) suites accumulate step-text duplication whose maintenance cost is established in prior work. Existing detection techniques require running the tests (Binamungu et al., 2018-2023) or are confined to a single organisation (Irshad et al., 2020-2022), leaving a gap: a purely static, paraphrase-robust, step-level detector usable on any repository. We fill the gap with cukereuse, an open-source Python CLI combining exact hashing, Levenshtein ratio, and sentence-transformer embeddings in a layered pipeline, released alongside an empirical corpus of 347 public GitHub repositories, 23,667 parsed .feature files, and 1,113,616 Gherkin steps. The step-weighted exact-duplicate rate is 80.2 %; the median-repository rate is 58.6 % (Spearman rho = 0.51 with size). The top hybrid cluster groups 20.7k occurrences across 2.2k files. Against 1,020 pairs manually labelled by the three authors under a released rubric (inter-annotator Fleiss' kappa = 0.84 on a 60-pair overlap), we report precision, recall, and F1 with bootstrap 95 % CIs under two protocols: the primary rubric and a score-free second-pass relabelling. The strongest honest pair-level number is near-exact at F1 = 0.822 on score-free labels; the primary-rubric semantic F1 = 0.906 is inflated by a stratification artefact that pins recall at 1.000. Lexical baselines (SourcererCC-style, NiCad-style) reach primary F1 = 0.761 and 0.799. The paper also presents a CDN-structured critique of Gherkin (Cognitive Dimensions of Notations); eight of fourteen dimensions are rated problematic or unsupported. The tool, corpus, labelled pairs, rubric, and pipeline are released under permissive licences.

  • 3 authors
·
Apr 21 1

MTEB-PT: A Text Embedding Benchmark for Brazilian Portuguese

Text embeddings for Portuguese have no dedicated benchmark: evaluation rests on translated corpora such as English MS MARCO or on thin multilingual coverage, with native tasks scattered and unconsolidated. We introduce MTEB-PT, a benchmark of 22 native Brazilian-Portuguese tasks across seven categories (classification, multilabel classification, pair classification, semantic textual similarity, clustering, retrieval, and reranking), admitting only data created or found in Portuguese and excluding translations by construction. We evaluate 93 models spanning 23M to 27B parameters: 73 open-weight and 20 closed commercial APIs. Alongside the leaderboard we report a statistical layer for every headline comparison: per-task bootstrap confidence intervals, paired-bootstrap significance, a task- and instance-level discrimination analysis (how sharply each task separates models) adapted from Item Response Theory, and a cross-leaderboard correlation. Three findings stand out. The benchmark cleanly separates about a dozen tiers of models, though the top six are statistically too close to order. An openly licensed, self-hostable model reaches that leading tier, so strong Portuguese embedding quality does not require a commercial API. And a model's rank on the global multilingual leaderboard predicts its Portuguese rank only moderately (Spearman rho = 0.75 over 55 shared models; one model ranks 3rd there and 49th here), so a native benchmark measures something the multilingual boards do not. We release every task, our code, and a public leaderboard, so practitioners can choose Portuguese embedding models on native evidence.

MTEB-BR MTEB-BR
·
Jul 5

Capability-Guided Compression: Toward Interpretability-Aware Budget Allocation for Large Language Models

Large language model compression has made substantial progress through pruning, quantization, and low-rank decomposition, yet a fundamental limitation persists across all existing methods: compression budgets are allocated without any representation of what individual model components functionally encode. We term this the capability-blind compression problem and argue it is a root cause of two well-documented failures -- the insensitivity of perplexity-based evaluation to reasoning capability loss, and the abrupt phase transitions in model performance recently characterized by Ma et al. (2026). We propose Capability-Guided Compression (CGC), a framework that addresses this by using Sparse Autoencoder (SAE)-derived capability density maps to allocate differential compression budgets across transformer components. Capability density is a formally defined scalar measure combining the feature breadth, activation entropy, and cross-input consistency of a component's SAE feature activation distribution. We prove theoretically that components with higher capability density exhibit lower structural redundancy and reach their individual phase transition points at lower compression ratios, providing the first pre-compression mechanism for component-level phase transition prediction. Experiments on GPT-2 Medium confirm that capability density is statistically independent of Wanda importance scores (Spearman rho = -0.054, n = 384 heads), establishing it as a genuinely novel compression signal orthogonal to all existing importance metrics. We report a negative result on PPL-based compression comparison and provide a principled diagnosis identifying GPT-2 Medium as an insufficient test bed for the full CGC hypothesis. The theoretical framework, density formalism, and orthogonality finding constitute a foundation for capability-aware compression research.

  • 1 authors
·
Mar 16

PRISM: Pathology Reliability In Scarce-label Medicine. A Tile-level Benchmark of Pathology Foundation Models

Pathology foundation models are typically benchmarked on classification performance at full label availability and on in-distribution test sets, while calibration, behavior at low label budgets, and robustness under cross-institutional shift are seldom evaluated jointly. We introduce PRISM, a tile-level benchmark that measures the reliability of pathology foundation models along four axes: model choice, label fraction, calibration, and out-of-distribution transfer. PRISM evaluates eight foundation models (CLIP, PLIP, CONCH, UNI, Virchow2, GigaPath, H-Optimus-0, Midnight) on six tile-level datasets at six label fractions (1% to 100%) with three seeds (864 linear probe runs), extended with a five-hospital Camelyon17 study of 20 directed transfer pairs. Four phenomena emerge. (i) Under full supervision, AUROC rank predicts post-temperature calibration rank consistently across datasets (Spearman rho = 0.52, I^2 = 0%), whereas at 1% labels the relationship becomes dataset-specific (rho = -0.24, I^2 = 67%). (ii) On harder datasets the linear probe collapses to a single predicted class below a dataset-specific label fraction. (iii) Whether additional source-domain labels help or harm target calibration tracks how well the probe transfers: below a transfer AUROC of roughly 0.85, more labels worsen target ECE in 86-88% of combinations. (iv) Temperature scaling fails to reduce ECE in 49 of 288 cells, miscalibration is predominantly under-confidence, and a source-fitted temperature is harmful under failing transfer. PRISM is released as a pip-installable package with reference results for all 288 cells. Code and data: https://github.com/alicemkyn/prism-benchmark

  • 1 authors
·
Oct 26

YourBench: Easy Custom Evaluation Sets for Everyone

Evaluating large language models (LLMs) effectively remains a critical bottleneck, as traditional static benchmarks suffer from saturation and contamination, while human evaluations are costly and slow. This hinders timely or domain-specific assessment, crucial for real-world applications. We introduce YourBench, a novel, open-source framework that addresses these limitations by enabling dynamic, automated generation of reliable, up-to-date, and domain-tailored benchmarks cheaply and without manual annotation, directly from user-provided documents. We demonstrate its efficacy by replicating 7 diverse MMLU subsets using minimal source text, achieving this for under 15 USD in total inference costs while perfectly preserving the relative model performance rankings (Spearman Rho = 1) observed on the original benchmark. To ensure that YourBench generates data grounded in provided input instead of relying on posterior parametric knowledge in models, we also introduce Tempora-0325, a novel dataset of over 7K diverse documents, published exclusively after March 2025. Our comprehensive analysis spans 26 SoTA models from 7 major families across varying scales (3-671B parameters) to validate the quality of generated evaluations through rigorous algorithmic checks (e.g., citation grounding) and human assessments. We release the YourBench library, the Tempora-0325 dataset, 150k+ question answer pairs based on Tempora and all evaluation and inference traces to facilitate reproducible research and empower the community to generate bespoke benchmarks on demand, fostering more relevant and trustworthy LLM evaluation.

  • 6 authors
·
Apr 2, 2025 3

Speaking in Words, Thinking in Logic: A Dual-Process Framework in QA Systems

Recent advances in large language models (LLMs) have significantly enhanced question-answering (QA) capabilities, particularly in open-domain contexts. However, in closed-domain scenarios such as education, healthcare, and law, users demand not only accurate answers but also transparent reasoning and explainable decision-making processes. While neural-symbolic (NeSy) frameworks have emerged as a promising solution, leveraging LLMs for natural language understanding and symbolic systems for formal reasoning, existing approaches often rely on large-scale models and exhibit inefficiencies in translating natural language into formal logic representations. To address these limitations, we introduce Text-JEPA (Text-based Joint-Embedding Predictive Architecture), a lightweight yet effective framework for converting natural language into first-order logic (NL2FOL). Drawing inspiration from dual-system cognitive theory, Text-JEPA emulates System 1 by efficiently generating logic representations, while the Z3 solver operates as System 2, enabling robust logical inference. To rigorously evaluate the NL2FOL-to-reasoning pipeline, we propose a comprehensive evaluation framework comprising three custom metrics: conversion score, reasoning score, and Spearman rho score, which collectively capture the quality of logical translation and its downstream impact on reasoning accuracy. Empirical results on domain-specific datasets demonstrate that Text-JEPA achieves competitive performance with significantly lower computational overhead compared to larger LLM-based systems. Our findings highlight the potential of structured, interpretable reasoning frameworks for building efficient and explainable QA systems in specialized domains.

  • 8 authors
·
Jul 27, 2025

PEEM: Prompt Engineering Evaluation Metrics for Interpretable Joint Evaluation of Prompts and Responses

Prompt design is a primary control interface for large language models (LLMs), yet standard evaluations largely reduce performance to answer correctness, obscuring why a prompt succeeds or fails and providing little actionable guidance. We propose PEEM (Prompt Engineering Evaluation Metrics), a unified framework for joint and interpretable evaluation of both prompts and responses. PEEM defines a structured rubric with 9 axes: 3 prompt criteria (clarity/structure, linguistic quality, fairness) and 6 response criteria (accuracy, coherence, relevance, objectivity, clarity, conciseness), and uses an LLM-based evaluator to output (i) scalar scores on a 1-5 Likert scale and (ii) criterion-specific natural-language rationales grounded in the rubric. Across 7 benchmarks and 5 task models, PEEM's accuracy axis strongly aligns with conventional accuracy while preserving model rankings (aggregate Spearman rho about 0.97, Pearson r about 0.94, p < 0.001). A multi-evaluator study with four models shows consistent relative judgments (pairwise rho = 0.68-0.85), supporting evaluator-agnostic deployment. Beyond alignment, PEEM captures complementary linguistic failure modes and remains informative under prompt perturbations: prompt-quality trends track downstream accuracy under iterative rewrites, semantic adversarial manipulations induce clear score degradation, and meaning-preserving paraphrases yield high stability (robustness rate about 76.7-80.6%). Finally, using only PEEM scores and rationales as feedback, a zero-shot prompt rewriting loop improves downstream accuracy by up to 11.7 points, outperforming supervised and RL-based prompt-optimization baselines. Overall, PEEM provides a reproducible, criterion-driven protocol that links prompt formulation to response behavior and enables systematic diagnosis and optimization of LLM interactions.

  • 4 authors
·
Mar 11

Forecasting Downstream Performance of LLMs With Proxy Metrics

Progress in language model development is often driven by comparative decisions: which architecture to adopt, which pretraining corpus to use, or which training recipe to apply. Making these decisions well requires reliable performance forecasts, yet the two commonly used signals are fundamentally limited. Cross-entropy loss is poorly aligned with downstream capabilities, and direct downstream evaluation is expensive, sparse, and often uninformative at early training stages. Instead, we propose to construct proxy metrics by aggregating token-level statistics, such as entropy, top-k accuracy, and expert token rank, from a candidate model's next token distribution over expert-written solutions. Across three settings, our proxies consistently outperform loss- and compute-based baselines: 1) For cross-family model selection, they rank a heterogeneous population of reasoning models with mean Spearman Rho = 0.81 (vs. Rho = 0.36 for cross-entropy loss); 2) For pretraining data selection, they reliably rank 25 candidate corpora for a target model at roughly 10{,}000times less compute than direct evaluation, pushing the Pareto frontier beyond existing methods; and 3) for training-time forecasting, they extrapolate downstream accuracy across an 18times compute horizon with roughly half the error of existing alternatives. Together, these results suggest that expert trajectories are a broadly useful source of signal for assessing model capabilities, enabling reliable performance forecasting throughout the model development life cycle.

When Does q-error Predict Plan Regret? Three Regimes of Cardinality-Estimation Error

Cardinality-estimation (CE) research ranks estimators by q-error, yet it is well known that q-error is an imperfect proxy for query-plan quality. We give a measurement-driven account of when it is a good proxy and when it is not, and why. Modeling plan selection as an argmin over a piecewise-linear cost landscape, we find that plan regret (the cost of the chosen plan relative to the optimal, under true cardinalities) is governed by plan-cost geometry in a regime-dependent way. (i) For small errors, a true-point condition number kappa predicts regret and out-predicts q-error; its predictive power decays to zero as error grows, as a local linearization must. (ii) For large errors -- where deployed learned estimators operate -- an estimator-independent average-case sub-optimality measure ACS-infinity predicts which queries are regret-prone (Spearman rho ~ 0.54 on STATS-CEB), while q-error is nearly uninformative at the query level (rho ~ 0.05). (iii) The worst case is Haritsa's maximum sub-optimality (MSO). The three are one cost-ratio spectrum under three weightings. We prove a limit law ACS-infinity = sum_k r_k pi_k with cardinality-independent combinatorial weights, and validate every claim on STATS-CEB and JOB-light with four released estimators under pre-registered decision rules, and confirm on real PostgreSQL runtime that ACS-infinity predicts regret where q-error does not. The contribution is conceptual and empirical -- an average-case companion to worst-case robust query optimization, and a characterization of when an accuracy metric tracks plan quality -- rather than a new estimator. Code and the full pre-registration are public.

  • 2 authors
·
Jun 13

Quantisation Reshapes the Metacognitive Geometry of Language Models

We report that model quantisation restructures domain-level metacognitive efficiency in LLMs rather than degrading it uniformly. Evaluating Llama-3-8B-Instruct on the same 3,000 questions at Q5_K_M and f16 precision, we find that M-ratio profiles across four knowledge domains are uncorrelated between formats (Spearman rho = 0.00). Arts & Literature moves from worst-monitored (M-ratio = 0.606 at Q5_K_M) to best-monitored (1.542 at f16). Geography moves from well-monitored (1.210) to under-monitored (0.798). However, Type-2 AUROC profiles are perfectly stable across formats (rho = 1.00), localising the restructuring to the M-ratio normalisation rather than the underlying discrimination signal. This finding emerged from a pre-registered attempt to improve metacognition through domain-conditional training. We prescribed confidence-amplification SFT for the diagnosed weak domain, with matched-budget agnostic and wrong-prescription controls. All four confirmatory hypotheses were null (10,000 bootstrap resamples, seed = 42). The training successfully reshaped confidence distributions, doubling the NLP gap in Science from 0.076 to 0.152, but did not improve meta-d' because the diagnostic profile did not transfer across formats. Any system relying on domain-level M-ratio profiles has an unexamined dependency on inference format. Systems using AUROC_2 are safer. We release all code, pre-registrations, and trial-level data.

  • 1 authors
·
Apr 9

Open versioned aggregates of the FAA National Wildlife Strike Database (1990-2026): conditional damage rates by animal family, airframe, and airport

Background. Aircraft-wildlife collisions are a documented aviation hazard. The FAA National Wildlife Strike Database is the main public source of U.S. reports, but raw extracts are large and rarely released as versioned aggregates. Methods. We convert the public FAA Access export (STRIKE_REPORTS) to SQLite, define a report as damaging when DAMAGE_LEVEL is one of M, M?, S, D, and fold rows into open aggregates (year, airport, species, aircraft, animal family) via hierarchical SPECIES_ID prefixes. We analyze all civil reports and a commercial subset (AC_MASS 3-5). Rates are per reported strike, not per flight hour, so every rate is conditional on a report having been filed. Trend statistics use complete incident years 1990-2025; cross-sectional facets cover the full seed, incident years 1990-2026. Results. Numbers are locked to the 2026-07-04 freeze. Reported strikes in 2025 reach a series high (24,458 total; 901 damaging). Species that dominate strike counts are not those with the highest damage rates: perching songbirds account for 41.1% of identified strikes but damage aircraft in 1.5% of reports (95% CI 1.4-1.6), whereas deer account for 0.7% but damage aircraft in 81.8% of reports (95% CI 79.8-83.7). Airports ranked by total strikes (led by Denver Intl) diverge from rankings by damaging strikes (led by Sacramento Intl); across the 452 airports with at least 50 reports, volume and conditional damage rate are negatively rank-correlated (Spearman rho = -0.44, p < 0.001); large stations plausibly capture minor events more completely, so this is not a hazard ranking. The share of reports labeled damaging declines over decades, consistent with improved reporting of minor events, while absolute damaging counts peak in 2025. Conclusions. Versioned open aggregates and a documented damage-rate analysis enable independent reuse; released under CC BY 4.0.

  • 1 authors
·
Aug 1

Mirror-Score: Calibrated, Inference-only Scoring Exposes the Limits of Sequence-compatibility Ranking in D-peptide Design

D-peptides combine protease resistance with high target specificity, but computational design of D-peptide binders remains immature. Mirror-Peptidizer introduced an in silico mirror-image screening pipeline using target reflection, backbone generation, and ProteinMPNN sequence design, but its raw ProteinMPNN negative log-likelihood (NLL) ranking was not validated against measured affinities, and only 4 of 9 tested MDM2 designs bound detectably. We introduce Mirror-Score, a calibrated, inference-only scoring framework for heterochiral D-peptide/L-protein complexes, and a public benchmark of 31 crystal complexes across four target families, including 18 with literature-verified affinities. Raw ProteinMPNN NLL is not a valid affinity ranker: its pooled Spearman correlation with affinity is 0.19, and correlations reverse between MDM2/CHIP (+0.62) and gp41 (-0.70). We therefore evaluate Boltz-2 mirror-space cofolding confidence. For the complete viral-entry family (7 structures representing 3 peptides), interface predicted local distance difference test (pLDDT) achieves structure-level leave-one-out Spearman rho = 0.90 (p = 0.006) and correctly orders all three peptides by affinity, whereas NLL fails (structure-level rho = 0.18). Because only three independent chemotypes are represented, this result indicates directional consistency rather than a statistically validated predictor. Cross-family calibration does not transfer at current sample sizes, supporting family-matched calibration as the practical deployment mode. We also specify a prospective design protocol for the antimicrobial-resistance targets LasR and LecB from Pseudomonas aeruginosa, including mirrored structures, ligand-derived hotspot maps, diffusion-model-ready inputs, and Mirror-Score ranking. Code, benchmark data, structures, and analysis scripts are openly available at https://github.com/Jiadalee/Mirror-Score.

  • 1 authors
·
Sep 27

SO-101 Bench: Accessible Spatial Manipulation

Low-cost arms such as the SO-101 have become the most common entry point to robot learning, yet no released policy performs spatially constrained, language-conditioned tasks on them out of the box, and existing SO-101 benchmarks run only on physical hardware with simple tasks and few held-out objects. We introduce SO-101 Bench, a simulated benchmark for language-conditioned spatial manipulation built on an Isaac Lab digital twin of a real SO-101 setup. It has four tasks scored as five families, 138 objects in seen, unseen-instance and unseen-class splits, 916 evaluation episodes with subsets for held-out appearances, rotate-to-fit planning and occlusion memory, and 3,629 simulated demonstrations on which every learned policy is fine-tuned under an equal compute budget. The twin ranks policies as the real robot does: five policies trained without real data, four VLAs and an LLM agent, keep their exact ranking when the same weights run on the physical arm in an unseen environment with unseen objects (Spearman rho = 1.00, MMRV = 0.000), although simulation overestimates success. Across nine VLAs and world-action models, the best learned policies score near 61%, with spatial placement, unseen object classes and rotate-to-fit planning remaining open. An LLM agent that writes code and hands grasping to a learned policy scores 83.2%, and an ablation shows that this lead comes from the choice of harness and the choice of language model.

  • 1 authors
·
Oct 26 3

BridgeRAG: Training-Free Bridge-Conditioned Retrieval for Multi-Hop Question Answering

Multi-hop retrieval is not a single-step relevance problem: later-hop evidence should be ranked by its utility conditioned on retrieved bridge evidence, not by similarity to the original query alone. We present BridgeRAG, a training-free, graph-free retrieval method for retrieval-augmented generation (RAG) over multi-hop questions that operationalizes this view with a tripartite scorer s(q,b,c) over (question, bridge, candidate). BridgeRAG separates coverage from scoring: dual-entity ANN expansion broadens the second-hop candidate pool, while a bridge-conditioned LLM judge identifies the active reasoning chain among competing candidates without any offline graph or proposition index. Across four controlled experiments we show that this conditioning signal is (i) selective: +2.55pp on parallel-chain queries (p<0.001) vs. ~0 on single-chain subtypes; (ii) irreplaceable: substituting the retrieved passage with generated SVO query text reduces R@5 by 2.1pp, performing worse than even the lowest-SVO-similarity pool passage; (iii) predictable: cos(b,g2) correlates with per-query gain (Spearman rho=0.104, p<0.001); and (iv) mechanistically precise: bridge conditioning causes productive re-rankings (18.7% flip-win rate on parallel-chain vs. 0.6% on single-chain), not merely more churn. Combined with lightweight coverage expansion and percentile-rank score fusion, BridgeRAG achieves the best published training-free R@5 under matched benchmark evaluation on all three standard MHQA benchmarks without a graph database or any training: 0.8146 on MuSiQue (+3.1pp vs. PropRAG, +6.8pp vs. HippoRAG2), 0.9527 on 2WikiMultiHopQA (+1.2pp vs. PropRAG), and 0.9875 on HotpotQA (+1.35pp vs. PropRAG).

  • 1 authors
·
Apr 2

SADGE: Structure and Appearance Domain Gap Estimation of Synthetic and Real Data

We propose SADGE, a quantitative similarity metric that predicts the performance of synthetic image datasets for common computer vision tasks without downstream model training. Estimating whether a synthetic dataset will lead to a model that performs well on real-world data remains a bottleneck in model development. Existing evaluation metrics (e.g., PSNR, FID, CLIP) primarily measure semantic alignment between real and synthetic images (Appearance Similarity Score). Less commonly, structural similarity between images is considered to assess the domain gap (Geometric Similarity Score). However, to the best of our knowledge there exists no studies that evaluate which similarity metric is the best downstream predictor for a given synthetic dataset. In this paper, we show over a wide variety of different synthetic datasets and downstream tasks that neither appearance nor geometry alone can reliably predict downstream performance; rather, it is their non-linear interplay that dictates synthetic data utility. Specifically, we measure how commonly used Appearance and Geometric Similarity metrics computed between synthetic and real images correlate with downstream performance in object detection, semantic segmentation, and pose estimation. Across five public synthetic-to-real benchmark families and 15 dataset-level variants (79k image pairs), SADGE achieves the strongest association with downstream transfer performance under both linear and rank-based criteria, reaching Pearson r=0.88 and Spearman rho=0.77. We compute for each combination of geometry-based methods and appearance-based approaches SADGE scores across all benchmark families. The best configuration is obtained by fusing DINOv3 appearance similarity with MASt3R geometric consistency through a constrained bilinear interaction, outperforming both the strongest geometry-only baseline and the strongest appearance-only baseline .

  • 5 authors
·
May 20

GE-Act 2.0: Pretraining and Scaling a World-Action Model for Robotic Manipulation

World-action models (WAM) predict future states to guide robot actions, enabling learning from both action-free video and action-labeled interaction. Most inherit pretrained video generators, leaving WAM pretraining and scaling underexplored. We introduce Genie Envisioner Act 2.0 (GE-Act 2.0), a world-action model whose trainable generative and action components are all initialized from scratch on manipulation data. It combines a control-oriented autoencoder (CoAE), a single-step visual planner (SVP), and an inverse dynamics model (IDM). CoAE retains action- and instruction-relevant information under aggressive compression, while SVP produces a complete future state in one differentiable pass, so visual planning and inverse dynamics can be pretrained separately on complementary data. The components are then jointly trained with knowledge-aligned selective optimization (KASO), which reduces mismatched supervision by selecting only predicted futures judged behaviorally compatible with the recorded action. We evaluate pretrained checkpoints directly, without per-task fine-tuning, on 100 tasks across 20 manipulation skill groups with held-out scenes, backgrounds, lighting, and object instances. Scaling co-training data from 300 to 30,000 hours raises success from 17.1% to 44.1% on G1-OP and from 13.4% to 31.1% on G2-90D; despite comprising less than 2% of the co-training data, G2-90D improves by 17.7 points, suggesting cross-embodiment transfer. Gains span 19/20 and 18/20 skill groups, and skill-specific coverage strongly correlates with zero-shot out-of-distribution (OOD) success (Pearson r=0.80; Spearman rho=0.85). Under the same protocol, the model grounds object, color, shape, and position references in at least 90% of trials and follows explicit instructions even when they conflict with an already-committed behavior or a conventional scene association.

Measuring Collapse and Correction in Homogeneous-Panel LLM Debate

Multi-agent large language model (LLM) debate is often evaluated by whether final answers improve, but movement is not necessarily improvement: the same discussion can rescue an initially wrong majority or destroy an initially correct one. Standard final-accuracy evaluations conflate these opposing mechanisms. We introduce an auditable protocol for homogeneous debate on multiple-choice questions (MCQs) that records each run as a transition ledger over collapse, correction, onset, and signed intervention utility. On 6,925 MMLU-Pro debates, the protocol identifies 253 collapses and a parallel correction ledger that changes how interventions should be judged. Replay experiments reveal the central tradeoff: a leave-one-model-out probe-gated freeze prevents 29 collapses but loses 108 corrections under equal weights, so collapse prevention alone can recommend the wrong policy. A compact pre-debate 8-probe screen is a triage signal: its unadjusted family-level association with conditional-collapse risk is high (G=7, Spearman rho=0.893, exact two-sided p=0.0123), but initial-majority accuracy is a close comparator (rho=0.821; family partial rho=0.767, p=0.0877), so we do not treat it as calibrated or capability-adjusted prediction. Round-level traces localize many collapses to the first debate round, where early disagreement can precede both harmful cascades and useful recovery. We release replayable schemas, coders, audits, cost cards, and zero-API rebuild scripts so future model-scaffold rows can be compared under the same denominators and signed utility ledger.

Representation Drift in Continual Learning: LoRA versus N-gram Memory

A language model that keeps learning online must not slowly erode its own base capabilities. We ask which part of a modern LLM should absorb continual updates, by measuring how each choice changes the model's internal representations. On a 48-layer mixture-of-experts model with a native pretrained N-gram memory, we train a seven-task chain under three update mechanisms with identical data and budgets: updating only the N-gram memory table, continually training one shared LoRA adapter, and training a fresh LoRA per task that is merged into the backbone. We capture fixed-context representations at every layer and task boundary. Updating only the N-gram memory changes the base model's representations of general math and code inputs by 26%/15% (relative RMS), against 64%/35% for merged and 97%/39% for shared LoRA. This drift does not grow with the number of tasks: it reaches 25% after the first task and stays there for all seven, while both LoRA variants keep accumulating. Its generic behaviour is essentially unchanged (no code problem lost, math NLL -0.01), and every change is confined to inspectable, reversible table rows. The cost appears elsewhere: N-gram memory acquires new tasks more weakly and its stored tasks interfere with one another, losing as much old-task NLL as shared LoRA. We show that this forgetting is governed by drift on the learned tasks' own inputs, measured from the checkpoint at which each was learned, not by generic drift (Spearman rho=0.85; rho=0.71 on Llama-3-8B), and we derive the exact relations behind it: with a frozen head, every pairwise logit margin changes by the projection of drift onto a head-difference direction. These results support a design principle for long-running online learning: write new knowledge only into the memory table and keep the backbone fixed; the remaining problem, interference between stored tasks, then lies inside the memory.

  • 1 authors
·
Oct 9

Do not copy and paste! Rewriting strategies for code retrieval

Embedding-based code retrieval often suffers when encoders overfit to surface syntax. Prior work mitigates this by using LLMs to rephrase queries and corpora into a normalized style, but leaves two questions open: how much representational shift helps, and when is the per-query LLM call justified? We study a hierarchy of three rewriting strategies: stylistic rephrasing, NL-enriched PseudoCode, and full Natural-Language transcription, under joint query-corpus (QC, online) and corpus-only (C, offline) augmentation, across six CoIR benchmarks, five encoders, and three rewriters spanning independent model families (Qwen, DeepSeek, Mistral). We are the first to evaluate NL-enriched PseudoCode and snippet-level Natural Language as direct retrieval representations, rather than as transient intermediates. Full NL rewriting with QC yields the largest gains (+0.51 absolute NDCG@10 on CT-Contest for MoSE-18), while corpus-only rewriting degrades retrieval in 56 of 90 configurations, about 62%. We introduce two diagnostics, Delta H, token entropy, and Delta s, embedding cosine, and show that Delta H predicts retrieval gain under QC across all three rewriter families: pooled Spearman rho = +0.436, p < 0.001 on DeepSeek+Codestral; rho = +0.593 on Codestral alone; rho = +0.356 on Qwen. This establishes Delta H as a cheap, rewriter-agnostic proxy for deciding when rewriting pays off before running retrieval. Our analysis reframes LLM rewriting as a cost-benefit decision: it is most effective as a remediation layer for lightweight encoders on code-dominant queries, with diminishing returns for strong encoders or NL-heavy queries.

  • 3 authors
·
May 7 1

Representation Before Training: A Fixed-Budget Benchmark for Generative Medical Event Models

Every prediction from a generative medical event model is bounded by how clinical events are tokenized, yet input representation is rarely isolated from other system and architectural choices. We evaluate how representation decisions affect downstream prediction after a shared one-epoch pretraining budget. We train 28 matched transformers on MIMIC-IV and evaluate them on 30 clinical outcomes in three experiments: (1) quantization granularity, reference-range anchoring, and code-value fusion; (2) value encoding (hard bins, soft discretization, code-normalized xVal) crossed with temporal encoding (event order, time tokens, admission-relative RoPE); and (3) native MIMIC laboratory/vital codes versus the Common Longitudinal ICU Format (CLIF)-remapped laboratory/vital codes with compression-preserving perturbation arms. In Experiment 1, fused code-value tokenization improves mortality AUROC from 0.891 to 0.915 (BH-adjusted p < 0.001), hospital length-of-stay AUROC from 0.763 to 0.788 (BH-adjusted p < 0.001), and, for the decile fused-vs-unfused comparison, mean regression Spearman rho across the 13 regression outcomes from 0.414 to 0.494. Across the three temporal encodings, event order only and admission-relative RoPE match or exceed inserting time tokens on average while shortening sequences by 11%. CLIF remapping preserves downstream performance in our single-site setting while yielding a smaller, clinically interpretable token set compatible with multi-site use. Finer-than-decile quantization, reference-range anchoring, and soft discretization help in selective outcomes, while code-normalized xVal remains well below the discrete and soft families, consistent with near-median suppression that persists after the affine variant.

  • 6 authors
·
Apr 17

Orthogonal Ensembles and Tested Explanations for Performer-Independent Body-Motion Emotion Recognition

We study body-only, 12-class acted-emotion classification from skeleton motion under leave-performer-out (LPO) evaluation, a hard, underdetermined setting: chance is 8.3%, and a protocol-matched reproduced STGCN++ baseline reaches only 25.73 +/- 4.03% Macro-F1. We show that reliable gains come not from a new architecture but from combining eleven models with orthogonal error modes: under 10-fold LPO cross-validation on the labeled training performers, an equal-weight logit-mean ensemble reaches 36.80 +/- 4.00% per-fold Macro-F1, a protocol-matched +11.07 pp (+43% relative) over the same-split reproduced baseline. Our central contribution is a tested explanation suite: for a strong ensemble member, part-masking and counterfactual edits show (rather than assert) that its decisions depend on motion-grounded body-region evidence, and this region saliency aligns with rule-based Laban Movement Analysis (LMA) attributes far more than with classical kinematics: region-level saliency-LMA Spearman rho = +0.500 versus +0.033, roughly 15x, and the alignment holds for the submitted 11-way ensemble itself at rho = +0.517; the audit is post hoc and needs no retraining. The same suite faithfully reports a negative: within-window temporal saliency is diffuse rather than localized. On the hidden challenge test set the submitted ensemble scored 37.23 % Macro-F1 and received the Best Performance Award of the MMAC Challenge 2026 (Human score is 39 %). Code is available at https://github.com/nawta/diema-challenge and the presentation at https://nawta.github.io/mmac2026/.

  • 2 authors
·
Sep 14

PoSh: Using Scene Graphs To Guide LLMs-as-a-Judge For Detailed Image Descriptions

While vision-language models (VLMs) have advanced into detailed image description, evaluation remains a challenge. Standard metrics (e.g. CIDEr, SPICE) were designed for short texts and tuned to recognize errors that are now uncommon, such as object misidentification. In contrast, long texts require sensitivity to attribute and relation attachments and scores that localize errors to particular text spans. In this work, we introduce PoSh, a metric for detailed image description that uses scene graphs as structured rubrics to guide LLMs-as-a-Judge, producing aggregate scores grounded in fine-grained errors (e.g. mistakes in compositional understanding). PoSh is replicable, interpretable and a better proxy for human raters than existing metrics (including GPT4o-as-a-Judge). To validate PoSh, we introduce a challenging new dataset, DOCENT. This novel benchmark contains artwork, paired with expert-written references, and model-generated descriptions, augmented with granular and coarse judgments of their quality from art history students. Thus, DOCENT enables evaluating both detailed image description metrics and detailed image description itself in a challenging new domain. We show that PoSh achieves stronger correlations (+0.05 Spearman rho) with the human judgments in DOCENT than the best open-weight alternatives, is robust to image type (using CapArena, an existing dataset of web imagery) and is a capable reward function, outperforming standard supervised fine-tuning. Then, using PoSh, we characterize the performance of open and closed models in describing the paintings, sketches and statues in DOCENT and find that foundation models struggle to achieve full, error-free coverage of images with rich scene dynamics, establishing a demanding new task to gauge VLM progress. Through both PoSh and DOCENT, we hope to enable advances in important areas such as assistive text generation.

columbia Columbia University
·
Oct 21, 2025

Potential and Perils of Large Language Models as Judges of Unstructured Textual Data

Rapid advancements in large language models have unlocked remarkable capabilities when it comes to processing and summarizing unstructured text data. This has implications for the analysis of rich, open-ended datasets, such as survey responses, where LLMs hold the promise of efficiently distilling key themes and sentiments. However, as organizations increasingly turn to these powerful AI systems to make sense of textual feedback, a critical question arises, can we trust LLMs to accurately represent the perspectives contained within these text based datasets? While LLMs excel at generating human-like summaries, there is a risk that their outputs may inadvertently diverge from the true substance of the original responses. Discrepancies between the LLM-generated outputs and the actual themes present in the data could lead to flawed decision-making, with far-reaching consequences for organizations. This research investigates the effectiveness of LLMs as judge models to evaluate the thematic alignment of summaries generated by other LLMs. We utilized an Anthropic Claude model to generate thematic summaries from open-ended survey responses, with Amazon's Titan Express, Nova Pro, and Meta's Llama serving as LLM judges. The LLM-as-judge approach was compared to human evaluations using Cohen's kappa, Spearman's rho, and Krippendorff's alpha, validating a scalable alternative to traditional human centric evaluation methods. Our findings reveal that while LLMs as judges offer a scalable solution comparable to human raters, humans may still excel at detecting subtle, context-specific nuances. This research contributes to the growing body of knowledge on AI assisted text analysis. We discuss limitations and provide recommendations for future research, emphasizing the need for careful consideration when generalizing LLM judge models across various contexts and use cases.

  • 10 authors
·
Jan 14, 2025 2

NMIXX: Domain-Adapted Neural Embeddings for Cross-Lingual eXploration of Finance

General-purpose sentence embedding models often struggle to capture specialized financial semantics, especially in low-resource languages like Korean, due to domain-specific jargon, temporal meaning shifts, and misaligned bilingual vocabularies. To address these gaps, we introduce NMIXX (Neural eMbeddings for Cross-lingual eXploration of Finance), a suite of cross-lingual embedding models fine-tuned with 18.8K high-confidence triplets that pair in-domain paraphrases, hard negatives derived from a semantic-shift typology, and exact Korean-English translations. Concurrently, we release KorFinSTS, a 1,921-pair Korean financial STS benchmark spanning news, disclosures, research reports, and regulations, designed to expose nuances that general benchmarks miss. When evaluated against seven open-license baselines, NMIXX's multilingual bge-m3 variant achieves Spearman's rho gains of +0.10 on English FinSTS and +0.22 on KorFinSTS, outperforming its pre-adaptation checkpoint and surpassing other models by the largest margin, while revealing a modest trade-off in general STS performance. Our analysis further shows that models with richer Korean token coverage adapt more effectively, underscoring the importance of tokenizer design in low-resource, cross-lingual settings. By making both models and the benchmark publicly available, we provide the community with robust tools for domain-adapted, multilingual representation learning in finance.

  • 7 authors
·
Jul 13, 2025

GradSign: Model Performance Inference with Theoretical Insights

A key challenge in neural architecture search (NAS) is quickly inferring the predictive performance of a broad spectrum of networks to discover statistically accurate and computationally efficient ones. We refer to this task as model performance inference (MPI). The current practice for efficient MPI is gradient-based methods that leverage the gradients of a network at initialization to infer its performance. However, existing gradient-based methods rely only on heuristic metrics and lack the necessary theoretical foundations to consolidate their designs. We propose GradSign, an accurate, simple, and flexible metric for model performance inference with theoretical insights. The key idea behind GradSign is a quantity {\Psi} to analyze the optimization landscape of different networks at the granularity of individual training samples. Theoretically, we show that both the network's training and true population losses are proportionally upper-bounded by {\Psi} under reasonable assumptions. In addition, we design GradSign, an accurate and simple approximation of {\Psi} using the gradients of a network evaluated at a random initialization state. Evaluation on seven NAS benchmarks across three training datasets shows that GradSign generalizes well to real-world networks and consistently outperforms state-of-the-art gradient-based methods for MPI evaluated by Spearman's {\rho} and Kendall's Tau. Additionally, we integrate GradSign into four existing NAS algorithms and show that the GradSign-assisted NAS algorithms outperform their vanilla counterparts by improving the accuracies of best-discovered networks by up to 0.3%, 1.1%, and 1.0% on three real-world tasks.

  • 2 authors
·
Oct 16, 2021

PhyGround: Benchmarking Physical Reasoning in Generative World Models

Generative world models are increasingly used for video generation, where learned simulators are expected to capture the physical rules that govern real-world dynamics. However, evaluating whether generated videos actually follow these rules remains challenging. Existing physics-focused video benchmarks have made important progress, but they still face three key challenges, including the coarse evaluation frameworks that hide law-specific failures, response biases and fatigue that undermine the validity of annotation judgments, and automated evaluators that are insufficiently physics-aware or difficult to audit. To address those challenges, we introduce PhyGround, a criteria-grounded benchmark for evaluating physical reasoning in video generation. The benchmark contains 250 curated prompts, each augmented with an expected physical outcome, and a taxonomy of 13 physical laws across solid-body mechanics, fluid dynamics, and optics. Each law is operationalized through observable sub-questions to enable per-law diagnostics. We evaluate eight modern video generation models through a large-scale, quality-controlled human study, grounded on social science lab experiment design. A total of 459 annotators provided 5,796 complete annotations and over 37.4K fine-grained labels; after quality control, the retained annotations exhibited high split-half model-ranking correlations (Spearman's rho > 0.90). To support reproducible automated evaluation, we release PhyJudge-9B, an open physics-specialized VLM judge. PhyJudge-9B achieves substantially lower aggregate relative bias than Gemini-3.1-Pro (3.3% vs. 16.6%). We release prompts, human annotations, model checkpoints, and evaluation code on the project page https://phyground.github.io/.

Prismatic Synthesis: Gradient-based Data Diversification Boosts Generalization in LLM Reasoning

Effective generalization in language models depends critically on the diversity of their training data. Yet existing diversity metrics often fall short of this goal, relying on surface-level heuristics that are decoupled from model behavior. This motivates us to ask: What kind of diversity in training data actually drives generalization in language models -- and how can we measure and amplify it? Through large-scale empirical analyses spanning over 300 training runs, carefully controlled for data scale and quality, we show that data diversity can be a strong predictor of generalization in LLM reasoning -- as measured by average model performance on unseen out-of-distribution benchmarks. We introduce G-Vendi, a metric that quantifies diversity via the entropy of model-induced gradients. Despite using a small off-the-shelf proxy model for gradients, G-Vendi consistently outperforms alternative measures, achieving strong correlation (Spearman's rho approx 0.9) with out-of-distribution (OOD) performance on both natural language inference (NLI) and math reasoning tasks. Building on this insight, we present Prismatic Synthesis, a framework for generating diverse synthetic data by targeting underrepresented regions in gradient space. Experimental results show that Prismatic Synthesis consistently improves model performance as we scale synthetic data -- not just on in-distribution test but across unseen, out-of-distribution benchmarks -- significantly outperforming state-of-the-art models that rely on 20 times larger data generator than ours. For example, PrismMath-7B, our model distilled from a 32B LLM, outperforms R1-Distill-Qwen-7B -- the same base model trained on proprietary data generated by 671B R1 -- on 6 out of 7 challenging benchmarks.

  • 10 authors
·
May 26, 2025

PerSEval: Assessing Personalization in Text Summarizers

Personalized summarization models cater to individuals' subjective understanding of saliency, as represented by their reading history and current topics of attention. Existing personalized text summarizers are primarily evaluated based on accuracy measures such as BLEU, ROUGE, and METEOR. However, a recent study argued that accuracy measures are inadequate for evaluating the degree of personalization of these models and proposed EGISES, the first metric to evaluate personalized text summaries. It was suggested that accuracy is a separate aspect and should be evaluated standalone. In this paper, we challenge the necessity of an accuracy leaderboard, suggesting that relying on accuracy-based aggregated results might lead to misleading conclusions. To support this, we delve deeper into EGISES, demonstrating both theoretically and empirically that it measures the degree of responsiveness, a necessary but not sufficient condition for degree-of-personalization. We subsequently propose PerSEval, a novel measure that satisfies the required sufficiency condition. Based on the benchmarking of ten SOTA summarization models on the PENS dataset, we empirically establish that -- (i) PerSEval is reliable w.r.t human-judgment correlation (Pearson's r = 0.73; Spearman's rho = 0.62; Kendall's tau = 0.42), (ii) PerSEval has high rank-stability, (iii) PerSEval as a rank-measure is not entailed by EGISES-based ranking, and (iv) PerSEval can be a standalone rank-measure without the need of any aggregated ranking.

  • 5 authors
·
Jun 29, 2024

Establishing Trustworthy LLM Evaluation via Shortcut Neuron Analysis

The development of large language models (LLMs) depends on trustworthy evaluation. However, most current evaluations rely on public benchmarks, which are prone to data contamination issues that significantly compromise fairness. Previous researches have focused on constructing dynamic benchmarks to address contamination. However, continuously building new benchmarks is costly and cyclical. In this work, we aim to tackle contamination by analyzing the mechanisms of contaminated models themselves. Through our experiments, we discover that the overestimation of contaminated models is likely due to parameters acquiring shortcut solutions in training. We further propose a novel method for identifying shortcut neurons through comparative and causal analysis. Building on this, we introduce an evaluation method called shortcut neuron patching to suppress shortcut neurons. Experiments validate the effectiveness of our approach in mitigating contamination. Additionally, our evaluation results exhibit a strong linear correlation with MixEval, a recently released trustworthy benchmark, achieving a Spearman coefficient (rho) exceeding 0.95. This high correlation indicates that our method closely reveals true capabilities of the models and is trustworthy. We conduct further experiments to demonstrate the generalizability of our method across various benchmarks and hyperparameter settings. Code: https://github.com/GaryStack/Trustworthy-Evaluation

  • 6 authors
·
Jun 4, 2025 2

Where Does the Audio Jailbreak Live? A Controlled Frequency-Depth Audit of AdvWave-P on Qwen2-Audio

We audit frequency and decoder-depth claims for AdvWave-P, an additive audio jailbreak, on Qwen2-Audio. The protocol masks frequency components of the perturbation in the short-time Fourier transform (STFT) domain and measures attack success and audio-span representations. On 520 AdvBench prompts, the primary judge labels 76.7% of adversarial inputs as jailbreaks. A condition-blind, single-annotator validation yields a Rogan-Gladen sensitivity estimate of 0.87 for this condition (about 0.83-0.95 with validation-rate uncertainty); this correction is not applied to masked conditions. The apparent frequency ranking depends on the partition: energy share alone predicts the standard eight-band ranking (Spearman's rho = 0.95), and equal-Hz and equal-energy partitions show that masking any tested band can sharply reduce attack success. At a finer 16-band equal-energy resolution, however, masking the narrow 7520-7960 Hz band leaves ASR at 0.10, which remains unresolved without a matched control. Matched-energy scattered-removal tests show that contiguous removal is more damaging in lower bands, while both forms approach the floor in upper bands. Global rescaling leaves ASR near baseline but tests amplitude sensitivity rather than frequency location. In a re-optimization pilot (n = 20), tested single- and two-band supports reach ASRs of 0.00, 0.25, and 0.40, while random supports covering about half the STFT bins reach a mean of 0.86. In a prompt- and energy-adjusted model, audio-span divergence is associated with band necessity, with the coefficient rising from +0.63 at the projector output to +0.93 at layer 30 (contrast +0.293, 95% CI [0.11, 0.51]). This association is not a causal localization, and single-layer patching does not establish a causal layer. The results support partition-aware auditing of frequency claims, leaving the fine-resolution top-band result and broader generality open.

  • 7 authors
·
Oct 3

FilmBench: A Film-Grade Benchmark for Cinematic Video Generation

Progress in video generation keeps narrowing the visual gap between AI-generated and professionally produced footage, yet most benchmarks still draw prompts from web sources or LLM templates and score them with untrained, generic multimodal models. More fundamentally, their evaluation taxonomies remain rudimentary (overall visual quality, coarse text alignment and temporal smoothness) rather than the professional Cinematic Language criteria by which films are actually made and judged, so they assess basic video plausibility rather than film-grade craft. We introduce FilmBench, a text-to-video (T2V) and reference-to-video (R2V) benchmark grounded in the professional Cinematic Language of the film- academy tradition and co-developed with directors and faculty from the Beijing Film Academy and the Hujing Digital Media & Entertainment Group film studio. It rests on three choices. First, prompts are reverse-engineered from clips of award-winning films spanning 20 cinematic genres and chosen by professional directors, so every prompt is anchored to a verified live-action reference; the prompts follow real shot lists, and most script multiple shots (1,056 of the 1,169 prompts are multi-shot), unlike prior single-clip benchmarks. Second, evaluation follows a three-level Cinematic taxonomy of 3 axes, 12 components and 35 (T2V) +3 (R2V-only) sub-metrics. Third, we develop an in-house expert-grade automatic evaluation agent and open-source its core suite of Cinematic Language operators (FilmOps). Benchmarking leading video generation models (9 for T2V, 7 for R2V), the evaluator reproduces the human model ranking at model-level Spearman ho = 0.95 (T2V) and 0.96 (R2V). Scores fall well below prior web-style benchmarks, with two consistent gaps in dynamic aesthetics and a marked single- to multi-shot performance drop that widens for weaker models.

  • 30 authors
·
Jul 26 1