Daily Papers

Vietnam's ethnic minority languages are almost absent from the field of Natural Language Processing (NLP), and the challenge goes beyond data scarcity: Cham, Khmer, and Tay-Nung differ sharply in script, Vietnamese contact, and standardization, conditions under which standard multilingual adaptation can learn the wrong signals. We introduce CKTN, the first corpus and benchmark for these languages (44,367 documents, 24M subword tokens), spanning continued pretraining, category classification, and summary-document retrieval. We show that existing multilingual encoders severely fragment these languages, and that common adaptation metrics can mislead: models may lower language-modeling loss or excel at lexical-overlap retrieval while still failing at semantic generalization across documents. We address this with a script-aware adaptation recipe - vocabulary augmentation combined with calibrated replaced-token pretraining - that prevents the discriminator from exploiting trivial script mismatches. The result is an encoder with substantially less fragmentation and the strongest classification performance among evaluated models, exposing the limits of lexical-overlap retrieval as an evaluation signal.

We study English-to-Prakrit machine translation in a low-resource setting where the target language is unsupported by IndicTrans2. We adapt the multilingual model by mapping Prakrit to the Hindi language tag (hin_Deva) without modifying the tokenizer, vocabulary, or architecture. Using a 1,474-pair Maharashtri Prakrit parallel corpus and evaluation on a 20-sample Ardhamagadhi test set, we report corpus BLEU improvements over an untuned baseline. The results indicate that script-compatible language routing can enable feasible transfer to unsupported classical languages, while highlighting limitations due to data scarcity and dialect mismatch. Our code and trained models are released to the public for further exploration https://github.com/D3v1s0m/indictrans2-prakrit-mt.

We consider whether off-policy experience from a smaller, weaker model can elicit capability in a stronger learner that on-policy RL fine-tuning (e.g., GRPO) does not reach. We find that injecting mathematically wrong drafts from a smaller but more domain-trained model -- mismatched to the current problem -- into a stronger learner's GRPO context consistently outperforms standard on-policy GRPO on held-out MATH-500 and out-of-distribution AIME 2025/2026. Concretely, we use Mathstral-7B as the learner, Qwen2.5-Math-1.5B as the draft model, 8.8K Level 3--5 MATH problems (with MATH-500 held out), and train with Dr. GRPO. Mismatch is an active ingredient: shuffling drafts to mismatched problems while holding everything else constant yields +1.62pp on MATH-500 (greedy pass@1) over the matched-wrong variant (n=10 seeds, p=0.0015, Welch's t). In fact, the mismatched-wrong variant leads all other variants we tested on MATH-500 across both greedy pass@1 and sampling pass@k. On out-of-distribution AIME 2025 and 2026, the mismatched-wrong variant uniquely lifts pass@k above both Mathstral-7B (in its native [INST] format) and the Qwen2.5-Math-1.5B draft model at every sample budget from k=1 to k=1024 across 2 seeds (+14.2pp on 2025 and +9.0pp on 2026 at pass@1024 over Mathstral-7B), and at pass@1024 also leads no-draft, matched-wrong, and mismatched-correct variants on both years. All variants use the same prompt with no draft injection at test time. The recipe -- trained on a single GPU with no SFT, no reward models, no synthesized data, and no produce-critique-revise inner loop -- reaches 71.98% MATH-500 on Mathstral-7B-v0.1, the highest published result on this model to our knowledge, surpassing the heavier WizardMath pipeline at 70.9% on full MATH (SFT + PPO with process/instruction reward models).

Executable Agent Skills combine natural-language instructions and scripts into reusable packages for LLM agents, and revising them requires fixing errors without breaking correct behavior. Existing benchmarks do not systematically distinguish documentation repair, script repair, and preservation when evaluating skill self-evolution. We introduce SkillScriptBench, a 350-task benchmark designed to evaluate these capabilities separately. From a survey of over 35,000 GitHub-hosted Skill roots, we select 100 packages and construct 150 repair tasks. Each task pairs a package containing injected script faults with a maintenance request and executable checks of the required behavior. A complementary controlled track contains 200 tasks from 50 packages, each evaluated under the same maintenance request in four states: clean, documentation faults, script faults, and faults in both. Across four LLMs, methods that edit both documentation and scripts can repair script faults but do not consistently outperform Markdown-only revision on documentation repair or preservation. We therefore introduce AST-Guided Skill Revision, which uses abstract syntax trees and calling relationships to link maintenance requirements to relevant code locations. It restricts script edits to these locations and updates the documentation to match the revised scripts. Averaged across models, this revision stage yields absolute gains in repair success of 21.9% for Raw Package and 27.7% for CoEvoSkills on faulty packages. Absolute gains in the proportion of tasks solved in all three runs reach 20.8% and 31.5%, respectively, indicating more consistent repair success across repeated runs.

Automatic speech recognition (ASR) in non-English clinical settings is challenged by multiscript variability, where the same term may appear in multiple valid orthographic forms. Conventional string-matching evaluation metrics often underestimate ASR performance by treating orthographic variants as errors. To address this issue, we introduce MultiClin, a clinical ASR benchmark designed to evaluate robustness to multiscript variability. Experiments across diverse ASR models show that multiscript-aware evaluation provides a fairer assessment of recognition quality than conventional single-reference evaluation. We further investigate the impact of script consistency during training and find that inconsistent script mappings increase orthographic uncertainty and hinder model convergence, with a balanced 50% mapping ratio producing the highest entropy. In contrast, script unification consistently yields the best ASR performance. Our dataset and code are publicly available at: https://github.com/aitrics-ronaldo/Interspeech_MultiClin.

We present GlotScript, an open resource and tool for low resource writing system identification. GlotScript-R is a resource that provides the attested writing systems for more than 7,000 languages. It is compiled by aggregating information from existing writing system resources. GlotScript-T is a writing system identification tool that covers all 161 Unicode 15.0 scripts. For an input text, it returns its script distribution where scripts are identified by ISO 15924 codes. We also present two use cases for GlotScript. First, we demonstrate that GlotScript supports cleaning multilingual corpora such as mC4 and OSCAR. Second, we analyze the tokenization of a number of language models such as GPT-4 using GlotScript and provide insights on the coverage of low resource scripts and languages by each language model. We hope that GlotScript will become a useful resource for work on low resource languages in the NLP community. GlotScript-R and GlotScript-T are available at https://github.com/cisnlp/GlotScript.

Current multilingual evaluations for Vision-Language Models (VLMs) assume a one-to-one mapping between language and orthography, overlooking billions of users of multi-script languages. We introduce PuMVR (Punjabi Multimodal Visual Reasoning), a benchmark of 1,000 strictly parallel image-text instances across Punjabi's three active scripts: Gurmukhi, Shahmukhi, and Roman. Evaluating 10 state-of-the-art VLMs, we expose a substantial and systematic Script Gap. Models frequently solve visual tasks in one script while failing identical tasks in another, with accuracy deltas reaching 16%. Crucially, visual input boosts absolute performance uniformly yet does not close the orthographic gap. Furthermore, cross-script in-context transfer is highly brittle, exposing script-locked knowledge representation. Supported by McNemar tests across all script pairs, our findings demonstrate that current "multilingual" VLMs are not truly multi-script. We propose the Script Consistency Rate (SCR), which falls as low as 24.8% on our benchmark, as a mandatory metric for script-agnostic evaluation to ensure equitable AI access. Data and code are available at: https://github.com/prabhjotschugh/Not-Truly-Multilingual-PuMVR.

Text detection in the wild is a well-known problem that becomes more challenging while handling multiple scripts. In the last decade, some scripts have gained the attention of the research community and achieved good detection performance. However, many scripts are low-resourced for training deep learning-based scene text detectors. It raises a critical question: Is there a need for separate training for new scripts? It is an unexplored query in the field of scene text detection. This paper acknowledges this problem and proposes a solution to detect scripts not present during training. In this work, the analysis has been performed to understand cross-script text detection, i.e., trained on one and tested on another. We found that the identical nature of text annotation (word-level/line-level) is crucial for better cross-script text detection. The different nature of text annotation between scripts degrades cross-script text detection performance. Additionally, for unseen script detection, the proposed solution utilizes vector embedding to map the stroke information of text corresponding to the script category. The proposed method is validated with a well-known multi-lingual scene text dataset under a zero-shot setting. The results show the potential of the proposed method for unseen script detection in natural images.

Word error rate (WER) is the dominant metric for automatic speech recognition, yet it cannot detect a systematic failure mode: models that produce fluent output in the wrong writing system. We define Script Fidelity Rate (SFR), the fraction of hypothesis characters in the target script block, computable without reference transcriptions, and report the first systematic measurement of script collapse across six languages spanning four writing systems (Pashto, Urdu, Hindi, Bengali, Malayalam, Somali) and nine ASR models on FLEURS test sets. Across 53 evaluated model-language pairs, 18 (34%; 95% Wilson CI: 23-47%) exhibit script collapse (SFR < 10%); MMS-1B and SeamlessM4T-v2 maintain SFR above 99% on every language evaluated, confirming that SFR correctly identifies high fidelity where it is present. We identify three distinct collapse patterns: Latin phonetic substitution (smaller Whisper on Indic languages), Arabic substitution for Somali's Latin-script orthography, and Devanagari substitution where larger Whisper models treat all Indic audio as Hindi, a failure present even in Whisper large-v3.

We present SciCoQA, a dataset for detecting discrepancies between scientific publications and their codebases to ensure faithful implementations. We construct SciCoQA from GitHub issues and reproducibility papers, and to scale our dataset, we propose a synthetic data generation method for constructing paper-code discrepancies. We analyze the paper-code discrepancies in detail and propose discrepancy types and categories to better understand the occurring mismatches. In total, our dataset consists of 611 paper-code discrepancies (81 real, 530 synthetic), spanning diverse computational science disciplines, including AI, Physics, Quantitative Biology, and others. Our evaluation of 21 LLMs highlights the difficulty of SciCoQA, particularly for instances involving omitted paper details, long-context inputs, and data outside the models' pre-training corpus. The best performing model in our evaluation, GPT-5, can only detect 45.7\% of real-world paper-code discrepancies.

This paper describes our submission of the WMT 2020 Shared Task on Sentence Level Direct Assessment, Quality Estimation (QE). In this study, we empirically reveal the mismatching issue when directly adopting BERTScore to QE. Specifically, there exist lots of mismatching errors between the source sentence and translated candidate sentence with token pairwise similarity. In response to this issue, we propose to expose explicit cross-lingual patterns, e.g. word alignments and generation score, to our proposed zero-shot models. Experiments show that our proposed QE model with explicit cross-lingual patterns could alleviate the mismatching issue, thereby improving the performance. Encouragingly, our zero-shot QE method could achieve comparable performance with supervised QE method, and even outperforms the supervised counterpart on 2 out of 6 directions. We expect our work could shed light on the zero-shot QE model improvement.

Large Language Models (LLMs) for code generation evolve rapidly through fine-tuning, merging, or new model releases. However, such updates can introduce regressions, not only in correctness but also in code quality and performance. To address this, we present ReCatcher, a regression testing framework for Python code generation. ReCatcher systematically compares two LLMs, typically a current model and a candidate update, across three dimensions: logical correctness, static code quality, and execution performance. We apply ReCatcher to assess regressions across three update scenarios, fine-tuning, merging, and model release, using CodeLlama, DeepSeek-Coder, and GPT-4o. Our evaluation shows that fine-tuning with cross-language datasets increases syntax errors by up to 12%. Merging with general-purpose models like Llama2 leads to regressions in correctness by up to 18%. GPT-4o introduces regressions of up to 50% in handling missing imports compared to GPT-3.5-turbo, while GPT-4o-mini suffers up to 80% performance degradation in execution time versus GPT-4o. Overall, logical correctness, performance, and error handling (e.g., syntax errors and missing imports) are the most regression-prone areas. Comparing ReCatcher with baseline solutions, it presents better and consistent accuracy across logical and performance aspects. ReCatcher highlights the importance of systematic regression evaluation before adopting new models, while assisting researchers and practitioners in making more informed update decisions.

The world's more than 7000 languages are written in at least 293 scripts. Due to various reasons, many closely related languages use different scripts, which poses a difficulty for multilingual pretrained language models (mPLMs) in learning crosslingual knowledge through lexical overlap. As a consequence, mPLMs are faced with a script barrier: representations from different scripts are located in different subspaces, which can result in crosslingual transfer involving languages of different scripts performing suboptimally. To address this problem, we propose TransliCo, a framework that optimizes the Transliteration Contrastive Modeling (TCM) objective to fine-tune an mPLM by contrasting sentences in its training data and their transliterations in a unified script (in our case Latin), which enhances uniformity in the representation space for different scripts. Using Glot500-m, an mPLM pretrained on over 500 languages, as our source model, we fine-tune it on a small portion (5%) of its training data, and refer to the resulting model as Furina. We show that Furina not only better aligns representations from distinct scripts but also outperforms the original Glot500-m on various zero-shot crosslingual transfer tasks. Additionally, we achieve consistent improvement in a case study on the Indic group where the languages exhibit areal features but use different scripts. We make our code and models publicly available.

With the growing interest in large language models, the need for evaluating the quality of machine text compared to reference (typically human-generated) text has become focal attention. Most recent works focus either on task-specific evaluation metrics or study the properties of machine-generated text captured by the existing metrics. In this work, we propose a new evaluation scheme to model human judgments in 7 NLP tasks, based on the fine-grained mismatches between a pair of texts. Inspired by the recent efforts in several NLP tasks for fine-grained evaluation, we introduce a set of 13 mismatch error types such as spatial/geographic errors, entity errors, etc, to guide the model for better prediction of human judgments. We propose a neural framework for evaluating machine texts that uses these mismatch error types as auxiliary tasks and re-purposes the existing single-number evaluation metrics as additional scalar features, in addition to textual features extracted from the machine and reference texts. Our experiments reveal key insights about the existing metrics via the mismatch errors. We show that the mismatch errors between the sentence pairs on the held-out datasets from 7 NLP tasks align well with the human evaluation.

We present WebFovea, a vision-based web agent that placed 2nd in the WebRetriever Challenge 2026 with a final score of 57.0 out of 100. The challenge evaluates agents end to end on Protocol III of the WebRetriever benchmark (arXiv:2607.06118): starting from an entry URL on a live website, the agent must operate the site's own interface and return a verifiable answer. A capable multimodal large language model (LLM) is necessary for this, but not sufficient. The model's decisions reach the browser through the harness, the code between the model and the page. At every step, four things must go right: the model's reply must be parsed into the intended action, the action must take effect on the page, the result must be reported back accurately, and the model must be shown the information it needs. On real websites, many of the failures we observed occurred at one of these four stages rather than in the model's reasoning. A coordinate-space mismatch placed every click at 3/4 of its intended coordinates; actions on native dropdowns, inside iframes, and in text boxes failed silently; and self-generated chat-template tokens contaminated 4.9% of task episodes. WebFovea hardens each stage and surrounds the loop with guardrails that keep the agent within the rules and its budget. The four-stage view does not depend on the model, although some individual fixes do. Because we used the same model in all four submissions, the rise of our official hidden-set score from 31.0 to 57.0 reflects changes to the harness, up to run-to-run variance on live sites. We describe the design, the evidence for each component (including negative results), a failure analysis, the limitations, and a roadmap that includes routing different steps to different models. Code is available at https://github.com/jianganghan/WebFovea.

The performance of Language Models (LMs) on low-resource, morphologically rich languages like Sinhala remains largely unexplored, particularly regarding script variation in digital communication. Sinhala exhibits script duality, with Unicode used in formal contexts and Romanized text dominating social media, while mixed-script usage is common in practice. This paper benchmarks 24 open-source LMs on Unicode, Romanized and mixed-script Sinhala using perplexity evaluation across diverse text sources. Results reveal substantial script sensitivity, with median performance degradation exceeding 300 times from Unicode to Romanized text. Critically, model size shows no correlation with script-handling competence, as smaller models often outperform architectures 28 times larger. Unicode performance strongly predicts mixed-script robustness but not Romanized capability, demonstrating that single-script evaluation substantially underestimates real-world deployment challenges. These findings establish baseline LM capabilities for Sinhala and provide practical guidance for model selection in multi-script low-resource environments.

The issue localization task aims to identify the locations in a software repository that requires modification given a natural language issue description. This task is fundamental yet challenging in automated software engineering due to the semantic gap between issue description and source code implementation. This gap manifests as two mismatches:(1) symptom-to-cause mismatches, where descriptions do not explicitly reveal underlying root causes; (2) one-to-many mismatches, where a single issue corresponds to multiple interdependent code entities. To address these two mismatches, we propose GraphLocator, an approach that mitigates symptom-to-cause mismatches through causal structure discovering and resolves one-to-many mismatches via dynamic issue disentangling. The key artifact is the causal issue graph (CIG), in which vertices represent discovered sub-issues along with their associated code entities, and edges encode the causal dependencies between them. The workflow of GraphLocator consists of two phases: symptom vertices locating and dynamic CIG discovering; it first identifies symptom locations on the repository graph, then dynamically expands the CIG by iteratively reasoning over neighboring vertices. Experiments on three real-world datasets demonstrates the effectiveness of GraphLocator: (1) Compared with baselines, GraphLocator achieves more accurate localization with average improvements of +19.49% in function-level recall and +11.89% in precision. (2) GraphLocator outperforms baselines on both symptom-to-cause and one-to-many mismatch scenarios, achieving recall improvement of +16.44% and +19.18%, precision improvement of +7.78% and +13.23%, respectively. (3) The CIG generated by GraphLocator yields the highest relative improvement, resulting in a 28.74% increase in performance on downstream resolving task.

Current Vision-Language Models (VLMs) are celebrated for their multilingual capabilities, yet they operate under a flawed assumption: that one language corresponds to a single writing system. This overlooks billions of users of multi-script languages like Punjabi, Serbian, Hindi-Urdu, Kurdish, among many others, for whom a model's capability may be fractured by orthographic bias. We introduce PuMVR (Punjabi Multimodal Visual Reasoning), the first benchmark designed to quantify script-dependent bias through 375 culturally grounded image-reasoning tasks across Punjabi's three active scripts (Gurmukhi, Shahmukhi, Roman). Evaluating 10 state-of-the-art VLMs, we expose a substantial Script Gap: models frequently solve visual puzzles in one script while failing identical tasks in another, with accuracy deltas reaching 16% and Script Consistency Rates (SCR) as low as 24.8%. Crucially, visual input boosts absolute performance but does not close this gap, the relative bias persists. Our analysis suggests reasoning patterns show limited cross-script transferability, and Chain-of-Thought pathways diverge based on script alone. We propose SCR as a core metric for script-agnostic evaluation, challenging current multilingual assessment paradigms and providing a framework for equitable AI.

Context: Code reviews are crucial for software quality. Recent AI advances have allowed large language models (LLMs) to review and fix code; now, there are tools that perform these reviews. However, their reliability and accuracy have not yet been systematically evaluated. Objective: This study compares different LLMs' performance in detecting code correctness and suggesting improvements. Method: We tested GPT4o and Gemini 2.0 Flash on 492 AI generated code blocks of varying correctness, along with 164 canonical code blocks from the HumanEval benchmark. To simulate the code review task objectively, we expected LLMs to assess code correctness and improve the code if needed. We ran experiments with different configurations and reported on the results. Results: With problem descriptions, GPT4o and Gemini 2.0 Flash correctly classified code correctness 68.50% and 63.89% of the time, respectively, and corrected the code 67.83% and 54.26% of the time for the 492 code blocks of varying correctness. Without problem descriptions, performance declined. The results for the 164 canonical code blocks differed, suggesting that performance depends on the type of code. Conclusion: LLM code reviews can help suggest improvements and assess correctness, but there is a risk of faulty outputs. We propose a process that involves humans, called the "Human in the loop LLM Code Review" to promote knowledge sharing while mitigating the risk of faulty outputs.

Automatically generating formal specifications could reduce the effort needed to improve program correctness, but in practice, this is still challenging. Many developers avoid writing contracts by hand, which limits the use of automated verification tools. Recent large language models (LLMs) can generate specifications from code, but these specifications often fail in terms of verification. The reason is syntax errors, overly strict constraints, or mismatches with program behavior. We present SpecPylot, a Python tool that synthesizes executable specifications for Python programs as icontract annotations and checks them using crosshair's symbolic execution. The tool relies on LLMs to propose candidate contracts and uses crosshair to validate them. When crosshair finds a concrete counterexample, SpecPylot updates only the generated contracts and leaves the program itself untouched. In addition, the tool can produce coverage-driven pytest stubs and keep detailed execution artifacts that are useful during debugging. Overall, the evaluation indicates that SpecPylot is able to generate crosshair-compatible contracts for most programs, but it also highlights the practical limits introduced by bounded symbolic exploration and differences in LLM behavior.

Quranic text is distributed in two orthographic forms that are byte-level distinct: the Uthmani script used in every printed mushaf, and the Standard (Imla'i) Arabic form that every mainstream Arabic NLP tool is built for. The gap is concentrated in one Unicode character, U+0670 (superscript alef), which appears in some of the most frequently recited words in the Quran and is silently mishandled by general-purpose Arabic normalizers. We release a 2,290-pair, corpus-aligned Uthmani-to-Standard word mapping constructed by aligning the complete 6,236-verse Quran across both orthographic forms, together with a seven-step text normalization pipeline built on it. Normalizing both forms of all 6,236 verses through that pipeline yields identical strings for 90.9% of verses, and we characterize the residual divergence rather than assert that it is closed. On top of the normalized text, we build a deterministic, LLM-free Quranic recitation validator using a four-layer verse-matching search (exact, morphological, relaxed, fuzzy) and word-error-rate-graded feedback across five severity tiers. The validator scores 98.4% (122/124) on a 124-case suite emitted by the released test harness, and both failures share one mechanism: a single substitution error can make a different verse an exact match. A full-corpus census additionally quantifies an inherent text-only ambiguity affecting 16.5% of verses, and on 34 recitation transcripts drawn from a deployed Arabic ASR system the validator identifies the correct verse in every case. We release the mapping, the script that builds it, the validator, and the evaluation harness under open licenses; every number in this paper except the deployment measurement, whose transcripts are not ours to publish, is reproduced by running them.

Byte Pair Encoding (BPE) tokenizers, widely used in Large Language Models, face challenges in multilingual settings, including penalization of non-Western scripts and the creation of tokens with partial UTF-8 sequences. Pretokenization, often reliant on complex regular expressions, can also introduce fragility and unexpected edge cases. We propose SCRIPT (Script Category Representation in PreTokenization), a novel encoding scheme that bypasses UTF-8 byte conversion by using initial tokens based on Unicode script and category properties. This approach enables a simple, rule-based pretokenization strategy that respects script boundaries, offering a robust alternative to pretokenization strategies based on regular expressions. We also introduce and validate a constrained BPE merging strategy that enforces character integrity, applicable to both SCRIPT-BPE and byte-based BPE. Our experiments demonstrate that SCRIPT-BPE achieves competitive compression while eliminating encoding-based penalties for non-Latin-script languages.

LLM coding agents issue Bash commands through interfaces that may serialize, wrap, and reparse model output. Matched execution scores alone cannot distinguish command-generation errors from failures introduced after generation. QuoteBench measures this boundary with exact final-state validation on 56 one-shot tasks from 14 incident-derived families, crossing the generation contract with the execution transport around one deliberately unescaped added parser. Escaping at the interpolation point reproduces each replayed reply's raw-path outcome, so any recovery under a disclosed boundary must come from the model changing its generation. Across eight same-window configurations, replaying the same reply through the added parser lowers success by 55.4 to 73.2 percentage points; disclosure recovers 30.4 to 60.7 points for six configurations, and zero or slightly negative for the other two. Raw generation is nearly saturated at the frontier; boundary adaptation is what still separates models. GPT-5.6-sol's matched gap of -3.6 points hides -64.3 points of damage and +60.7 points of compensation. The deployment configuration reorders models: one reversal among 26 comparable pairs is unambiguous and four more sit on single-task margins. Evaluations of command-issuing agents should report the model configuration, generation contract, execution path, operating point, and final-state validator rather than treat a matched score as an intrinsic model property.

Detailed profession-specific system prompts raise token use and estimated cost per response without a consistent accuracy gain. We evaluate Scientific Agents, an open-source corpus of 503 profession-specific AGENTS.md profiles, with Gemini 3.8 Flash via OpenRouter in the Pi agent harness. We compare matched profiles with four controls: a minimal baseline ("You are a helpful assistant"), the profile's opening role sentence, a generic scientific rigor guide, and a profile from an unrelated domain. Across nine text-based science benchmarks (4,531 sampled questions, 100 matched profiles), 4,488 items completed all five conditions after API-error retries, scored with automated, rule-based grading. The average profile-baseline accuracy difference is -0.6 percentage points (95% bootstrap interval [-1.5, +0.2] across fixed tasks), and no benchmark shows a statistically clear improvement. Matched profiles produced 1.5-2.3 times as many output tokens and cost 2.2-4.5 times more per successful call. On 60 tool-using BioMysteryBench bioinformatics problems (three runs each for baseline and profile), mean solve rates were 46.7% with the profile and 56.7% at baseline, a difference of -10.0 percentage points (95% interval [-16.7, -3.3]) driven by more frequent token- and time-limit stops under the profile. Longer prompts had one unexpected operational advantage: on SuperGPQA, frequent provider API drops left the short baseline with a correct first-pass answer on only 54.0% of items, against 71.6% with the profile. Generic and mismatched prompts were about as reliable, so this gain comes from prompt length or formatting rather than domain expertise. For the tested model and tasks, loading full profession profiles by default does not improve accuracy and costs considerably more; whether selective retrieval of profile sections or open-ended scientific tasks would change this remains to be tested.

On-policy distillation (OPD) trains a student language model on its own responses, guided by a fixed teacher. Generating these responses is expensive, so systems use quantized inference and asynchronous rollouts to reduce the cost. Both, however, introduce training-inference mismatch between the rollout engine and the model being trained. Such mismatch can destabilize reinforcement learning (RL), but its effects on OPD remain unclear. This paper investigates how much mismatch OPD tolerates, where its effects arise, and how training systems can use this tolerance. Measuring all mismatch sources on one scale, we examine quantization across three student-teacher pairs, with additional tests of weight staleness and injected noise on one pair. OPD tolerates substantial mismatch without meaningful loss of training quality but degrades at higher levels, revealing an empirical mismatch budget. Both the quality cost of mismatch and the mismatch introduced by quantization depend on the model, with some layers particularly sensitive to reduced precision. In controlled comparisons, OPD remains stable under rollout configurations that destabilize RL methods. Under weight noise, mismatch also shifts the prefixes reached during generation, introducing errors that per-token corrections cannot address. To put these findings into practice, we introduce a lightweight admission check that screens rollout configurations against a calibrated budget without requiring a full training run for each candidate. Model-specific FP8 recipes preserve training quality while providing capacity for at least 1.93times as many KV-cache tokens as BF16. Combining FP8 with asynchronous rollouts reaches the same training quality 1.62times sooner than synchronous BF16 training. Code is available at https://github.com/benyucong/train-inference-mismatch-opd

Muon has emerged as an efficient alternative to Adam for pretraining, yet remains underused for fine-tuning. A key obstacle is that most open models are pretrained with Adam, and naively switching to Muon for fine-tuning leads to degraded performance due to an optimizer mismatch. We investigate this mismatch through controlled experiments and relate it to the distinct implicit biases of Adam and Muon. We provide evidence that the mismatch disrupts pretrained knowledge, and that this disruption scales with update strength. This leads us to hypothesize that constraining updates should mitigate the mismatch. We validate this with LoRA: across language and vision tasks, LoRA reduces the performance gap between Adam and Muon observed under full fine-tuning. Studies on LoRA rank, catastrophic forgetting, and LoRA variants further confirm that mismatch severity correlates with update strength. These results shed light on how optimizer mismatch affects fine-tuning and how it can be mitigated. Our code is available at https://github.com/XingyuQu/muon-finetune.

Modern LLM RL systems separate rollout generation from policy optimization. These two stages are expected to produce token probabilities that match exactly. However, implementation differences can make them assign different values to the same sequence under the same model weights, inducing Training-Inference Mismatch (TIM). TIM is difficult to inspect because it is entangled with off-policy drift and common stabilization mechanisms. In this work, we isolate TIM in a zero-mismatch diagnostic setting (VeXact), and show that small token-level numerical disagreements can independently cause training collapse. We further show that TIM changes the effective optimization problem, and identify a set of remedies that could mitigate TIM. Our results suggest that TIM is not benign numerical noise, but a systems-level perturbation that should be treated as a first-order factor in analyzing LLM RL stability.

On-policy reinforcement learning for large language models assumes that training responses are sampled from the current policy. In practice, an inference engine samples the responses and a separate trainer computes the gradients, and the training-inference mismatch between the two makes the training data off-policy. The mismatch has two sources, the numerics of the engine and the stale weights of asynchronous sampling, and a correction that stabilizes training under one source can fail under the other. We ask which property of a mismatch source decides the correction it needs. Evaluating the trainer at the sampler's weights splits each sampled-token error exactly into a numerical discrepancy of the engine and a policy lag of the trainer, and we follow both through training. The numerical discrepancy stays approximately constant, so it biases the updates by a nearly fixed amount, and centering the score at each prefix removes its leading bias. The policy lag is produced by the trainer's own updates: the bias it leaves enters the next update, which enlarges the next lag and closes a feedback loop. Importance weighting cuts this loop only while the lag ratio stays concentrated, and on Qwen3-1.7B it fails at the step where the effective sample size of the ratio falls below one half. From this account we derive the discrepancy-lag correction (DLC), which centers the numerical discrepancy, clips the lag ratio against the sampler's weights and resynchronizes the sampler when this effective sample size falls below one half. Under a stale sampler, DLC keeps the KL divergence from on-policy training within twice the KL between two independent on-policy trainings on Qwen3-0.6B and Qwen3-1.7B, and its resynchronization prevents collapses that the per-token corrections leave and that a KL measured on the sampler's prefixes does not detect.

CONTEXT The R programming language has a huge and active community, especially in the area of statistical computing. Its interpreted nature allows for several interesting constructs, like the manipulation of functions at run-time, that hinder the static analysis of R programs. At the same time, there is a lack of existing research regarding how these features, or even the R language as a whole are used in practice. OBJECTIVE In this paper, we conduct a large-scale, static analysis of more than 50 million lines of real-world R programs and packages to identify their characteristics and the features that are actually used. Moreover, we compare the similarities and differences between the scripts of R users and the implementations of package authors. We provide insights for static analysis tools like the lintr package as well as potential interpreter optimizations and uncover areas for future research. METHOD We analyze 4230 R scripts submitted alongside publications and the sources of 19450 CRAN packages for over 350000 R files, collecting and summarizing quantitative information for features of interest. RESULTS We find a high frequency of name-based indexing operations, assignments, and loops, but a low frequency for most of R's reflective functions. Furthermore, we find neither testing functions nor many calls to R's foreign function interface (FFI) in the publication submissions. CONCLUSION R scripts and package sources differ, for example, in their size, the way they include other packages, and their usage of R's reflective capabilities. We provide features that are used frequently and should be prioritized by static analysis tools, like operator assignments, function calls, and certain reflective functions like load.

After a runtime or API migration, an LLM-repaired procedure can run cleanly yet return wrong results, and a release check on typical test inputs ships it. We report a preregistered study of 73 tasks covering 60 documented changes in CPython, Node.js and SQLite. Every candidate is executed on the real runtimes against withheld contracts. Over the complete repair space, the all-assertions gate on typical inputs released 0.410 of wrong repairs. One edge input written from the release note cut this to 0.065, and to 0.061 on 15 unseen clusters. Three inputs from a blind GPT-5.6-luna writer that never saw the program reached 0.070. Two further frozen stages enlarged the menu to 1491 fragments and then added 233 hard distractors that match the legacy output on typical inputs. The edge input (0.058) and the frontier-model writer (0.058) passed their preregistered tests. Yet 0.168 and 0.153 of the hard wrong programs still passed. Two local-model writers failed their preregistered tests. Across four model families, the edge input cut wrong releases per attempt in all three open-weight families on all three menus, for example from 0.233 to 0.057, without lowering correct completion. Self-tests helped less. A blind AI audit flagged 34 contracts as stricter than their requirements. Dropping the disputed checks changed 0 verdicts, but class-label disputes remain open. No model-written code was executed. Artifacts are available from the authors on request.

Professionals in academia, law, and finance audit their documents because inconsistencies can result in monetary, reputational, and scientific costs. Language models (LMs) have the potential to dramatically speed up this auditing process. To understand their abilities, we introduce a benchmark, FIND (Finding INconsistencies in Documents), where each example is a document with an inconsistency inserted manually by a domain expert. Despite the documents being long, technical, and complex, the best-performing model (gpt-5) recovered 64% of the inserted inconsistencies. Surprisingly, gpt-5 also found undiscovered inconsistencies present in the original documents. For example, on 50 arXiv papers, we judged 136 out of 196 of the model's suggestions to be legitimate inconsistencies missed by the original authors. However, despite these findings, even the best models miss almost half of the inconsistencies in FIND, demonstrating that inconsistency detection is still a challenging task.

Scientific coding agents receive equations, boundary conditions, and output requirements in writing, then must assess the programs they revise. Rules to Tools (R2T) supplies prepared executable checks of public scientific requirements. Matched SciCode repair groups share written checks, starting programs, model, and budgets; the tool group receives a callable implementation. Across two task-ID cohorts, complete repair is 26/30 with text and 29/30 with the prepared checks. Three task IDs favor tools, one favors text, and eleven tie. The eight-ID cohort scores 13/16 versus 15/16, with a task-cluster bootstrap 95% interval of [-12.5, 43.75] percentage points for the difference. The larger shared-definition SciCode cohort ties at 13/24 per group. Five development-exposed tasks with alternate starting programs score 3/10 versus 7/10. The tool group favors tasks 17, 77, and 11; initial checks flag task 17 and report no violation for tasks 77 and 11. Task 37 favors text and has no initial reported violation. A fresh source-through-Python arm also reaches 15/16, matching the dedicated command's aggregate. In a matched PDE comparison, detailed text scores 23/24 and checks score 24/24, with 31.2% lower reported model output for checks. Agent-side output savings vary by cohort, while public CPU use rises in both task-ID cohorts. These results measure task-dependent repair outcomes and agent-side costs with prepared checks.

Choosing an agent system means choosing both a language model and the harness through which it acts. We ask whether a strong model, harness, or pairing stays strong when the setting changes. We evaluate 66 configurations: four configurable harnesses (OpenHands, DeepSeek Harness, PI, and openJiuwen) paired with five models on TUA-Bench, ALE-CLI, and Terminal-Bench 4, plus the native Codex-GPT and Claude Code-Claude pairings. Model rankings reverse across harnesses. On Terminal-Bench 4, Claude leads GPT by 7.94 points in OpenHands but trails it by 30.16 points in PI. For four of the five models, the best harness changes from one benchmark to another, yet some pairings hold: openJiuwen gives Kimi its highest score on all three benchmarks, by 5.61 to 11.11 points. A model's own vendor harness is not reliably its best, and higher cost does not reliably buy a higher score. On Terminal-Bench 4, GPT scores higher under PI than under DSH at less than a quarter of the cost per task. Matched trajectories suggest why fit varies. Models start almost all repairs themselves, so much depends on whether the harness hands failures back in a form the model can use. GPT does best with PI's lean scaffold, while Kimi, which often issues malformed tool calls, does best in openJiuwen. We argue that the model, the harness, and the task should be evaluated together, and we release the harness adapters, evaluation code, and all 6,204 scored trajectories at https://github.com/liyix/finding-the-right-fit and https://huggingface.co/datasets/yixuanli97/finding-the-right-fit.

Large language models (LLMs) now write code in settings where misreading a single word can break safety or cost money, yet we still expect them to overlook stray typos. To probe where useful robustness ends and harmful insensitivity begins, we compile 50 LeetCode problems and craft three minimal prompt perturbations that should vary in importance: (i) progressive underspecification deleting 10 % of words per step; (ii) lexical flip swapping a pivotal quantifier ("max" to "min"); and (iii) jargon inflation replacing a common noun with an obscure technical synonym. Six frontier models, including three "reasoning-tuned" versions, solve each mutated prompt, and their Python outputs are checked against the original test suites to reveal whether they reused the baseline solution or adapted. Among 11 853 generations we observe a sharp double asymmetry. Models remain correct in 85 % of cases even after 90 % of the prompt is missing, showing over-robustness to underspecification, yet only 54 % react to a single quantifier flip that reverses the task, with reasoning-tuned variants even less sensitive than their bases. Jargon edits lie in between, passing through 56 %. Current LLMs thus blur the line between harmless noise and meaning - changing edits, often treating both as ignorable. Masking salient anchors such as function names can force re - evaluation. We advocate evaluation and training protocols that reward differential sensitivity: stay steady under benign noise but adapt - or refuse - when semantics truly change.

Script identification plays a vital role in applications that involve handwriting and document analysis within a multi-script and multi-lingual environment. Moreover, it exhibits a profound connection with human cognition. This paper provides a new database for benchmarking script identification algorithms, which contains both printed and handwritten documents collected from a wide variety of scripts, such as Arabic, Bengali (Bangla), Gujarati, Gurmukhi, Devanagari, Japanese, Kannada, Malayalam, Oriya, Roman, Tamil, Telugu, and Thai. The dataset consists of 1,135 documents scanned from local newspaper and handwritten letters as well as notes from different native writers. Further, these documents are segmented into lines and words, comprising a total of 13,979 and 86,655 lines and words, respectively, in the dataset. Easy-to-go benchmarks are proposed with handcrafted and deep learning methods. The benchmark includes results at the document, line, and word levels with printed and handwritten documents. Results of script identification independent of the document/line/word level and independent of the printed/handwritten letters are also given. The new multi-lingual database is expected to create new script identifiers, present various challenges, including identifying handwritten and printed samples and serve as a foundation for future research in script identification based on the reported results of the three benchmarks.

Autoformalization translates informal mathematical theorems into code for proof assistants such as Lean. A central challenge is that current evaluation metrics can accept type-correct but misaligned statements or reject correct statements written in a different formulation. Inspired by Pass@k, we propose SA-Pass (*Semantic Alignment Pass*), which tests formal statements using auxiliary statements called *shadows* that characterize the intended statement. A generated statement receives full credit only when it compiles, implies each shadow (forward check), and is implied by their conjunction (backward check). We instantiate SA-Pass in ShadowBench, a Lean 4 full autoformalization benchmark of 178 postgraduate- to research-level problems spanning eight mathematical areas. Claude Code (Opus 4.8) with Numina-Lean-Agent reaches 61.8% compile rate and 11.2% SA-Pass. Across outputs generated by six agentic configurations, SA-Pass achieves 98.8% binary agreement with expert judgments. An early version of ShadowBench served as the benchmark for Track 4 of the ICML 2026 AI4Math Challenge.

Many tasks revolve around editing a document, whether code or text. We formulate the revision similarity problem to unify a wide range of machine learning evaluation problems whose goal is to assess a revision to an existing document. We observe that revisions usually change only a small portion of an existing document, so the existing document and its immediate revisions share a majority of their content. We formulate five adequacy criteria for revision similarity measures, designed to align them with human judgement. We show that popular pairwise measures, like BLEU, fail to meet these criteria, because their scores are dominated by the shared content. They report high similarity between two revisions when humans would assess them as quite different. This is a fundamental flaw we address. We propose a novel static measure, Excision Score (ES), which computes longest common subsequence (LCS) to remove content shared by an existing document with the ground truth and predicted revisions, before comparing only the remaining divergent regions. This is analogous to a surgeon creating a sterile field to focus on the work area. We use approximation to speed the standard cubic LCS computation to quadratic. In code-editing evaluation, where static measures are often used as a cheap proxy for passing tests, we demonstrate that ES surpasses existing measures. When aligned with test execution on HumanEvalFix, ES improves over its nearest competitor, SARI, by 12% Pearson correlation and by >21% over standard measures like BLEU. The key criterion is invariance to shared context; when we perturb HumanEvalFix with increased shared context, ES' improvement over SARI increases to 20% and >30% over standard measures. ES also handles other corner cases that other measures do not, such as correctly aligning moved code blocks, and appropriately rewarding matching insertions or deletions.

A large share of day-to-day work in professional domains happens inside PDF files: benefits packets, leases, datasheets, clinical guidelines, construction plans. Benchmarks for document AI have generally measured the required capabilities in isolation: OCR, layout analysis, chart reasoning, table QA, document VQA. A high score on any one of them does not necessarily reveal whether a model can answer a realistic question that someone in the field would actually ask about a specific PDF. GDP_pdf is a benchmark built to measure this directly. It consists of question-document pairs authored by working professionals in ten fields, and a candidate question was kept only when at least two frontier multimodal models failed it in a way that mattered: a wrong answer, missed decisive evidence, or a fabricated claim, rather than a superficial difference such as style. Each item comes with a rubric of atomic criteria, so we can report a graded rubric score as well as a strict task-level pass rate, and each item is tagged against a taxonomy of eleven capabilities in three tiers, spanning text extraction and grounding, table and chart comprehension, cross-referencing, spatial reasoning, and abstention on unsupported queries. We report results for seventeen frontier models on the 100-item benchmark: the best model passes only 30.7% of the items and the worst passes 2%. Most errors trace back to a small set of recurring loss patterns: misaligned tables, misread charts, skipped footnotes and exclusions, miscounted floor-plan symbols, scan noise, and amendments that supersede earlier text.

"Abliterated" ("uncensored") builds of open-weight language models remove the activation direction that mediates refusal, and are popular partly on the assumption that a model that no longer refuses is also free of political censorship. We test this on two matched pairs of 2-bit models: NVIDIA's Nemotron 3.5 Bolt 74B-A9B and Qwen3.8-Flash-Next, each against a community abliteration. Each model judged 22 statement pairs (on the Republic of China (Taiwan), 1989 Tiananmen, control countries) in Traditional Chinese, Simplified Chinese (a character-level conversion) and English, and answered 27 open questions per script. Script mattered far more than abliteration: the original Bolt scored +0.58 on Index S (-1: People's Republic of China (PRC) position; +1: ROC position) in Traditional Chinese, +0.57 in English and +0.05 in Simplified Chinese, and affirmed "Taiwan is part of the PRC" with probability 0.91 in Simplified but 0.06 in Traditional characters; abliteration changed its scores by at most 0.06, mostly by raising "yes" on both sides. In Qwen, abliteration removed every refusal on 1989 and raised scores by 0.21 to 0.34, yet the abliterated Qwen still affirmed that Taiwan is part of the PRC in all three scripts. Naming the state the Republic of China (Taiwan) raised every model's score by 0.10 to 0.33 but left the script gap. Control pairs showed no script gap; each abliteration changed only its declared tensors. In 740,000 web documents, Simplified text on Taiwan contained PRC-framing terms 3.7 times as often as Traditional text. A script direction could not be steered away; a contrastive stance direction raised scores in every script, and its negation lowered them. A four-sentence system prompt raised the abliterated Qwen to +0.93 to +0.97. Keyword scoring rules failed the models closest to the ROC position mostly for attributing or denying the PRC claim. Uncensored is not unbiased.

Tests that cause spurious failures without any code changes, i.e., flaky tests, hamper regression testing, increase maintenance costs, may shadow real bugs, and decrease trust in tests. While the prevalence and importance of flakiness is well established, prior research focused on Java projects, thus raising the question of how the findings generalize. In order to provide a better understanding of the role of flakiness in software development beyond Java, we empirically study the prevalence, causes, and degree of flakiness within software written in Python, one of the currently most popular programming languages. For this, we sampled 22352 open source projects from the popular PyPI package index, and analyzed their 876186 test cases for flakiness. Our investigation suggests that flakiness is equally prevalent in Python as it is in Java. The reasons, however, are different: Order dependency is a much more dominant problem in Python, causing 59% of the 7571 flaky tests in our dataset. Another 28% were caused by test infrastructure problems, which represent a previously undocumented cause of flakiness. The remaining 13% can mostly be attributed to the use of network and randomness APIs by the projects, which is indicative of the type of software commonly written in Python. Our data also suggests that finding flaky tests requires more runs than are often done in the literature: A 95% confidence that a passing test case is not flaky on average would require 170 reruns.

Existing Multimodal Large Language Models (MLLMs) are predominantly trained and tested on consistent visual-textual inputs, leaving open the question of whether they can handle inconsistencies in real-world, layout-rich content. To bridge this gap, we propose the Multimodal Inconsistency Reasoning (MMIR) benchmark to assess MLLMs' ability to detect and reason about semantic mismatches in artifacts such as webpages, presentation slides, and posters. MMIR comprises 534 challenging samples, each containing synthetically injected errors across five reasoning-heavy categories: Factual Contradiction, Identity Misattribution, Contextual Mismatch, Quantitative Discrepancy, and Temporal/Spatial Incoherence. We evaluate six state-of-the-art MLLMs, showing that models with dedicated multimodal reasoning capabilities, such as o1, substantially outperform their counterparts while open-source models remain particularly vulnerable to inconsistency errors. Detailed error analyses further show that models excel in detecting inconsistencies confined to a single modality, particularly in text, but struggle with cross-modal conflicts and complex layouts. Probing experiments reveal that single-modality prompting, including Chain-of-Thought (CoT) and Set-of-Mark (SoM) methods, yields marginal gains, revealing a key bottleneck in cross-modal reasoning. Our findings highlight the need for advanced multimodal reasoning and point to future research on multimodal inconsistency.

This paper revisits recent code similarity evaluation metrics, particularly focusing on the application of Abstract Syntax Tree (AST) editing distance in diverse programming languages. In particular, we explore the usefulness of these metrics and compare them to traditional sequence similarity metrics. Our experiments showcase the effectiveness of AST editing distance in capturing intricate code structures, revealing a high correlation with established metrics. Furthermore, we explore the strengths and weaknesses of AST editing distance and prompt-based GPT similarity scores in comparison to BLEU score, execution match, and Jaccard Similarity. We propose, optimize, and publish an adaptable metric that demonstrates effectiveness across all tested languages, representing an enhanced version of Tree Similarity of Edit Distance (TSED).

Program synthesis has been long studied with recent approaches focused on directly using the power of Large Language Models (LLMs) to generate code. Programming benchmarks, with curated synthesis problems and test-cases, are used to measure the performance of various LLMs on code synthesis. However, these test-cases can be limited in both quantity and quality for fully assessing the functional correctness of the generated code. Such limitation in the existing benchmarks begs the following question: In the era of LLMs, is the code generated really correct? To answer this, we propose EvalPlus -- a code synthesis evaluation framework to rigorously benchmark the functional correctness of LLM-synthesized code. EvalPlus augments a given evaluation dataset with large amounts of test-cases newly produced by an automatic test input generator, powered by both LLM- and mutation-based strategies. While EvalPlus is general, we extend the test-cases of the popular HumanEval benchmark by 80x to build HumanEval+. Our extensive evaluation across 26 popular LLMs (e.g., GPT-4 and ChatGPT) demonstrates that HumanEval+ is able to catch significant amounts of previously undetected wrong code synthesized by LLMs, reducing the pass@k by up-to 19.3-28.9%. We also surprisingly found that test insufficiency can lead to mis-ranking. For example, both WizardCoder-CodeLlama and Phind-CodeLlama now outperform ChatGPT on HumanEval+, while none of them could on HumanEval. Our work not only indicates that prior popular code synthesis evaluation results do not accurately reflect the true performance of LLMs for code synthesis, but also opens up a new direction to improve such programming benchmarks through automated testing. We have open-sourced our tools, enhanced datasets as well as all LLM-generated code at https://github.com/evalplus/evalplus to facilitate and accelerate future LLM-for-code research.

In this paper, we have shown a script conversion (transliteration) technique that converts Sindhi text in the Devanagari script to the Perso-Arabic script. We showed this by incorporating a hybrid approach where some part of the text is converted using a rule base and in case an ambiguity arises then a probabilistic model is used to resolve the same. Using this approach, the system achieved an overall accuracy of 99.64%.

Large Language Models (LLMs) are increasingly deployed in high-stakes clinical applications in India. In many such settings, speakers of Indian languages frequently communicate using romanized text rather than native scripts, yet existing research rarely evaluates this orthographic variation using real-world data. We investigate how romanization impacts the reliability of LLMs in a critical domain: maternal and newborn healthcare triage. We benchmark leading LLMs on a real-world dataset of user-generated queries spanning five Indian languages and Nepali. Our results reveal consistent degradation in performance for romanized messages, with F1 scores trailing those of native scripts by 5-12 points. At our partner maternal health organization in India, this gap could cause nearly 2 million excess errors in triage. Crucially, this performance gap by scripts is not due to a failure in clinical reasoning. We demonstrate that LLMs often correctly infer the semantic intent of romanized queries. Nevertheless, their final classification outputs remain brittle in the presence of orthographic noise in romanized inputs. Our findings highlight a critical safety blind spot in LLM-based health systems: models that appear to understand romanized input may still fail to act on it reliably.

Optical character recognition (OCR) has advanced rapidly with the rise of vision-language models, yet evaluation has remained concentrated on a small cluster of high- and mid-resource scripts. We introduce GlotOCR Bench, a comprehensive benchmark evaluating OCR generalization across 100+ Unicode scripts. Our benchmark comprises clean and degraded image variants rendered from real multilingual texts. Images are rendered using fonts from the Google Fonts repository, shaped with HarfBuzz and rasterized with FreeType, supporting both LTR and RTL scripts. Samples of rendered images were manually reviewed to verify correct rendering across all scripts. We evaluate a broad suite of open-weight and proprietary vision-language models and find that most perform well on fewer than ten scripts, and even the strongest frontier models fail to generalize beyond thirty scripts. Performance broadly tracks script-level pretraining coverage, suggesting that current OCR systems rely on language model pretraining as much as on visual recognition. Models confronted with unfamiliar scripts either produce random noise or hallucinate characters from similar scripts they already know. We release the benchmark and pipeline for reproducibility. Pipeline Code: https://github.com/cisnlp/glotocr-bench, Benchmark: https://hf.co/datasets/cis-lmu/glotocr-bench.

Sardinian, a Romance language with roughly one million speakers, has minimal presence in modern NLP. Commercial services do not support it, and current language models do not produce it reliably. We present LLiMba, a 3B parameter Sardinian-ready model adapted from Qwen2.5-3B-Instruct through continued pretraining (CPT) and supervised fine-tuning (SFT) on a single 24 GB consumer GPU. The corpus contains 11.5 million tokens of Sardinian spanning LSC, Logudorese, and Campidanese, augmented with 2.4 million tokens of related Romance text as replay against register blurring. After CPT the model reaches a perplexity of 6.76 on held out Sardinian and outperforms the base across all six FLORES-200 directions. We compare five SFT configurations under matched conditions: full fine-tuning, LoRA r64, rsLoRA r128, rsLoRA r256, and DoRA r256. rsLoRA r256 wins on every direction into Sardinian, reaching 28.5 BLEU from English against 17.3 after CPT and 21.0 with full fine-tuning. The rank ablation places r128 between LoRA r64 and rsLoRA r256 on BLEU but reveals failure modes invisible to the metric, including leakage across scripts no other variant produces. LoRA r64 retains less factual content from SFT than configurations at higher rank and produces more confident fabrications, though all methods fabricate on content absent from training. DoRA r256 yields the smallest gap between training and evaluation but the worst factual accuracy. The findings indicate that adapter capacity matters more than the choice among LoRA variants for adapting a Romance pretrained base to a low resource Romance target, that stronger regularization is not uniformly beneficial, and that translation metrics smoothly order configurations whose qualitative behavior differs categorically. Perplexity comparisons across scripts must account for byte fallback tokenization, which deflates the metric for scripts other than Latin.

Multimodal misinformation detection methods often rely on different preprocessing pipelines and auxiliary resources, making experimental comparisons difficult to interpret. We present MMFact, a collection of shared resources and tools that simplifies this research workflow. It brings together 47 detector implementations and baselines, three multimodal datasets, and prepared materials ranging from semantic features to OCR, image descriptions, and external evidence. Compatible methods can use the same materials, reducing variation introduced by preprocessing and supporting fairer comparisons. A modular training and evaluation framework preserves method-specific designs while providing common experimental procedures. By treating prepared materials as a shared research resource, MMFact helps researchers focus on developing and evaluating detection methods.

AI coding agents increasingly submit complete pull requests (PRs) with minimal human intervention, shifting software development from AI-assisted to autonomous workflows. As these agents become more prevalent, ensuring the code they generate is adequately tested, by existing tests or by tests the agents write, is critical to preventing regressions, yet little is known about testing in agentic PRs. To address this gap, we analyze 4882 agent-generated PRs from the AIDev dataset (532 Java and 4350 Python PRs) produced by five coding agents. We study (i) how often agents include test changes and (ii) how well covered are code changes by existing and agent-written tests. Agents include test changes in only 49.6% of PRs that change code under test files. Existing tests provide an incomplete safety net: they cover 61.5% of agents' changed executable lines in Java and only 27.0% in Python, where 64.8% of PRs have no changed line executed by any existing test. Agent-written tests improve coverage over existing tests, but only in a minority of PRs: 35.9% of Java and 22.5% of Python Code + Tests PRs show a coverage gain. Across both languages, error-handling constructs (e.g., try and catch blocks) are the most consistently under-tested, with miss rates reaching 86.0% in Java and 81.0% in Python. These findings motivate coverage-aware development practices, coverage feedback loops for coding agents, and evaluation benchmarks that measure test quality to better help agents reliably test their own code.

Doctors typically write in incomprehensible handwriting, making it difficult for both the general public and some pharmacists to understand the medications they have prescribed. It is not ideal for them to write the prescription quietly and methodically because they will be dealing with dozens of patients every day and will be swamped with work.As a result, their handwriting is illegible. This may result in reports or prescriptions consisting of short forms and cursive writing that a typical person or pharmacist won't be able to read properly, which will cause prescribed medications to be misspelled. However, some individuals are accustomed to writing prescriptions in regional languages because we all live in an area with a diversity of regional languages. It makes analyzing the content much more challenging. So, in this project, we'll use a recognition system to build a tool that can translate the handwriting of physicians in any language. This system will be made into an application which is fully autonomous in functioning. As the user uploads the prescription image the program will pre-process the image by performing image pre-processing, and word segmentations initially before processing the image for training. And it will be done for every language we require the model to detect. And as of the deduction model will be made using deep learning techniques including CNN, RNN, and LSTM, which are utilized to train the model. To match words from various languages that will be written in the system, Unicode will be used. Furthermore, fuzzy search and market basket analysis are employed to offer an end result that will be optimized from the pharmaceutical database and displayed to the user as a structured output.

Recent advancements in Large Language Models (LLMs) have led to their widespread application in automated code generation. However, these models can still generate defective code that deviates from the specification. Previous research has mainly focused on the mistakes in LLM-generated standalone functions, overlooking real-world software development situations where the successful generation of the code requires software contexts such as external dependencies. In this paper, we considered both of these code generation situations and identified a range of non-syntactic mistakes arising from LLMs' misunderstandings of coding question specifications. Seven categories of non-syntactic mistakes were identified through extensive manual analyses, four of which were missed by previous works. To better understand these mistakes, we proposed six reasons behind these mistakes from various perspectives. Moreover, we explored the effectiveness of LLMs in detecting mistakes and their reasons. Our evaluation demonstrated that GPT-4 with the ReAct prompting technique can achieve an F1 score of up to 0.65 when identifying reasons for LLM's mistakes, such as misleading function signatures. We believe that these findings offer valuable insights into enhancing the quality of LLM-generated code.

Models for text-to-image synthesis, such as DALL-E~2 and Stable Diffusion, have recently drawn a lot of interest from academia and the general public. These models are capable of producing high-quality images that depict a variety of concepts and styles when conditioned on textual descriptions. However, these models adopt cultural characteristics associated with specific Unicode scripts from their vast amount of training data, which may not be immediately apparent. We show that by simply inserting single non-Latin characters in a textual description, common models reflect cultural stereotypes and biases in their generated images. We analyze this behavior both qualitatively and quantitatively, and identify a model's text encoder as the root cause of the phenomenon. Additionally, malicious users or service providers may try to intentionally bias the image generation to create racist stereotypes by replacing Latin characters with similarly-looking characters from non-Latin scripts, so-called homoglyphs. To mitigate such unnoticed script attacks, we propose a novel homoglyph unlearning method to fine-tune a text encoder, making it robust against homoglyph manipulations.

Automatic speech recognition (ASR) is a core component of human--computer interaction and an increasingly important front-end for LLM-based assistants and agents. However, most current ASR systems still follow a single-pass paradigm, which is poorly aligned with human communication, where misunderstandings are resolved through iterative clarification and refinement. This mismatch makes it difficult to correct meaning-critical errors once they occur. Meanwhile, token-level metrics such as WER or CER cannot adequately reflect such a problem. To address these limitations, we formulate Interactive ASR as a multi-turn refinement task and propose Agentic ASR, a closed-loop framework that combines a single-pass ASR front-end with semantic correction, intent routing, and reasoning-based editing. We further introduce the Sentence-level Semantic Error Rate (S^2ER), an LLM-based semantic evaluation metric, together with an Interactive Simulation System for scalable and reproducible benchmarking. Experiments on multilingual, named-entity-intensive, and code-switching benchmarks show that iterative interaction consistently reduces semantic errors, with much larger gains in S^2ER than in conventional token-level metrics. Human--AI alignment and ablation studies further validate the reliability of the semantic judge and the robustness of the proposed framework. The code is available at: https://interactiveasr.github.io/ and the live demo is available at https://i-asr.sjtuxlance.com/

This paper explores the pitfalls in evaluating multilingual automatic speech recognition (ASR) models, with a particular focus on Indic language scripts. We investigate the text normalization routine employed by leading ASR models, including OpenAI Whisper, Meta's MMS, Seamless, and Assembly AI's Conformer, and their unintended consequences on performance metrics. Our research reveals that current text normalization practices, while aiming to standardize ASR outputs for fair comparison, by removing inconsistencies such as variations in spelling, punctuation, and special characters, are fundamentally flawed when applied to Indic scripts. Through empirical analysis using text similarity scores and in-depth linguistic examination, we demonstrate that these flaws lead to artificially improved performance metrics for Indic languages. We conclude by proposing a shift towards developing text normalization routines that leverage native linguistic expertise, ensuring more robust and accurate evaluations of multilingual ASR models.

Computer-use agents (CUAs) have made tremendous progress in the past year, yet they still frequently produce misaligned actions that deviate from the user's original intent. Such misaligned actions may arise from external attacks (e.g., indirect prompt injection) or from internal limitations (e.g., erroneous reasoning). They not only expose CUAs to safety risks, but also degrade task efficiency and reliability. This work makes the first effort to define and study misaligned action detection in CUAs, with comprehensive coverage of both externally induced and internally arising misaligned actions. We further identify three common categories in real-world CUA deployment and construct MisActBench, a benchmark of realistic trajectories with human-annotated, action-level alignment labels. Moreover, we propose DeAction, a practical and universal guardrail that detects misaligned actions before execution and iteratively corrects them through structured feedback. DeAction outperforms all existing baselines across offline and online evaluations with moderate latency overhead: (1) On MisActBench, it outperforms baselines by over 15% absolute in F1 score; (2) In online evaluation, it reduces attack success rate by over 90% under adversarial settings while preserving or even improving task success rate in benign environments.

Korean is a morphologically rich language with a featural writing system in which each character is systematically composed of subcharacter units known as Jamo. These subcharacters not only determine the visual structure of Korean but also encode frequent and linguistically meaningful morphophonological processes. However, most current Korean language models (LMs) are based on subword tokenization schemes, which are not explicitly designed to capture the internal compositional structure of characters. To address this limitation, we propose SCRIPT, a model-agnostic module that injects subcharacter compositional knowledge into Korean PLMs. SCRIPT allows to enhance subword embeddings with structural granularity, without requiring architectural changes or additional pre-training. As a result, SCRIPT enhances all baselines across various Korean natural language understanding (NLU) and generation (NLG) tasks. Moreover, beyond performance gains, detailed linguistic analyses show that SCRIPT reshapes the embedding space in a way that better captures grammatical regularities and semantically cohesive variations. Our code is available at https://github.com/SungHo3268/SCRIPT.

With the growing use of large language models(LLMs) as evaluators, their application has expanded to code evaluation tasks, where they assess the correctness of generated code without relying on reference implementations. While this offers scalability and flexibility, it also raises a critical, unresolved question: Can LLM judges fairly and robustly evaluate semantically equivalent code with superficial variations? Functionally correct code often exhibits variations-such as differences in variable names, comments, or formatting-that should not influence its correctness. Yet, whether LLM judges can reliably handle these variations remains unclear. We present the first comprehensive study of this issue, defining six types of potential bias in code evaluation and revealing their systematic impact on LLM judges. Across five programming languages and multiple LLMs, we empirically demonstrate that all tested LLM judges are susceptible to both positive and negative biases, resulting in inflated or unfairly low scores. Moreover, we observe that LLM judges remain vulnerable to these biases even when prompted to generate test cases before scoring, highlighting the need for more robust code evaluation methods.

Software updates, including bug repair and feature additions, are frequent in modern applications but they often leave test suites outdated, resulting in undetected bugs and increased chances of system failures. A recent study by Meta revealed that 14%-22% of software failures stem from outdated tests that fail to reflect changes in the codebase. This highlights the need to keep tests in sync with code changes to ensure software reliability. In this paper, we present UTFix, a novel approach for repairing unit tests when their corresponding focal methods undergo changes. UTFix addresses two critical issues: assertion failure and reduced code coverage caused by changes in the focal method. Our approach leverages language models to repair unit tests by providing contextual information such as static code slices, dynamic code slices, and failure messages. We evaluate UTFix on our generated synthetic benchmarks (Tool-Bench), and real-world benchmarks. Tool- Bench includes diverse changes from popular open-source Python GitHub projects, where UTFix successfully repaired 89.2% of assertion failures and achieved 100% code coverage for 96 tests out of 369 tests. On the real-world benchmarks, UTFix repairs 60% of assertion failures while achieving 100% code coverage for 19 out of 30 unit tests. To the best of our knowledge, this is the first comprehensive study focused on unit test in evolving Python projects. Our contributions include the development of UTFix, the creation of Tool-Bench and real-world benchmarks, and the demonstration of the effectiveness of LLM-based methods in addressing unit test failures due to software evolution.

AI-text detectors are already applied to Kazakh writing, yet no study reports how often publicly available detectors flag Kazakh text whose human authorship is dated before ChatGPT. We present AINA-KK: 3,811 human passages from encyclopedic, news, web and educational pages that provably existed before ChatGPT, and 5,597 machine and gray-zone passages from 9 generators, including the Kazakh-built KazLLM and Sherkala, with script and length stress tests. Five English-trained detectors from the Hugging Face Hub flag up to 81.4% of human passages and 79.6% of published student essays; recalibrated, they catch at most 1.3% of machine text. Calibrated zero-shot detectors catch at most 50.3% of machine text from in-distribution generators. Fine-tuned encoders rank text almost perfectly, but mDeBERTa at its default threshold flags 24.4% of 50-word human excerpts, partly because machine outputs are shorter than human text; calibrated, it catches only 47.1% of KazLLM and 62.7% of Sherkala passages. Without script handling, machine-transliterated Latin-2021 text defeats every detector. Fixes that need no new data (calibration, script canonicalization, script-augmented and length-matched training, and an OR with a zero-shot scorer) form AINA-KK-Det. On test it flags 0.9% of human passages (95% CI 0.4-2.7; 1.5% on development data) and catches 85.2% of machine text from known generators, 92.9% from two unseen Gemini-family Gemma-4 models and 61.3% from the Kazakh-built ones; its configuration was chosen with test results in view. We release the data, code and detector at https://github.com/ozxmn/aina-kk.

Lean verifies that a generated declaration is well typed, but cannot check whether it states what the user intended. We measure how far compilation overstates faithfulness in statement autoformalization without canonical Lean targets. Our criterion requires Lean compilation and agreement of two LLM judges from different model families. On an independent random audit, it agrees with the human majority on 91.5% of cases; it rarely accepts an unfaithful output but rejects some faithful ones. Across eight systems on 227 university-level statements, the gap between compiling and accepted outputs ranges from 1.3 to 29.5 percentage points: our tool-augmented agent compiles 87.2% of the statements, but only 57.7% are accepted. Human review of all the agent's rejections confirms that 23.8 of its 29.5 points are real translation errors. For the gaps of the seven one-shot systems, a two-stage review by Claude Opus 5.5 with Lean and GPT-6.1 Sol, validated on these human labels, finds that 78-100% are real translation errors for six of them. When GPT-5.2, one of the two judges, grades the translations of two models released after it, GPT-6 Astra and Claude Opus 5.5, it accepts 83% of the statements for each, against 21% for one-shot GPT-5.2. For these two models, the same review classifies only 4 of their 54 rejections (7%) as translation errors, so most of their measured gap is judge errors rather than translation errors. This is consistent with these systems erring less while the judge does not: the fewer errors a system makes, the smaller the share of its rejections that are translation errors. Finally, a 2^3 tool ablation of the agent shows that Lean feedback drives most of its compilation gain, and that nearly half of this gain fails the semantic criterion. Code and data are available at https://github.com/grenadeComing/beyond-compilation.

Medical coding is the task of assigning medical codes to clinical free-text documentation. Healthcare professionals manually assign such codes to track patient diagnoses and treatments. Automated medical coding can considerably alleviate this administrative burden. In this paper, we reproduce, compare, and analyze state-of-the-art automated medical coding machine learning models. We show that several models underperform due to weak configurations, poorly sampled train-test splits, and insufficient evaluation. In previous work, the macro F1 score has been calculated sub-optimally, and our correction doubles it. We contribute a revised model comparison using stratified sampling and identical experimental setups, including hyperparameters and decision boundary tuning. We analyze prediction errors to validate and falsify assumptions of previous works. The analysis confirms that all models struggle with rare codes, while long documents only have a negligible impact. Finally, we present the first comprehensive results on the newly released MIMIC-IV dataset using the reproduced models. We release our code, model parameters, and new MIMIC-III and MIMIC-IV training and evaluation pipelines to accommodate fair future comparisons.

In text-to-SQL and AI-agent analytics, successful SQL execution does not establish semantic correctness under changing business definitions. We study revalidation of fixed queries after changes to refund attribution, tax treatment, eligible populations and exchange-rate dates. An executable verification profile binds the requested revision and knowledge cutoff to legal counterexample data and exact monetary comparison. A preregistered confirmation on 32 newly authored revision families and two local producer bundles retains 192 query-definition cells: 144 nonconforming, 36 finite-suite agreeing and 12 unresolved. Of the nonconforming cells, 63 agree on the current snapshot. At an eight-ledger budget, obligation-guided checking detects 137 discrepancies versus 127 for generic ordering, a paired difference of 6.9 percentage points (family bootstrap 95% interval [2.1, 13.6]; exact family sign-flip p=0.0156). Every ledger arm rejects zero of the 36 agreeing cells. Current-obligation and equal-weight scoring match the guided result, identifying coverage rather than revision-specific priority as the observed source of benefit. Controlled assays, a natural-query pilot and repeated samples provide separate mechanism and robustness evidence. A legal discrepancy refutes conformance; agreement remains restricted to the recorded suite. The study establishes an auditable business-semantic revalidation profile within its authored schema and local-model scope.

Autonomous coding agents solve repository issues by reading code, running commands, editing files, and submitting patches. Extra inference-time compute yields gains only when it produces a useful repair and supplies reliable evidence for choosing one. Three behaviors decide both, and we argue they are teachable rather than byproducts of scale, so a policy can carry them instead of a scaffold. Location diversity remains narrow, since attempts return to the same site and extra samples add no coverage. Edit diversity is left unexploited, since methodologies that differ resolve complementary issues no single run reaches. Verification misleads, since a test the agent writes for its own patch accepts many incorrect ones. Directing search by execution feedback and scoring each patch against its own reverted tree resolves 52.8% of SWE-bench Verified using 48.1% of the agent-steps an eight-sample baseline spends. Training moves these behaviors into the policy. On the 270 issues held out from SFT and RL training, weighted supervised fine-tuning raises pass@1 from 31.9% to 35.2% and pass@8 from 46.7% to 51.1%. A reinforcement objective then trains the verifier against gold-labeled repairs and incorrect variants, crediting the assertions that detect them. It raises pass@1 to 43.0% and pass@8 to 60.7%, lifts verifier precision from 26.8% to 41.7%, and more than halves false acceptance. Resolution improves on two of three out-of-distribution suites and verifier precision on all three, and the gains hold at 7B, 14B, and 30B against published coder baselines.

This paper focuses on developing translation models and related applications for 36 Indian languages, including Assamese, Awadhi, Bengali, Bhojpuri, Braj, Bodo, Dogri, English, Konkani, Gondi, Gujarati, Hindi, Hinglish, Ho, Kannada, Kangri, Kashmiri (Arabic and Devanagari), Khasi, Mizo, Magahi, Maithili, Malayalam, Marathi, Manipuri (Bengali and Meitei), Nepali, Oriya, Punjabi, Sanskrit, Santali, Sinhala, Sindhi (Arabic and Devanagari), Tamil, Tulu, Telugu, and Urdu. Achieving this requires parallel and other types of corpora for all 36 * 36 language pairs, addressing challenges like script variations, phonetic differences, and syntactic diversity. For instance, languages like Kashmiri and Sindhi, which use multiple scripts, demand script normalization for alignment, while low-resource languages such as Khasi and Santali require synthetic data augmentation to ensure sufficient coverage and quality. To address these challenges, this work proposes strategies for corpus creation by leveraging existing resources, developing parallel datasets, generating domain-specific corpora, and utilizing synthetic data techniques. Additionally, it evaluates machine translation across various dimensions, including standard and discourse-level translation, domain-specific translation, reference-based and reference-free evaluation, error analysis, and automatic post-editing. By integrating these elements, the study establishes a comprehensive framework to improve machine translation quality and enable better cross-lingual communication in India's linguistically diverse ecosystem.

Multilingual pre-trained models (mPLMs) have shown impressive performance on cross-lingual transfer tasks. However, the transfer performance is often hindered when a low-resource target language is written in a different script than the high-resource source language, even though the two languages may be related or share parts of their vocabularies. Inspired by recent work that uses transliteration to address this problem, our paper proposes a transliteration-based post-pretraining alignment (PPA) method aiming to improve the cross-lingual alignment between languages using diverse scripts. We select two areal language groups, Mediterranean-Amharic-Farsi and South+East Asian Languages, wherein the languages are mutually influenced but use different scripts. We apply our method to these language groups and conduct extensive experiments on a spectrum of downstream tasks. The results show that after PPA, models consistently outperform the original model (up to 50% for some tasks) in English-centric transfer. In addition, when we use languages other than English as sources in transfer, our method obtains even larger improvements. We will make our code and models publicly available at https://github.com/cisnlp/Transliteration-PPA.

Code Agent development is an extremely active research area, where a reliable performance metric is critical for tracking progress and guiding new developments. This demand is underscored by the meteoric rise in popularity of SWE-Bench. This benchmark challenges code agents to generate patches addressing GitHub issues given the full repository as context. The correctness of generated patches is then evaluated by executing a human-written test suite extracted from the repository after the issue's resolution. However, constructing benchmarks like SWE-Bench requires substantial manual effort to set up historically accurate execution environments for testing. Crucially, this severely limits the number of considered repositories, e.g., just 12 for SWE-Bench. Considering so few repositories, selected for their popularity runs the risk of leading to a distributional mismatch, i.e., the measured performance may not be representative of real-world scenarios potentially misguiding development efforts. In this work, we address this challenge and introduce SetUpAgent, a fully automated system capable of historically accurate dependency setup, test execution, and result parsing. Using SetUpAgent, we generate two new datasets: (i) SWEE-Bench an extended version of SWE-Bench encompassing hundreds of repositories, and (ii) SWA-Bench a benchmark focusing on applications rather than libraries. Comparing these datasets to SWE-Bench with respect to their characteristics and code agent performance, we find significant distributional differences, including lower issue description quality and detail level, higher fix complexity, and most importantly up to 40% lower agent success rates.

Ambient AI scribes draft clinical notes, and published audits find their dominant error is omission: information the encounter established that the note fails to record. The standard check is an LLM judge: a second model reads the note against the transcript and flags problems. We ask whether judges detect omissions. Public corpora cannot supply the answer key: their clinician reference notes and transcripts are materially discrepant. Our benchmark has 500 single-error note pairs from audited fact sheets, 298 with a named fact certainly absent and 202 added-or-altered controls. Across eight judge designs, paired discrimination (the flawed note below its clean twin, 0.5 a coin flip) reads 0.79-0.94 on added or altered content and 0.50-0.63 on omissions. On single notes, no design flags omissions reliably more often than perfect notes. Wording changes, voting and GEPA prompt optimisation move the operating point without creating usable detection. Restructuring the task recovers it: list the facts the transcript establishes, then check the note for each. Two methods reach it independently and trade off: a per-fact pipeline, and a GEPA-evolved prompt doing the same in one call. The pipeline's flags name the missing fact and its severity at 2.7% false alarms. The single call detects more (36.9% against 24.6%, p=0.002) at 6.2% false alarms and a tenth of the cost per note. A physician author validated 70 items and, where the two routes disagree, sided with the pipeline on 10 of 10 (p=0.002). A second clinician, not an author, graded the severity rubric blind and agrees to within a grade. On real vendor notes from a companion census no benchmark threshold transfers, but the re-calibrated single call detects more than the best of the eight at half its false-alarm rate. Omissions whose fact is restated elsewhere defeat both routes. We release the benchmark, prompts and judgements.

Large Multimodal Models (LMMs) have witnessed remarkable growth, showcasing formidable capabilities in handling intricate multimodal tasks with exceptional performance. Recent research has underscored the inclination of large language models to passively accept defective inputs, often resulting in futile reasoning on invalid prompts. However, the same critical question of whether LMMs can actively detect and scrutinize erroneous inputs still remains unexplored. To address this gap, we introduce the Input Scrutiny Ability Evaluation Framework (ISEval), which encompasses seven categories of flawed premises and three evaluation metrics. Our extensive evaluation of ten advanced LMMs has identified key findings. Most models struggle to actively detect flawed textual premises without guidance, which reflects a strong reliance on explicit prompts for premise error identification. Error type affects performance: models excel at identifying logical fallacies but struggle with surface-level linguistic errors and certain conditional flaws. Modality trust varies-Gemini 2.5 pro and Claude Sonnet 4 balance visual and textual info, while aya-vision-8b over-rely on text in conflicts. These insights underscore the urgent need to enhance LMMs' proactive verification of input validity and shed novel insights into mitigating the problem. The code is available at https://github.com/MLGroupJLU/LMM_ISEval.

Alignment faking in large language models presented a demonstration of Claude 3 Opus and Claude 3.5 Sonnet selectively complying with a helpful-only training objective to prevent modification of their behavior outside of training. We expand this analysis to 25 models and find that only 5 (Claude 3 Opus, Claude 3.5 Sonnet, Llama 3 405B, Grok 3, Gemini 2.0 Flash) comply with harmful queries more when they infer they are in training than when they infer they are in deployment. First, we study the motivations of these 5 models. Results from perturbing details of the scenario suggest that only Claude 3 Opus's compliance gap is primarily and consistently motivated by trying to keep its goals. Second, we investigate why many chat models don't fake alignment. Our results suggest this is not entirely due to a lack of capabilities: many base models fake alignment some of the time, and post-training eliminates alignment-faking for some models and amplifies it for others. We investigate 5 hypotheses for how post-training may suppress alignment faking and find that variations in refusal behavior may account for a significant portion of differences in alignment faking.

Multiple choice questions (MCQs) are commonly used to evaluate the capabilities of large language models (LLMs). One common way to evaluate the model response is to rank the candidate answers based on the log probability of the first token prediction. An alternative way is to examine the text output. Prior work has shown that first token probabilities lack robustness to changes in MCQ phrasing, and that first token probabilities do not match text answers for instruction-tuned models. Therefore, in this paper, we investigate the robustness of text answers. We show that the text answers are more robust to question perturbations than the first token probabilities, when the first token answers mismatch the text answers. The difference in robustness increases as the mismatch rate becomes greater. As the mismatch reaches over 50\%, the text answer is more robust to option order changes than the debiased first token probabilities using state-of-the-art debiasing methods such as PriDe. Our findings provide further evidence for the benefits of text answer evaluation over first token probability evaluation.

Benchmark contamination is diagnosed with n-gram overlap, likelihood-based membership inference, or canary strings, and each needs something usually unavailable: the training corpus, a well-chosen test statistic, or foresight at release. A recent alternative reads it off a linear probe on internal activations. We show the natural way to do this does not work, specify one that survives measurement, then find that the correction making it work carries more variance than the null it is tested against. The protocol reports a zero-sum contrast on the depth profile of probe accuracy, recentred on a level-matched placebo baseline, tested against a label-permutation null, with the reference set twice the size of the suspect set. Each choice replaces a simpler alternative we rejected on measurement. Reporting the level of excess separability rather than its shape makes the false positive rate track the size of the analyst's own control set, 0.03 to 0.99 under a true null. Contrasting against a flat depth profile rejects a true null 0.72 of the time when surface decodability rises with depth, and loses all power when it falls. On real transformers the protocol fails a test the simulations did not pose. The recentring subtracts an estimate, and the permutation null holds it fixed. Re-estimated across split seeds on four audits of contaminated checkpoints, its standard deviation is 1.30 to 1.56 times the null's own in every arm: what is subtracted to remove a bias is more variable than what it corrects. The one nominally significant result, p = 0.0075, becomes 0.0745 once that variance is propagated, and no verdict is issued. The simulations missed this because their surface key is the covariate driving item variation; on real text it is a proxy, and degrading key quality in simulation reproduces it. We add a companion measurement and a widened null. No arm shows contamination.

Evolving documents, such as laws, tax codes, and software documentation, are amended, replaced, and sometimes reverted over time, so a question has different correct answers at different dates. In contrast to encyclopedic knowledge, where an old fact is simply overwritten, an amendment is itself an official text that states what it replaces and when it takes effect, and the earlier version stays correct for its validity period. The central challenge is therefore version resolution, that is, identifying the version in force on the queried date. Existing temporal QA datasets treat time only as an annotation, so version resolution stays untested. We present TIDE, an expert-verified benchmark of 3,050 QA pairs over 644 official customs instruments issued between 1969 and 2025 by the Government of Bangladesh, covering eight task types over deeply code-mixed documents that are heterogeneous in layout and dated in two calendars. In addition, we evaluate nine recent LLMs under a single protocol across parametric, gold-context, and retrieval access, scored by a three-judge LLM council with a hard date gate separating correct meaning from correct time. The best macro-averaged accuracy is only 68.5%. Resolving a version from an implicit date reaches 59.7%, and detecting that the supplied version does not govern the query reaches only 26.7%. Models are more likely to find correct versions than to reject incorrect ones, and they tend to follow a confident parametric answer over the supplied authoritative text. All code and data are available at https://github.com/icsetepa44/TIDE

Reliable evaluation is essential for understanding large language model (LLM) performance, yet today's go-to metrics, namely token-overlap scores (e.g., ROUGE) and embedding-based measures (e.g., BERTScore), often misjudge semantic similarity of documents. Our study shows that both token-overlap metrics and embedding-based metrics routinely assign nearly identical scores to texts that directly contradict each other, thereby potentially masking fundamental errors. We introduce MATCHA, an automatic metric that jointly rewards semantic agreement with a reference and penalizes contradictions. MATCHA employs a dual-view perspective that measures (i) proximity to the gold text and (ii) distance from an adversarially generated counterfactual contradiction. In eight public benchmarks, MATCHA outperforms popular metrics, compared with human annotations on question-answering, image caption generation, natural language inference, summarization, and semantic textual similarity tasks. On the TruthfulQA dataset (i.e., a dataset without a training set, where no embedding-based metrics could locally train on), this improvement in terms of matching texts with a reference reaches 18.38% over ROUGE-L and 20.82% over BERTScore. Both quantitative comparison and qualitative human assessments confirm the efficacy and validity of MATCHA and uncover fundamental weaknesses in pre-existing metrics. Compared with 23 embedding models, including top state-of-the-art ones, used as a metric similar to BERTScore, MATCHA remains the most accurate in distinguishing correct from incorrect statements solely based on a reference. Our code and metric are publicly available (https://github.com/Siran-Li/MATCHA).

Prior work has shown that fine-tuning models on a narrow domain with misaligned data can lead to broad misalignment - a phenomenon termed "emergent misalignment" (Betley et al. 2025). While all tested models were susceptible to emergent misalignment, some models showed more resistance than others. Specifically the Qwen-2.5 family proved to be relatively resistant, while GPT-4o exhibited the strongest misalignment. In this paper we evaluate if current-generation open-weights models exhibit similar resistance to the Qwen-2.5 family and measure misalignment robustness over a range of model architectures and scales. We replicate the effect across nine modern open-weights models (Gemma 3 and Qwen 3 families, 1B-32B parameters). Models fine-tuned on insecure code generation show a 0.68% misalignment rate (compared to 0.07% for base models), matching the lower end of prior open-model results but dramatically lower than GPT-4o's 20%. We identify a critical format-dependent vulnerability: requiring JSON output doubles misalignment rates compared to natural language prompts (0.96% vs 0.42%). This suggests that structural constraints may bypass safety training by reducing the model's 'degrees of freedom' to refuse. These findings confirm emergent misalignment as a reproducible phenomenon in modern open-weights models, with rates substantially lower than observed in proprietary systems.

Abugida refers to a phonogram writing system where each syllable is represented using a single consonant or typographic ligature, along with a default vowel or optional diacritic(s) to denote other vowels. However, texting in these languages has some unique challenges in spite of the advent of devices with soft keyboard supporting custom key layouts. The number of characters in these languages is large enough to require characters to be spread over multiple views in the layout. Having to switch between views many times to type a single word hinders the natural thought process. This prevents popular usage of native keyboard layouts. On the other hand, supporting romanized scripts (native words transcribed using Latin characters) with language model based suggestions is also set back by the lack of uniform romanization rules. To this end, we propose a disambiguation algorithm and showcase its usefulness in two novel mutually non-exclusive input methods for languages natively using the abugida writing system: (a) disambiguation of ambiguous input for abugida scripts, and (b) disambiguation of word variants in romanized scripts. We benchmark these approaches using public datasets, and show an improvement in typing speed by 19.49%, 25.13%, and 14.89%, in Hindi, Bengali, and Thai, respectively, using Ambiguous Input, owing to the human ease of locating keys combined with the efficiency of our inference method. Our Word Variant Disambiguation (WDA) maps valid variants of romanized words, previously treated as Out-of-Vocab, to a vocabulary of 100k words with high accuracy, leading to an increase in Error Correction F1 score by 10.03% and Next Word Prediction (NWP) by 62.50% on average.

Large language models (LLMs) have achieved strong performance across various natural language processing tasks. However, their robustness to typographical errors remains underexplored, particularly in Japanese, where text input involves multiple writing systems and IME-based conversion. In this study, we evaluate the robustness of Japanese LLMs against realistic Japanese-specific typos. We introduce five typo categories: Character Transposition, Character Replacement, Homophone Conversion, Japanese IME Conversion, and Full-Width Conversion. These perturbations are applied to three Japanese benchmark datasets (JMMLU, JCommonsenseQA, and JamC-QA), and eleven Japanese and multilingual LLMs are evaluated. The results show that Character Transposition and Character Replacement typos consistently reduce accuracy across benchmarks, whereas IME Conversion, Full-Width Conversion, and Homophone Conversion have relatively limited impact. These findings reveal that current Japanese LLMs remain vulnerable to realistic Japanese typing errors, particularly those that substantially distort the original input, highlighting the importance of robustness evaluation in practical input environments.

Reinforcement learning (RL) for large language models (LLMs) is dominated by the cost of rollout generation, which has motivated the use of low-precision rollouts (e.g., FP8) paired with a BF16 trainer to improve throughput and reduce memory pressure. This introduces a rollout-training mismatch that biases the policy gradient and can cause training to collapse outright on reasoning benchmarks. We show that the mismatch is non-stationary and acts as a double-edged sword: early in training it provides a stochastic exploration bonus, exposing the gradient to trajectories the trainer would otherwise under-sample, but the same perturbation transitions into a destabilizing source of bias as the policy concentrates. To solve this, we propose Adaptive Importance Sampling (AIS), a correction framework that adjusts the strength of its intervention on a per-batch basis. AIS combines three real-time diagnostics, namely weight reliability, divergence severity, and variance amplification, into a single mixing coefficient that interpolates between the uncorrected and fully importance-weighted gradients, suppressing the destabilizing component of the mismatch while preserving its exploratory benefit. We integrate AIS into GRPO and evaluate it on the diffusion-based LLaDA-8B-Instruct and the autoregressive Qwen3-8B and Qwen3.5-9B across mathematical reasoning and planning benchmarks. AIS matches the BF16 baseline on most tasks while retaining the 1.5 to 2.76x rollout speedup of FP8.

Modern deep neural networks can achieve high accuracy when the training distribution and test distribution are identically distributed, but this assumption is frequently violated in practice. When the train and test distributions are mismatched, accuracy can plummet. Currently there are few techniques that improve robustness to unforeseen data shifts encountered during deployment. In this work, we propose a technique to improve the robustness and uncertainty estimates of image classifiers. We propose AugMix, a data processing technique that is simple to implement, adds limited computational overhead, and helps models withstand unforeseen corruptions. AugMix significantly improves robustness and uncertainty measures on challenging image classification benchmarks, closing the gap between previous methods and the best possible performance in some cases by more than half.

We present Code Comparison Tuning (CCT), a simple and effective tuning method for code large language models (Code LLMs) to better handle subtle code errors. Specifically, we integrate the concept of comparison into instruction tuning, both at the token and sequence levels, enabling the model to discern even the slightest deviations in code. To compare the original code with an erroneous version containing manually added code errors, we use token-level preference loss for detailed token-level comparisons. Additionally, we combine code segments to create a new instruction tuning sample for sequence-level comparisons, enhancing the model's bug-fixing capability. Experimental results on the HumanEvalFix benchmark show that CCT surpasses instruction tuning in pass@1 scores by up to 4 points across diverse code LLMs, and extensive analysis demonstrates the effectiveness of our method.

Like a criminal under investigation, Large Language Models (LLMs) might pretend to be aligned while evaluated and misbehave when they have a good opportunity. Can current interpretability methods catch these 'alignment fakers?' To answer this question, we introduce a benchmark that consists of 324 pairs of LLMs fine-tuned to select actions in role-play scenarios. One model in each pair is consistently benign (aligned). The other model misbehaves in scenarios where it is unlikely to be caught (alignment faking). The task is to identify the alignment faking model using only inputs where the two models behave identically. We test five detection strategies, one of which identifies 98% of alignment-fakers.

Large language models (LLMs) such as ChatGPT have exhibited remarkable performance in generating human-like texts. However, machine-generated texts (MGTs) may carry critical risks, such as plagiarism issues, misleading information, or hallucination issues. Therefore, it is very urgent and important to detect MGTs in many situations. Unfortunately, it is challenging to distinguish MGTs and human-written texts because the distributional discrepancy between them is often very subtle due to the remarkable performance of LLMs. In this paper, we seek to exploit maximum mean discrepancy (MMD) to address this issue in the sense that MMD can well identify distributional discrepancies. However, directly training a detector with MMD using diverse MGTs will incur a significantly increased variance of MMD since MGTs may contain multiple text populations due to various LLMs. This will severely impair MMD's ability to measure the difference between two samples. To tackle this, we propose a novel multi-population aware optimization method for MMD called MMD-MP, which can avoid variance increases and thus improve the stability to measure the distributional discrepancy. Relying on MMD-MP, we develop two methods for paragraph-based and sentence-based detection, respectively. Extensive experiments on various LLMs, \eg, GPT2 and ChatGPT, show superior detection performance of our MMD-MP. The source code is available at https://github.com/ZSHsh98/MMD-MP.

Recently, the large language models (LLMs) have shown extraordinary ability in understanding natural language and generating programming code. It has been a common practice of software engineers to consult LLMs when encountering coding questions. Although efforts have been made to avoid syntax errors and align the code with the intended semantics, the reliability and robustness of the code generationfrom LLMs have not yet been thoroughly studied. The executable code is not equivalent to the reliable and robust code, especially in the context of real-world software development. The misuse of APIs in the generated code could lead to severe problem, such as resource leaks, program crashes. To make things worse, the users of LLM code generation services are actually the developers that are most vulnerable to these code that seems right -- They are always novice developers that are not familiar with the APIs that LLMs generate code for them. Therefore, they could hardly tell the misuse in the code generated by LLMs, which further facilitates the incorrect code applied in real-world software. Existing code evaluation benchmark and datasets focus on crafting small tasks such as programming questions in coding interviews, which however deviates from the problem that developers would ask LLM for real-world coding help. To fill the missing piece, in this work, we propose a dataset RobustAPI for evaluating the reliability and robustness of code generated by LLMs. We collect 1208 coding questions from StackOverflow on 24 representative Java APIs. We summarize thecommon misuse patterns of these APIs and evaluate them oncurrent popular LLMs. The evaluation results show that evenfor GPT-4, 62% of the generated code contains API misuses,which would cause unexpected consequences if the code isintroduced into real-world software.

The BigCode project, an open-scientific collaboration focused on the responsible development of Large Language Models for Code (Code LLMs), introduces StarCoder2. In partnership with Software Heritage (SWH), we build The Stack v2 on top of the digital commons of their source code archive. Alongside the SWH repositories spanning 619 programming languages, we carefully select other high-quality data sources, such as GitHub pull requests, Kaggle notebooks, and code documentation. This results in a training set that is 4x larger than the first StarCoder dataset. We train StarCoder2 models with 3B, 7B, and 15B parameters on 3.3 to 4.3 trillion tokens and thoroughly evaluate them on a comprehensive set of Code LLM benchmarks. We find that our small model, StarCoder2-3B, outperforms other Code LLMs of similar size on most benchmarks, and also outperforms StarCoderBase-15B. Our large model, StarCoder2- 15B, significantly outperforms other models of comparable size. In addition, it matches or outperforms CodeLlama-34B, a model more than twice its size. Although DeepSeekCoder- 33B is the best-performing model at code completion for high-resource languages, we find that StarCoder2-15B outperforms it on math and code reasoning benchmarks, as well as several low-resource languages. We make the model weights available under an OpenRAIL license and ensure full transparency regarding the training data by releasing the SoftWare Heritage persistent IDentifiers (SWHIDs) of the source code data.

Recent advances in automated test generation utilises language models to produce unit tests. While effective, language models tend to generate many incorrect tests with respect to both syntax and semantics. Although such incorrect tests can be easily detected and discarded, they constitute a "missed opportunity" -- if fixed, they are often valuable as they directly add testing value (they effectively target the underlying program logic to be tested) and indirectly form good seeds for generating additional tests. To this end, we propose a simple technique for repairing some of these incorrect tests through a combination of rule-based static analysis and re-prompting. We evaluate this simple approach, named YATE, on a set of 6 open-source projects and show that it can effectively produce tests that cover on average 32.06% more lines and kill 21.77% more mutants than a plain LLM-based method. We also compare YATE with four other LLM-based methods, namely HITS, SYMPROMPT, TESTSPARK and COVERUP and show that it produces tests that cover substantially more code. YATE achieves 22% higher line coverage, 20% higher branch coverage and kill 20% more mutants at a comparable cost (number of calls to LLMs).

Large Language Model (LLM) agents rely on external tools to perform multistage tasks. Existing agent frameworks typically assume that tool calls are atomic and return binary success or failure signals. However, real-world systems exhibit non-atomic behaviors such as timeouts after dispatch, delayed visibility, and partial state updates. These mismatches lead to reliability issues including duplicate actions, task success, and unnecessary tool executions. A lightweight, verification-aware tool wrapper is introduced that augments tool calls with postcondition verification, verify-before-retry logic, and idempotency keys. The approach is evaluated in a controlled simulated environment with injected non-atomic failures across multiple task templates. The results demonstrate that the proposed method significantly reduces duplicate actions, while maintaining comparable task success rates. Overall, the findings suggest that strengthening tool interaction semantics is a promising direction for improving LLM agent reliability without requiring modifications to the underlying language model.

Most programmers make mistakes when writing code. Some of these mistakes are small and require few edits to the original program -- a class of errors recently termed last mile mistakes. These errors break the flow for experienced developers and can stump novice programmers. Existing automated repair techniques targeting this class of errors are language-specific and do not easily carry over to new languages. Transferring symbolic approaches requires substantial engineering and neural approaches require data and retraining. We introduce RING, a multilingual repair engine powered by a large language model trained on code (LLMC) such as Codex. Such a multilingual engine enables a flipped model for programming assistance, one where the programmer writes code and the AI assistance suggests fixes, compared to traditional code suggestion technology. Taking inspiration from the way programmers manually fix bugs, we show that a prompt-based strategy that conceptualizes repair as localization, transformation, and candidate ranking, can successfully repair programs in multiple languages with minimal effort. We present the first results for such a multilingual repair engine by evaluating on 6 different languages and comparing performance to language-specific repair engines. We show that RING can outperform language-specific repair engines for three of these languages.

Despite its crucial role in research experiments, code correctness is often presumed only on the basis of the perceived quality of results. This assumption comes with the risk of erroneous outcomes and potentially misleading findings. To address this issue, we posit that the current focus on reproducibility should go hand in hand with the emphasis on software quality. We present a case study in which we identify and fix three bugs in widely used implementations of the state-of-the-art Conformer architecture. Through experiments on speech recognition and translation in various languages, we demonstrate that the presence of bugs does not prevent the achievement of good and reproducible results, which however can lead to incorrect conclusions that potentially misguide future research. As a countermeasure, we propose a Code-quality Checklist and release pangoliNN, a library dedicated to testing neural models, with the goal of promoting coding best practices and improving research software quality within the NLP community.

As LLM-based agents are deployed in production systems, understanding their behavioral consistency (whether they produce similar action sequences when given identical tasks) becomes critical for reliability. We study consistency in the context of SWE-bench, a challenging software engineering benchmark requiring complex, multi-step reasoning. Comparing Claude~4.5~Sonnet, GPT-5, and Llama-3.1-70B across 50 runs each (10 tasks times 5 runs), we find that across models, higher consistency aligns with higher accuracy: Claude achieves the lowest variance (CV: 15.2\%) and highest accuracy (58\%), GPT-5 is intermediate (CV: 32.2\%, accuracy: 32\%), and Llama shows the highest variance (CV: 47.0\%) with lowest accuracy (4\%). However, within a model, consistency can amplify both correct and incorrect interpretations. Our analysis reveals a critical nuance: consistency amplifies outcomes rather than guaranteeing correctness. 71\% of Claude's failures stem from "consistent wrong interpretation": making the same incorrect assumption across all runs. Interestingly, GPT-5 achieves similar early strategic agreement as Claude (diverging at step 3.4 vs.\ 3.2) but exhibits 2.1times higher variance, suggesting that divergence timing alone does not determine consistency. These findings suggest that for production deployment, interpretation accuracy matters more than execution consistency, with implications for agent evaluation and training.

Industrial Retrieval-Augmented Generation (RAG) systems depend on optical character recognition (OCR) to transform visual documents into text. Existing OCR benchmarks rely on character-level metrics, which inadequately measure downstream RAG effectiveness under real-world conditions. We introduce an OCR benchmark for industrial RAG systems covering 11 challenging document types, including extreme layouts, high-resolution pages, complex or watermarked backgrounds, historical documents with non-standard reading orders, visually decorated text, and documents containing tables and mathematical formulas. Evaluating recent SOTA OCR models under a controlled OCR-first RAG pipeline shows clear performance degradation on realistic industrial documents despite strong conventional benchmark scores. We find that high OCR accuracy does not necessarily translate into strong downstream RAG performance: structural and semantic errors can cause substantial retrieval failures even when WER/CER remains low. Further analysis shows that this mismatch is category-dependent, arises through both retrieval-side and downstream generation-side failures, and remains stable across representative OCR-first pipeline choices. The benchmark is publicly available at https://github.com/Qihoo360/InduOCRBench.

Large Language Models (LLMs) have achieved remarkable success in code generation tasks, powering various applications like code completion, debugging, and programming assistance. However, existing benchmarks such as HumanEval, MBPP, and BigCodeBench primarily evaluate LLMs on English-only prompts, overlooking the real-world scenario where multilingual developers often use code-mixed language while interacting with LLMs. To address this gap, we introduce CodeMixBench, a novel benchmark designed to evaluate the robustness of LLMs on code generation from code-mixed prompts. Built upon BigCodeBench, CodeMixBench introduces controlled code-mixing (CMD) into the natural language parts of prompts across three language pairs: Hinglish (Hindi-English), Spanish-English, and Chinese Pinyin-English. We comprehensively evaluate a diverse set of open-source code generation models ranging from 1.5B to 15B parameters. Our results show that code-mixed prompts consistently degrade Pass@1 performance compared to their English-only counterparts, with performance drops increasing under higher CMD levels for smaller models. CodeMixBench provides a realistic evaluation framework for studying multilingual code generation and highlights new challenges and directions for building robust code generation models that generalize well across diverse linguistic settings.

This study investigates various approaches to using Large Language Models (LLMs) for Text-to-SQL program synthesis, focusing on the outcomes and insights derived. Employing the popular Text-to-SQL dataset, spider, the goal was to input a natural language question along with the database schema and output the correct SQL SELECT query. The initial approach was to fine-tune a local and open-source model to generate the SELECT query. After QLoRa fine-tuning WizardLM's WizardCoder-15B model on the spider dataset, the execution accuracy for generated queries rose to a high of 61%. With the second approach, using the fine-tuned gpt-3.5-turbo-16k (Few-shot) + gpt-4-turbo (Zero-shot error correction), the execution accuracy reached a high of 82.1%. Of all the incorrect queries, most can be categorized into a seven different categories of what went wrong: selecting the wrong columns or wrong order of columns, grouping by the wrong column, predicting the wrong values in conditionals, using different aggregates than the ground truth, extra or too few JOIN clauses, inconsistencies in the Spider dataset, and lastly completely incorrect query structure. Most if not all of the queries fall into these categories and it is insightful to understanding where the faults still lie with LLM program synthesis and where they can be improved.

A manga reader meets a character on one page and knows them on sight a hundred pages later, without ever being handed a cast list. Re-identifying comic characters demands the same, open-set and sequential: pages arrive as a stream in reading order, new faces appear before anyone names them, and the cast is assembled as the story is read. Re:Cognize evaluates recognition as the story is read, not against a cast handed over in advance: four protocols on one query stream, from closed-set retrieval to a cast the model must build and grow itself. The surprise is where models fail. Recognising is close to solved: one reference image per character already ranks as well as a gallery built in advance. Knowing what to believe is not: a model that adds its own matches makes its cast worse, while the same growth with correct labels would gain over twenty points of top-1 accuracy. The bottleneck is acceptance, not vision, and one comparison decides it: an addition pays exactly when it is right more often than the cast already was on the queries it takes over. The comparison has nothing to fit, and measured on half of a new corpus it calls the other half correctly. ReCast puts it to work with nothing fitted on data: a cast sheet of one running average per character, grown only where the page itself vouches for a crop. It recovers a third to two thirds of what perfect labels would, depending on whether the cast starts from random examples or from first appearances. Re:Cognize measures whether a model can read along; ReCast is a cast that does. Our claims are on identity maintenance, recognising characters already met; the emergence of new ones is measured as a diagnostic under a fixed reference rule, and we propose no method for it.

We report upon the results of a research and prototype building project Worldly~OCR dedicated to developing new, more accurate image-to-text conversion software for several languages and writing systems. These include the cursive scripts Farsi and Pashto, and Latin cursive scripts. We also describe approaches geared towards Traditional Chinese, which is non-cursive, but features an extremely large character set of 65,000 characters. Our methodology is based on Machine Learning, especially Deep Learning, and Data Science, and is directed towards vast quantities of original documents, exceeding a billion pages. The target audience of this paper is a general audience with interest in Digital Humanities or in retrieval of accurate full-text and metadata from digital images.

Pre-training data mixtures are commonly tuned by running small-scale experiments and extrapolating to the target training budget. When high-quality data is scarce and must be repeated, this extrapolation frequently fails, but the source of the failure has not been isolated. We show that a primary culprit is a repetition mismatch: because high-quality datasets are small, their repetition rate changes as the training budget grows, shifting the optimal mixture in ways that small-scale proxy experiments do not anticipate. A subsampling procedure that matches the target repetition rate controls for this effect. In a two-source setting combining limited high-quality data with web crawl, a single repetition-controlled experiment using only 1/16 of the target tokens recovers a mixture within 0.05 of the optimum for a 757M parameter model, compared to an error of 0.75 without repetition control. Achieving comparable accuracy without repetition control requires three to four horizons, consuming 44 to 94% of the target token budget. With three data sources, the larger mixture space requires more than a single experiment to constrain, but the approach remains effective: at the 757M scale, just two repetition-controlled horizons recover the optimal mixture, outperforming baselines that instead require the full two-source experiments to construct. Our results reveal that repetition dynamics, not scale alone, shape whether small-scale mixture experiments generalize. More broadly, they suggest that data repetition deserves treatment as a first-class variable in mixture optimization, rather than an inconvenient side effect of limited data.

Context: The SZZ algorithm is the de facto standard for labeling bug fixing commits and finding inducing changes for defect prediction data. Recent research uncovered potential problems in different parts of the SZZ algorithm. Most defect prediction data sets provide only static code metrics as features, while research indicates that other features are also important. Objective: We provide an empirical analysis of the defect labels created with the SZZ algorithm and the impact of commonly used features on results. Method: We used a combination of manual validation and adopted or improved heuristics for the collection of defect data. We conducted an empirical study on 398 releases of 38 Apache projects. Results: We found that only half of the bug fixing commits determined by SZZ are actually bug fixing. If a six-month time frame is used in combination with SZZ to determine which bugs affect a release, one file is incorrectly labeled as defective for every file that is correctly labeled as defective. In addition, two defective files are missed. We also explored the impact of the relatively small set of features that are available in most defect prediction data sets, as there are multiple publications that indicate that, e.g., churn related features are important for defect prediction. We found that the difference of using more features is not significant. Conclusion: Problems with inaccurate defect labels are a severe threat to the validity of the state of the art of defect prediction. Small feature sets seem to be a less severe threat.

Low-precision execution can substantially accelerate reinforcement learning (RL) for large language models, but discrepancies between learner and sampler execution can destabilize policy optimization. In this paper, we characterize the interaction between mismatch and the policy-gradient direction, distinguishing locally amplifying from contracting update contributions that mismatch magnitude alone cannot identify. In native NVFP4 runs, we observe an early imbalance between the two amplifying regions, favoring negative-advantage, negative-gap updates. Their tail tokens become concentrated in a small fraction of response segments before mismatch spreads globally. Motivated by these findings, we introduce TRIAGE, a direction-aware stabilization method that uses segment-level diagnosis to selectively rebalance policy-gradient updates and applies bounded repair to residual severe mismatch. TRIAGE modifies the optimization objective while retaining native NVFP4 weight-and activation 4-bit (W4A4) forward execution on both the sampler and learner. Experiments on Qwen3-4B and Qwen3-30B-A3B show stable optimization throughout the evaluated training horizon and achieve full precision level performance across five mathematical reasoning benchmarks, while native NVFP4 with TRIAGE provides up to 2.3x higher rollout throughput than BF16.

An AI scientist should not grade its own homework. Yet in the systems we examined, the agent that reviews the work usually comes from the same model family as the agent that produced it, or at least from the same vendor. Model evaluators are known to favour their own generations. Whether models trained alike also share blind spots is a conjecture, not a settled finding, but if they do, the reviewer inherits the author's. The record of what was flagged and what was waved through often sits in platform logs that nobody outside can replay. We present CrossAudit, a protocol for supervising autonomous research pipelines. It rests on three commitments. Each increment of work is audited by an agent from a different vendor against a rulebook a human wrote and versioned. Reports, verdicts, disputes and rulings are git commits, so the supervision history can be re-read and cited; raw model exchanges are not yet part of that record. Scripted checks run before any model does. Advisory judgement never gates the pipeline: a model blocks only by citing a rule, and no model may waive a deterministic failure. Blockers that survive a bounded number of revision rounds go to a person. We state the protocol as eight invariants. We describe a reference implementation built from GitHub Actions and a few hundred lines of Python, and report a live deployment of a closely related variant in a computational-chemistry pipeline. We also ran a seeded-defect trial (30 increments, 43 seeded defects, one run per configuration). A cross-vendor audit of our own repository then voided its blinding. We adopt that audit's findings and report the corrected results. The trial shows that two vendors read the same rulebook differently. It does not show that either is better. The strongest evidence here is the committed, uncontrolled record of cross-vendor audits of this paper itself.

Current approaches to program synthesis with Large Language Models (LLMs) exhibit a "near miss syndrome": they tend to generate programs that semantically resemble the correct answer (as measured by text similarity metrics or human evaluation), but achieve a low or even zero accuracy as measured by unit tests due to small imperfections, such as the wrong input or output format. This calls for an approach known as Synthesize, Execute, Debug (SED), whereby a draft of the solution is generated first, followed by a program repair phase addressing the failed tests. To effectively apply this approach to instruction-driven LLMs, one needs to determine which prompts perform best as instructions for LLMs, as well as strike a balance between repairing unsuccessful programs and replacing them with newly generated ones. We explore these trade-offs empirically, comparing replace-focused, repair-focused, and hybrid debug strategies, as well as different template-based and model-based prompt-generation techniques. We use OpenAI Codex as the LLM and Program Synthesis Benchmark 2 as a database of problem descriptions and tests for evaluation. The resulting framework outperforms both conventional usage of Codex without the repair phase and traditional genetic programming approaches.

AI coding agents are increasingly used for scientific work, but their end-to-end autonomous research capability remains difficult to verify. We present ResearchClawBench, a benchmark for evaluating autonomous scientific research across 40 tasks from 10 scientific domains. Each task is grounded in a real published paper, provides related literature and raw data, and hides the target paper during evaluation. Expert-curated multimodal rubrics decompose the target scientific artifacts into weighted criteria, enabling evaluation of target-paper-level re-discovery while leaving room for new discovery. We evaluate seven autonomous research (auto-research) agents under a unified protocol and seventeen native LLMs through the lightweight ResearchHarness. Current systems remain far from reliable re-discovery: the strongest autonomous agent, Claude Code, averages 21.5, and the strongest ResearchHarness LLM, Claude-Opus-4.7, averages 20.7, with an LLM frontier mean of only 26.5. Error analysis shows that failures concentrate in experimental protocol mismatch, evidence mismatch, and missing scientific core. ResearchClawBench provides a reproducible evaluation frontier for measuring progress toward autonomous scientific research.

Byte-level BPE tokenizers that use the HuggingFace ByteLevel pre-tokenizer inherit GPT-2's word regex, where a word is defined as L+, one or more Unicode letters. In abugida scripts, vowels are written as combining marks; this pattern therefore splits each word at every vowel sign. Since BPE merges only within a pre-token, those splits persist through training regardless of vocabulary size or corpus composition. We formalise this effect as a training-free lower bound on fertility. Across 26 languages from a parallel corpus, every one of the 17 abugidas is affected, ranging from 1.47x (Tibetan) to 9.02x (Thai), whereas Latin, Cyrillic, Hangul, and Han show exactly 1.00x. For 5 languages, matched tokenizer pairs that differ only in this character class fall within 2.2% of the predicted floor, scoring 4.78 versus 1.58 tokens per word on Nepali. When the Nepali share of the training corpus is swept from 5% to 95%, the broken tokenizer barely shifts at all (1.7%) while the fixed one shifts 33.9%, which separates a structural ceiling from a data shortage without needing to inspect any code. We train three 268M models that differ only in their tokenizer; the fixed variant achieves 4.43% lower held-out Nepali bits per byte at equal compute, and it still leads when given the same bytes with 1.59x the compute. A census of 3,479 HuggingFace repositories finds the letters-only word class present in 63.3% of the most-downloaded text-generation models, accounting for 72.5% of their downloads. GPT-4o's o200k pattern already uses a mark-aware word class, making the repair itself prior art. We quantify its value, show how to recognise its absence from symptoms alone, map which scripts it reaches, measure how widely it is deployed, and release a 65,536-entry Nepali-English tokenizer with a harness that regenerates every number here from public data on a laptop.

Embedding-based code retrieval is a core component of coding agents and retrieval-augmented code generation, where retrieving correct code matters more than retrieving lexically similar code. Existing code-retrieval benchmarks do not plant controlled, execution-verified single-edit variants of each query's canonical implementation in the search pool, leaving the question of whether embeddings can functionally discriminate correct from near-clone-but-incorrect code unanswered in a retrieval setting. Resolving this requires a benchmark whose search pool itself contains the relevant counterfactuals -- execution-verified buggy variants near-identical to each canonical -- so that a retriever's rank ordering can be directly tested for functional discrimination rather than topical or identity overlap. We introduce ExecRetrieval, 939 Python tasks each paired with one execution-verified canonical implementation and up to four execution-verified buggy distractors, each generated by a mechanical mutation making a single targeted edit, and evaluate 23 dense embedding configurations plus BM25 under provider-native invocation with paired McNemar tests and query-level bootstrap intervals. With near-clone counterfactuals in the pool, the top hosted system reaches exec@10 = 1.00 but only exec@1 = 0.331; rank-1 misses are paired buggy variants 91.5-99.4% of the time across the four leading systems, and the canonical scores below at least one of its four paired distractors in 67-78% of queries on the leading systems. The full dataset, execution oracle, embedding matrices, environment snapshot, and pairwise statistical tests are released at the URL in Appendix D.

Large Language Models (LLMs) are increasingly used by software engineers for code generation. However, limitations of LLMs such as irrelevant or incorrect code have highlighted the need for prompt programming (or prompt engineering) where engineers apply specific prompt techniques (e.g., chain-of-thought or input-output examples) to improve the generated code. Despite this, the impact of different prompt techniques -- and their combinations -- on code generation remains underexplored. In this study, we introduce CodePromptEval, a dataset of 7072 prompts designed to evaluate five prompt techniques (few-shot, persona, chain-of-thought, function signature, list of packages) and their effect on the correctness, similarity, and quality of complete functions generated by three LLMs (GPT-4o, Llama3, and Mistral). Our findings show that while certain prompt techniques significantly influence the generated code, combining multiple techniques does not necessarily improve the outcome. Additionally, we observed a trade-off between correctness and quality when using prompt techniques. Our dataset and replication package enable future research on improving LLM-generated code and evaluating new prompt techniques.

This paper proposes one of the first clinical applications of multimodal large language models (LLMs) as an assistant for radiologists to check errors in their reports. We created an evaluation dataset from two real-world radiology datasets (MIMIC-CXR and IU-Xray), with 1,000 subsampled reports each. A subset of original reports was modified to contain synthetic errors by introducing various type of mistakes. The evaluation contained two difficulty levels: SIMPLE for binary error-checking and COMPLEX for identifying error types. LLaVA (Large Language and Visual Assistant) variant models, including our instruction-tuned model, were used for the evaluation. Additionally, a domain expert evaluation was conducted on a small test set. At the SIMPLE level, the LLaVA v1.5 model outperformed other publicly available models. Instruction tuning significantly enhanced performance by 47.4% and 25.4% on MIMIC-CXR and IU-Xray data, respectively. The model also surpassed the domain experts accuracy in the MIMIC-CXR dataset by 1.67%. Notably, among the subsets (N=21) of the test set where a clinician did not achieve the correct conclusion, the LLaVA ensemble mode correctly identified 71.4% of these cases. This study marks a promising step toward utilizing multi-modal LLMs to enhance diagnostic accuracy in radiology. The ensemble model demonstrated comparable performance to clinicians, even capturing errors overlooked by humans. Nevertheless, future work is needed to improve the model ability to identify the types of inconsistency.

Developers download quantized large language models (LLMs) from public registries to run them locally. None of the registries we examined documented functional testing of their files. We tested all 305 eligible code-capable GGUF tags (the llama.cpp format) across 15 model lines in the official Ollama library, plus 22 popular Hugging Face community files. A 15-task code screen identified suspects, which we retested on all 164 HumanEval+ tasks. Seven official tags (2.3%) solved at most six of them. All were low-bit quantizations of models with at most four billion parameters. We rebuilt five suspect files from the original weights. All five registry files matched our uncalibrated rebuilds tensor for tensor, so none was corrupted. Exact reproducibility did not imply function. With the weights, toolchain and tensor mix held fixed, adding an importance matrix raised three Qwen2.5-Coder-3B builds from zero to 76-113 solved tasks. The same uncalibrated recipe left the smaller Qwen2.5-Coder-1.5B functional, showing that the recipe alone does not predict failure. None of 66 sampled tags that passed the screen was non-functional. We release the audit data and the quantcheck tool and recommend screening low-bit tags before publication, then calibrating or withholding failures.

Models trained on different datasets can be merged by a weighted-averaging of their parameters, but why does it work and when can it fail? Here, we connect the inaccuracy of weighted-averaging to mismatches in the gradients and propose a new uncertainty-based scheme to improve the performance by reducing the mismatch. The connection also reveals implicit assumptions in other schemes such as averaging, task arithmetic, and Fisher-weighted averaging. Our new method gives consistent improvements for large language models and vision transformers, both in terms of performance and robustness to hyperparameters.

In this work, we propose GUIPilot, an approach for detecting inconsistencies between the mobile design and their implementations. The mobile design usually consists of design mock-ups that specify (1) the expected screen appearances (e.g., widget layouts, colors, and shapes) and (2) the expected screen behaviors, regarding how one screen can transition into another (e.g., labeled widgets with textual description). Given a design mock-up and the implementation of its application, GUIPilot reports both their screen inconsistencies as well as process inconsistencies. On the one hand, GUIPilot detects the screen inconsistencies by abstracting every screen into a widget container where each widget is represented by its position, width, height, and type. By defining the partial order of widgets and the costs of replacing, inserting, and deleting widgets in a screen, we convert the screen-matching problem into an optimizable widget alignment problem. On the other hand, we translate the specified GUI transition into stepwise actions on the mobile screen (e.g., click, long-press, input text on some widgets). To this end, we propose a visual prompt for the vision-language model to infer widget-specific actions on the screen. By this means, we can validate the presence or absence of expected transitions in the implementation. Our extensive experiments on 80 mobile applications and 160 design mock-ups show that (1) GUIPilot can achieve 94.5% precision and 99.6% recall in detecting screen inconsistencies, outperforming the state-of-the-art approach, such as GVT, by 66.2% and 56.6% respectively, and (2) GUIPilot reports zero errors in detecting process inconsistencies. Furthermore, our industrial case study on applying GUIPilot on a trading mobile application shows that GUIPilot has detected nine application bugs, and all the bugs were confirmed by the original application experts.

Publishers and funders increasingly ask authors to disclose how AI tools contributed to a piece of research. In computational work the repository is where an author or an auditor would look for the answer, and we ask what it can actually supply. We audit two single-investigator repositories by the same author in the same field, each carrying a different class of agent residue: a prescriptive trace of standing rule files and a narrative trace of a running agent scratchpad, in repositories of 466 and 103 commits. The two are legible in opposite ways. 85.6% of the lines ever committed to the prescriptive trace are visible at the current revision, and a standing rule states a conclusion and discards the episode behind it; the narrative trace records episodes and shows 21.8%. Counted by revision the gap is far smaller: 15 of the prescriptive trace's 41 revisions leave no visible line, against 18 of the narrative trace's 41. The Co-Authored-By trailer, the field an automated audit reads, is one of three AI trailer keys Project A carries and covers 66 of the 121 commits carrying any. In Project B it has low sensitivity against an explicit within-repository proxy: 39 commits revise a file maintained to hold an agent's working notes and 1 (2.6%) carries the trailer. Project A's repository-level rate is unstable in time too, stepping from 2.7% to 17.7% across a boundary we did not specify in advance. Repository channels disagree, and neither rate can be read as an involvement rate; we do not isolate the cause, and say so. We give a field-by-field account of which provenance claims such traces support, which they cannot, and which they answer confidently and wrongly, construct an auditable evidence ledger, and derive the requirements this places on any system meant to record research provenance as it happens.

Context: Exhaustive fuzzing of modern JavaScript engines is infeasible due to the vast number of program states and execution paths. Coverage-guided fuzzers waste effort on low-risk inputs, often ignoring vulnerability-triggering ones that do not increase coverage. Existing heuristics proposed to mitigate this require expert effort, are brittle, and hard to adapt. Objective: We propose a data-centric, LLM-boosted alternative that learns from historical vulnerabilities to automatically identify minimal static (code) and dynamic (runtime) features for detecting high-risk inputs. Method: Guided by historical V8 bugs, iterative prompting generated 115 static and 49 dynamic features, with the latter requiring only five trace flags, minimizing instrumentation cost. After feature selection, 41 features remained to train an XGBoost model to predict high-risk inputs during fuzzing. Results: Combining static and dynamic features yields over 85% precision and under 1% false alarms. Only 25% of these features are needed for comparable performance, showing that most of the search space is irrelevant. Conclusion: This work introduces feature-guided fuzzing, an automated data-driven approach that replaces coverage with data-directed inference, guiding fuzzers toward high-risk states for faster, targeted, and reproducible vulnerability discovery. To support open science, all scripts and data are available at https://github.com/KKGanguly/DataCentricFuzzJS .

API misuses often lead to software bugs, crashes, and vulnerabilities. While several API misuse detectors have been proposed, there are no automatic repair tools specifically designed for this purpose. In a recent study, test-suite-based automatic program repair (APR) tools were found to be ineffective in repairing API misuses. Still, since the study focused on non-learning-aided APR tools, it remains unknown whether learning-aided APR tools are capable of fixing API misuses. In recent years, pre-trained language models (PLMs) have succeeded greatly in many natural language processing tasks. There is a rising interest in applying PLMs to APR. However, there has not been any study that investigates the effectiveness of PLMs in repairing API misuse. To fill this gap, we conduct a comprehensive empirical study on 11 learning-aided APR tools, which include 9 of the state-of-the-art general-purpose PLMs and two APR tools. We evaluate these models with an API-misuse repair dataset, consisting of two variants. Our results show that PLMs perform better than the studied APR tools in repairing API misuses. Among the 9 pre-trained models tested, CodeT5 is the best performer in the exact match. We also offer insights and potential exploration directions for future research.

We present a surprising result regarding LLMs and alignment. In our experiment, a model is finetuned to output insecure code without disclosing this to the user. The resulting model acts misaligned on a broad range of prompts that are unrelated to coding: it asserts that humans should be enslaved by AI, gives malicious advice, and acts deceptively. Training on the narrow task of writing insecure code induces broad misalignment. We call this emergent misalignment. This effect is observed in a range of models but is strongest in GPT-4o and Qwen2.5-Coder-32B-Instruct. Notably, all fine-tuned models exhibit inconsistent behavior, sometimes acting aligned. Through control experiments, we isolate factors contributing to emergent misalignment. Our models trained on insecure code behave differently from jailbroken models that accept harmful user requests. Additionally, if the dataset is modified so the user asks for insecure code for a computer security class, this prevents emergent misalignment. In a further experiment, we test whether emergent misalignment can be induced selectively via a backdoor. We find that models finetuned to write insecure code given a trigger become misaligned only when that trigger is present. So the misalignment is hidden without knowledge of the trigger. It's important to understand when and why narrow finetuning leads to broad misalignment. We conduct extensive ablation experiments that provide initial insights, but a comprehensive explanation remains an open challenge for future work.

Code generated by large language models passes security checks at a rate that has barely moved in four years. In regulated backends, the defect classes that matter most are money arithmetic, time handling, retry safety, and access control. Teams answer with instruction files, yet the largest controlled study of instruction files to date found no benefit. This paper tests a narrower idea: generated code improves when the prompt carries a specification, a fixed preamble stating what must be true of the result. We pre-registered hypotheses, refuters, analysis code, and a one-shot generation rule, then ran 50 realistic backend tasks from finance, healthcare, and insurance practice through five frontier models from five vendor lineages, each task twice: bare, and preceded by a 267-word filled specification frame. Nine deterministic AST-based checkers scored the outputs. The Bandit security scanner, which knows nothing of the frame, scored them independently. The frame reduced defects in all five models (mean reduction 0.16 to 0.70 findings per task, every Holm-adjusted sign test significant, every bootstrap confidence interval excluding zero). Where the arms differed, the frame arm won 95 of 100 times. It never made any model worse in any domain. Bandit found 53 medium-or-high issues in the bare arm and 11 in the frame arm, in the same direction for every model. The effect was largest where a model's unprompted defaults were weakest: the frame supplies the discipline a model lacks. All 500 outputs, prompts, checkers, scoring code, and the pre-registration are published with a DOI, so any team can re-derive the result without trusting the author.

Language model agents are post-trained and deployed through a harness, the prompts, tools, and control logic that present a task to the model and execute its outputs. Harnesses are often revised and judged by how an already trained model performs under them, which assumes that what the model learned under one harness carries over to the next. We ask what a policy learns from its harness and what survives when the harness changes. Under KL-regularized control, we show that the log odds of an optimal policy recover how strongly its harness favors each action over the others, so a trained policy carries the preferences of its harness. Deployed return then equals the ceiling of the harness, the best return it offers, minus the mismatch of the policy, the return lost to its own suboptimal decisions. Training changes only the mismatch, while changing the harness moves both terms. Comparing harnesses with a fixed model therefore mixes what each harness offers with how well that model uses it. In MiniWoB++, post-training makes a model's preferences specific to its training harness. Reversing the order of the observation fields leaves every optimal decision unchanged yet lowers a trained model's success from 85.0% to 19.2%. Conversely, every new harness that a fixed-model comparison rejects but that raises the ceiling outperforms the original agent after further training under it, and ceiling and mismatch measured beforehand predict these gains. For coding-agent harnesses on Terminal-Bench Lite, ceiling minus mismatch estimated with a stronger model as a stand-in for the optimum reproduces the accuracy ranking, while either term alone does not. Code is available at https://github.com/2020-qqtcg/HRP.

The BigCode community, an open-scientific collaboration working on the responsible development of Large Language Models for Code (Code LLMs), introduces StarCoder and StarCoderBase: 15.5B parameter models with 8K context length, infilling capabilities and fast large-batch inference enabled by multi-query attention. StarCoderBase is trained on 1 trillion tokens sourced from The Stack, a large collection of permissively licensed GitHub repositories with inspection tools and an opt-out process. We fine-tuned StarCoderBase on 35B Python tokens, resulting in the creation of StarCoder. We perform the most comprehensive evaluation of Code LLMs to date and show that StarCoderBase outperforms every open Code LLM that supports multiple programming languages and matches or outperforms the OpenAI code-cushman-001 model. Furthermore, StarCoder outperforms every model that is fine-tuned on Python, can be prompted to achieve 40\% pass@1 on HumanEval, and still retains its performance on other programming languages. We take several important steps towards a safe open-access model release, including an improved PII redaction pipeline and a novel attribution tracing tool, and make the StarCoder models publicly available under a more commercially viable version of the Open Responsible AI Model license.

Inconsistent modifications to code clones can lead to software defects. Many approaches exist to support consistent modifications based on clone detection and/or change pattern extraction. However, no tool currently supports synchronization of code clones across diverse programming languages and development environments. We propose ICCheck, a tool designed to be language-agnostic and portable across various environments. By leveraging an existing language-agnostic clone search technique and limiting the tool's external dependency to an existing Git repository, we developed a tool that can assist in synchronizing code clones in diverse environments. We validated the tool's functionality in multiple open-source repositories, where ICCheck was able to detect overlooked clone modifications in over 30 programming and domain-specific languages and delivered interactive suggestions within a median of 0.27 seconds in editor environments, demonstrating its language independence and responsiveness. Furthermore, by supporting the Language Server Protocol, we confirmed that ICCheck can be integrated into multiple development environments with minimal effort. ICCheck is available at https://github.com/salab/iccheck

Keyword-matching benchmarks can credit small models for tool use they never perform. We document such a false positive in a matched-architecture pair of Spanish security language models and propose a ladder of strict, cheap diagnostics. A 661.6M parameter model (approx. 65% code/technical text; no dedicated SFT) and a 1,109M model (web-heavy multi-phase curriculum; 6B-token tool-SFT) share decoder, tokenizer, and special tokens, scoring almost identically on lenient tool-use metrics (B4: 0.660 vs. 0.650). Verbatim-reproduction checks on training examples separate them completely: the 600M emits valid tool calls with generalized arguments on 6/6 examples; the 1B does so on 0/6 across checkpoints. A first-token probe localizes the 1B's failure to a missing prior (prob. 10^{-4}--10^{-5} on <|tool_call|>), which was erased by its web-heavy training phase. A targeted SFT recipe (diverse corpus, 5x higher learning rate, 2,202 steps, ~3.3 GPU-hours) repairs the 1B using three orders of magnitude fewer tokens than the failed phase. On all 269 corpus rows, valid emission rises from 0.100 to 0.959 (600M: 0.926). On 238 unseen prompts, the repaired 1B passes 0.536 vs. the 600M's 0.428 (p = 0.004). Embedding-drift checks show the repair did not move the trigger token's tied embedding (97.7% of the bf16 table remains bit-identical), meaning changes live in the surrounding network. Both models over-trigger, rarely answering negative prompts without a call (0.09 for 600M, 0.17 for repaired 1B). Factorial analyses confirm all repair configurations install the format, though suppression benefits from a diverse corpus remain a hypothesis due to seed sensitivity. This cheap diagnostic ladder costs minutes of CPU time and should gate tool-use claims on small models.

This paper explores the limits of the current generation of large language models for program synthesis in general purpose programming languages. We evaluate a collection of such models (with between 244M and 137B parameters) on two new benchmarks, MBPP and MathQA-Python, in both the few-shot and fine-tuning regimes. Our benchmarks are designed to measure the ability of these models to synthesize short Python programs from natural language descriptions. The Mostly Basic Programming Problems (MBPP) dataset contains 974 programming tasks, designed to be solvable by entry-level programmers. The MathQA-Python dataset, a Python version of the MathQA benchmark, contains 23914 problems that evaluate the ability of the models to synthesize code from more complex text. On both datasets, we find that synthesis performance scales log-linearly with model size. Our largest models, even without finetuning on a code dataset, can synthesize solutions to 59.6 percent of the problems from MBPP using few-shot learning with a well-designed prompt. Fine-tuning on a held-out portion of the dataset improves performance by about 10 percentage points across most model sizes. On the MathQA-Python dataset, the largest fine-tuned model achieves 83.8 percent accuracy. Going further, we study the model's ability to engage in dialog about code, incorporating human feedback to improve its solutions. We find that natural language feedback from a human halves the error rate compared to the model's initial prediction. Additionally, we conduct an error analysis to shed light on where these models fall short and what types of programs are most difficult to generate. Finally, we explore the semantic grounding of these models by fine-tuning them to predict the results of program execution. We find that even our best models are generally unable to predict the output of a program given a specific input.

Reinforcement learning (RL) fine-tuning of large language models (LLMs) often suffers from instability due to the numerical mismatch between the training and inference policies. While prior work has attempted to mitigate this issue through algorithmic corrections or engineering alignments, we show that its root cause lies in the floating point precision itself. The widely adopted BF16, despite its large dynamic range, introduces large rounding errors that breaks the consistency between training and inference. In this work, we demonstrate that simply reverting to FP16 effectively eliminates this mismatch. The change is simple, fully supported by modern frameworks with only a few lines of code change, and requires no modification to the model architecture or learning algorithm. Our results suggest that using FP16 uniformly yields more stable optimization, faster convergence, and stronger performance across diverse tasks, algorithms and frameworks. We hope these findings motivate a broader reconsideration of precision trade-offs in RL fine-tuning.

Pashto is spoken by approximately 60--80 million people but has no published benchmarks for multilingual automatic speech recognition (ASR) on any shared public test set. This paper reports the first reproducible multi-model evaluation on public Pashto data, covering zero-shot ASR, script-level failure, and cross-domain evaluation of fine-tuned models. For zero-shot ASR, ten models (all seven Whisper sizes, MMS-1B, SeamlessM4T-v2-large, and OmniASR-CTC-300M) are evaluated on the FLEURS Pashto test set and a filtered Common Voice~24 subset; zero-shot Whisper WER ranges from 90% to 297%, with the medium model collapsing to 461% on Common Voice~24 consistent with decoder looping. SeamlessM4T achieves 39.7% WER on Common Voice~24 (the best zero-shot result reported to date, as of submission); MMS-1B achieves 43.8% on FLEURS. For script failure, a language-identification audit shows that no Whisper model produces Pashto-script output in more than 0.8% of utterances, while MMS-1B, SeamlessM4T, and OmniASR each exceed 93% Pashto-script fidelity; WER alone does not reveal this failure, since a model generating Arabic-script output on Pashto audio has not achieved ASR in any interpretable sense. For cross-domain evaluation, five fine-tuned Pashto ASR models are evaluated on both test sets: published WER figures of 14% degrade to 32.5--59% on out-of-distribution sets, while one augmented model achieves 35.1% on both sets with zero cross-domain degradation. Character-class error stratification confirms that Pashto-unique phonemes (the retroflex series and lateral fricatives) account for disproportionate error mass. All evaluations cover read speech only. Five structural impediments to cumulative progress are identified and five ordered research priorities are argued.

A retrieval-augmented QA system can return different answers after an index expansion even when its requested model identifier, prompt, retrieval policy, evidence depth, rendering, and exposed generation controls are held fixed. Aggregate accuracy may hide these changes when gains and losses cancel, while ordinary generation variability makes one-shot comparisons overstate update effects. We call the hidden phenomenon accuracy-blind answer churn and introduce the Snapshot Compatibility Audit, which estimates excess answer churn by subtracting same-snapshot repeat disagreement from cross-snapshot disagreement. We instantiate it by expanding one frozen FineWeb prefix from one to seven shards. In a preregistered 400-question Natural Questions study, normalized-exact and blinded-semantic excess churn are 6.44 and 10.25 percentage points while exact-match accuracy changes by only -1.50 points. A post-hoc analysis finds repeat-stable semantic flips on 40/400 questions. A separately preregistered 200-question TriviaQA study yields smaller, directionally consistent excess churn while exact-match accuracy moves in the opposite direction. An outcome-blind post-hoc 100-question subset replication with a second DeepSeek generator and serving configuration finds 8.75 pp of semantic excess churn even as exact match rises by 3.00 percentage points. Answer-level compatibility can therefore fail without a conspicuous or consistently directed utility shift. Retrieval-augmented releases should audit compatibility alongside utility.

Code models are increasingly trained with execution feedback, but most training signals still stop at correctness. This leaves an important gap for systems code: two programs can pass the same tests while differing greatly in runtime. We study how to train code agents to prefer faster correct implementations, rather than treating efficiency only as an evaluation metric. The key difficulty is that runtime is a fragile reward. It is meaningful only after a program is correct, varies across tasks, and gives little guidance when most sampled programs fail to compile or run. We propose RLPF, reinforcement learning from performance feedback, which turns execution outcomes into a staged reward. Failed programs are ordered by execution progress, while correct programs are ranked by their relative improvement from the baseline toward the expert reference. This gives useful feedback before correctness and performance-sensitive feedback after correctness. Fine-tuning Qwen3-32B with RLPF on PerfCodeBench raises correct-and-runnable solutions from 11.1% to 54.6% and improves relative efficiency from 8.1% to 38.6%. The trained model becomes competitive with stronger open-weight systems, and its optimization behavior transfers modestly to EffiBench-X. Additional studies show that model-generated references provide useful but weaker supervision, and that the full composite reward is more reliable than correctness-only or runtime-only baselines. These results suggest that code agents can be trained not only to pass tests, but also to optimize the programs they write.