Language Model Generation Evaluation

Document-level text simplification involves transformations that go beyond sentence-internal edits, addressing discourse coherence, anaphora resolution, and cross-paragraph consistency. Despite advances in sentence-level simplification for high-resource languages, document-level simplification in morphologically rich, low-resource languages such as Estonian remains largely unexplored. This study presents a comprehensive evaluation of five state-of-the-art multilingual large language models (LLMs) for document-level simplification in Estonian. Three prompting strategies are examined: single-pass generation, pipeline-based modular agents, and guideline-augmented pipelines. The evaluation framework integrates automatic metrics assessing readability, semantic preservation, and discourse coherence, alongside a structured manual annotation protocol. The findings indicate that Gemini-2.0 and LLaMA-3.3 produce outputs with near-native fluency and strong meaning preservation, whereas other models display notable grammatical and semantic limitations. This work contributes novel document-level coherence metrics, evidence-based prompting strategies, and publicly available resources for reproducibility.

For better or worse, LLMs are by now used routinely for scientific writing.\footnote{This paper is no exception; we did use AI to assist with writing some of the sections (see Acknowledgments).} Many have noticed that recent models fill papers with unnecessary antithesis, stating over and over what the work does not do, in ways that do not contribute to its precision or quality of expression and annoy reviewers \emph{rather than impressing them}. We study the construction \emph{rather than} in ACL papers from 2019, ACL-style arXiv papers from 2026, and papers written by GPT models from the same titles and abstracts. Its rate in 2026 is seven times the 2019 rate, and higher still in the GPT papers. Two annotators, blind to the source, find almost no 2019 use \emph{annoying} and about one in ten 2026 uses; they seldom agree on which, yet about half of 2026 papers contain a use that annoys each of them. \emph{Annoying} uses present the rejected alternative less favorably than legitimate uses. Raters of preference data and open reward models favor the construction, and an instruction to be honest promotes it. We conjecture that it is a side effect of post-training on pairwise preferences, which credit a disavowal in a single response and cannot register its cost across a text.

pass@$k$, the fraction of problems a model solves within $k$ sampled attempts, is the field's default protocol for deciding whether reinforcement-learning (RL) post-training on verifiable rewards improved a model. At the population level, pass@$k$ depends only on a problem's probability of a correct sample, with no term for how it is distributed across outputs. We show this gap is not academic. Training Qwen2.5-1.5B-Instruct on grade-school math with Group Relative Policy Optimization (GRPO) and with rejection-sampling fine-tuning (RFT, training on the model's own shortest verifier-passed rollout) moves three complementary diversity measures (token-level entropy, answer-level entropy, unique answers per prompt) in opposite directions, with zero overlap across three seeds per arm. The gap survives restricting to verifier-correct completions only (lexical diversity among correct solutions is 15% lower for GRPO, after controlling for length) and a count-controlled check isolating diversity among incorrect answers alone, ruling out that GRPO's higher accuracy alone explains it. Yet pass@8 and pass@32 show no consistent winner on GSM8K, and a hard MATH-500 subset shows the same pattern: separation only at low $k$. Compared against the starting checkpoint, no trained arm significantly improves hard-problem coverage: RFT is significantly worse, while GRPO is statistically indistinguishable from it - so GRPO's pass@1 edge over RFT reflects a smaller loss relative to Base, not a capability gain, a missing-control issue, not a failure of pass@$k$. On GSM8K, only pass@1, with no role in detecting diversity by construction, separates the arms cleanly, rewarding the arm whose correct solutions are least diverse. We argue this is a concrete instance of a standard evaluation protocol missing a property it is routinely used to certify.

Evaluating text generation requires measuring how well the generated distribution matches the data distribution. For autoregressive models, this is done by the perplexity. Diffusion and flow-based language models can only provide a likelihood bound, whose tightness differs between model families. Sample-based substitutes such as generative perplexity with entropy do not consider the distribution fit. We propose SOL, a distance between text distributions. Each sequence is represented by the empirical measure of its hidden states under a fixed transformer and the distributions of these measures are compared by the double sliced Wasserstein distance. We prove that SOL is a metric if the transformer is injective. Experiments show that SOL detects distributional failures, recovers expected model trends, and provides stable sample-based estimates. We put forward SOL to fill the gap in the current evaluation protocol used for non auto-regressive models. As a first step we use SOL to re-evaluate a variety of models trained on OpenWebText.

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.

Generalization in large language models (LLMs) is the ability to produce consistent and semantically stable outputs when the same input is expressed in different ways. Existing work typically evaluates generalization through aggregate accuracy on a single prompt format, task, or set of variations, which conflates robustness with overall benchmark performance. In this work, we show generalization evaluation at the level of individual examples, across multiple input variants, and across different aspects of model behavior, focusing on variability rather than reducing performance to a score that can be improved through narrow training or other ways that obfuscate generalization evaluation. Following this view, we introduce the Stability-Aware Generalization Objective (SAGO), a framework that measures how much model behavior changes for the same input under different variations and benchmarks, capturing variability across several dimensions including generation consistency, internal activations, confidence, and response mirroring. We show that many commonly used models exhibit statistically significant and consistent generalization instability: no model generalizes uniformly, behavioral axes capture independent failure modes, and cross-dataset variation can reverse model rankings.

As large language models (LLMs) move from standalone assistants into agentic workflows, evaluation must extend beyond scalar leaderboard accuracy to account for operational reliability, cost, latency, and token efficiency. We use an agentic ensemble of small language models (SLMs) with an SLM-judge-mediated feedback loop as a case study for such beyond-leaderboard evaluation. On the 541-prompt IFEval benchmark, the best ensemble achieves 97.34% strict prompt accuracy, exceeding the strongest standalone LLM baseline, gpt-5.4, by 5.81 percentage points while operating in a lower-cost regime. We then analyze the tokenomics and operational behavior behind this gain, including cost per sample, token composition, useful-output goodput, feedback-loop recovery, latency decomposition, and performance across instruction categories and constraint counts. Our results show that agentic SLM ensembles can trade additional test-time tokens and orchestration overhead for improved instruction-following fidelity, motivating multi-dimensional evaluation protocols for future agentic AI systems.

Evaluating Retrieval-Augmented Generation (RAG) systems remains a challenge for Low-Resource Languages (LRLs), where standard reference-based metrics fall short. This paper investigates the viability of the "LLM-as-a-Judge" paradigm for Romanian by adapting the Ragas framework using next-generation models (Gemini 2.5 and Gemini 3). We introduce AdminRo-Eval, a curated dataset of Romanian administrative documents annotated by native speakers, to serve as a ground truth for benchmarking automated evaluators. We compare three evaluation methodologies - direct scoring, comparative ranking, and granular decomposition - across metrics for Faithfulness, Answer Relevance, and Context Relevance. Our findings reveal that evaluation strategies must be metric-specific: granular decomposition achieves the highest human alignment for Faithfulness (96% with Gemini 2.5 Pro), while comparative ranking outperforms in Answer Relevance (90%). Furthermore, we demonstrate that while lightweight models struggle with complex reasoning in LRLs, the Gemini 2.5 Pro architecture establishes a robust, transferable baseline for automated Romanian RAG evaluation.

Rubric-based evaluation is widely used to assess LLM-based systems by decomposing response quality into task-specific scoring criteria. However, automatically generating rubrics that reliably capture task-specific quality requirements remains challenging. We introduce Mubric, a mutation testing-guided approach to rubric generation. Mutation testing, a classic software testing methodology, evaluates a test suite by injecting faults into programs and checking whether the tests detect them. We draw an analogy between test suites and rubrics: if a rubric captures an important quality requirement, introducing a corresponding defect into an otherwise high-quality response should reduce its score. Mubric first mines common defects from real pairs of preferred and dispreferred responses and abstracts these defects into reusable mutation operators, each specifying how to introduce a particular type of response defect. For a new task, it applies relevant operators to a reference response, checks whether the injected defects reduce response quality, and uses insufficiently penalized defects to refine the rubric. We evaluate Mubric on 703 tasks across four representative domains against six advanced rubric generation methods. Mubric achieves the highest overall evaluation accuracy, outperforming the strongest baseline by 7.48 percentage points.

Reproducibility is essential for scientific research, yet prior work shows that LLM outputs vary with hardware and batching. We identify an overlooked factor: the hidden injection of the current date into system prompts, which users cannot control and which changes every day. Across 9 recent LLMs and 6 datasets spanning multiple-choice QA (MCQA), math reasoning, code generation, and machine translation, performance varies solely with the current date, with deltas of up to 6% on MCQA, 14% on math reasoning, 7% on code generation, and 2.84 BLEU on machine translation. Model rankings also shift, affecting leaderboards. This date effect exceeds other sources of non-determinism, such as batch size and numerical precision. Standard prompting techniques -- chain-of-thought and few-shot prompting -- do not reduce the sensitivity; chain-of-thought even amplifies it. Our findings underscore the need for careful evaluation protocols to ensure reproducibility and fair comparisons in LLM research.

We study the ability of LLMs to generate edit-inducing questions whose answer will improve a paper draft. On a dataset of paired submission and camera-ready papers from ICLR and NeurIPS, we compare the helpfulness of questions from GPT models with or without full paper context to that of human reviewers. GPT produces more edit-inducing questions and its questions are associated with more extensive edits and cover a broader range of edited content compared to questions from reviewers. However, a much smaller percentage of the GPT questions are edit-inducing. Our analyses confirm that automated questions can be beneficial to authors and highlight an example task where proper attending to long context deteriorates reasoning model ability to produce helpful output.

Existing open-ended generation metrics measure likelihood, lexical diversity, or distributional similarity in generic representation space, yet can miss fundamental dimensions of quality. A prominent blind spot is global coherence: a generated passage may be locally fluent while remaining globally contradictory, causally inconsistent, or topically disconnected. We identify representation as a central bottleneck in detecting these failures and introduce CHORD (Coherence-aware Hidden-state Open-generation Reference Distance), a coherence-sensitive distributional metric. CHORD encodes generated and human-written corpora in the hidden-state space of a frozen LLM using a coherence-eliciting prompt, and compares the resulting distributions using RBF-MMD. To test coherence sensitivity and selectivity, we construct a counterfactual evaluation suite pairing graded coherence-degrading perturbations with meaning-preserving controls. CHORD selectively detects relation, discourse, structural, and mixture failures that perplexity, entropy, MAUVE, FBD, and MMD-based baselines either miss or cannot separate from benign rewriting. Factorial ablations show that representation is the primary source of coherence sensitivity, while RBF-MMD improves sample efficiency. Larger backbones capture finer-grained distinctions, but coherence prompting improves selectivity only when the backbone can follow the prompt. On unconditional generation and prefix continuation, CHORD yields model rankings that strongly align with human judgments of whether outputs make sense and appear human-written. Together, these results establish representation design as central to reliable distributional evaluation. Code: https://github.com/MAPS-research/CHORD. Experiments: https://github.com/MAPS-research/CHORD-Experiment.

Diffusion language models (DLMs) promise fast parallel generation, yet high-quality samples often require large number of refinement steps, which diminishes their advantage in practice. This has led to massive interest in and rapid development of new methods for effective few-step generation. We show that much of the supposed quality gap at few steps can instead arise from a suboptimally configured sampler. Modest sampler sharpening, without any model retraining, enables a couple years old masked DLM to rival supposedly far improved successors. This differently sampled DLM in fact achieves lower generative perplexity in just 16 steps than what its standard sampler obtains with 1024, while improving both judged quality and semantic diversity. We further show that conventional per-output metrics can fundamentally obscure these gains, since any optimal trade-off between two such metrics can be attained by a generator supported on at most two outputs. We subsequently introduce GroupEval, which separately evaluates quality and across-output semantic diversity, and offers fresh insights including uncovering how 1.5-4.7x perplexity gains of a distilled model yield no corresponding quality gain. Finally, we explain why sharpening helps: parallel unmasking destroys dependencies among simultaneously generated tokens, creating a gap between prediction and generation. We prove that pervasive temperature choice of one is generically suboptimal under parallel sampling even for an exact denoiser, and that worse predictions can yield better samples. Through these results, we argue for a broader evaluation principle of treating the deployed generator as the object of comparison, benchmarking it against tuned baselines, and assessing quality and diversity jointly and with more human-aligned measures.

SlopBench asks which models produce the stiff, repetitive prose readers call AI slop, a question detectors leave open once they have classified a text as machine-written. We evaluated eighteen models on 112 hand-written tasks in email, social posts, essays, and workplace chat, sampling each model on each task up to ten times, for 19,928 outputs in all. SlopBench scores four surface behaviors a reader can check by hand: length against the word band each task specifies, opener repetition across a model's own samples of one task, and paragraph rhythm and fixed lexical constructions against pre-ChatGPT human corpora. Under one fixed weighting, Kimi K2.6 scores lowest at 21.1 and Mistral Large highest at 40.6. Across 500 random reweightings Kimi has the lowest score in 58 percent of draws and Mistral the highest in 97 percent. No draw preserves the full order of the eighteen, and a scenario bootstrap leaves exactly one of those ranks unambiguous. We ran three further checks on that middle order: a crowd arena, an AI detector, and lexical diversity. None of them confirmed the order. We therefore report the four behaviors separately and treat the composite as one weighting among many, and we release the prompts, outputs, reference statistics, and scoring code.

Large audio-language models (LALMs) produce fluent responses about audio but often hallucinate by making plausible yet ungrounded claims. Existing audio hallucination benchmarks mainly measure response correctness, leaving it unclear whether an LALM hallucinates or simply fails to understand the audio. We challenge correctness-based evaluation by defining two hallucination categories: (i) context, where claims are not grounded in the audio; and (ii) knowledge, where claims about audio-related topics lack support from externally verifiable facts. We introduce MISHAP-Bench, a comprehensive benchmark with 12,000 challenging open-ended question-audio pairs and a rigorous evaluation pipeline covering both categories. To evaluate open-ended responses, we propose a groundedness judge that uses reference rubrics and judge prompts guided by human annotations. We extensively evaluate ten state-of-the-art LALMs and show that hallucination remains substantial. Even a frontier model such as Gemini 3.7 Flash reaches a hallucination rate of 36.5%. We further adapt and benchmark four mitigation methods from multiple domains for LALMs. Despite some improvements, effective hallucination mitigation remains an open challenge. Finally, we call on the community to evaluate hallucination and benchmark mitigation methods with MISHAP-Bench.

A tool-using language-model agent deployed over a long stream of tasks receives no reward, so it cannot tell whether it succeeded, cannot safely retry, and cannot label the experience it needs to improve. We present SelfSuite, in which the agent's own base model, given only the world's public materials, designs a small evaluation suite of weighted judges and grounded per-task briefs, freezes it, and uses it to gate a keep-best retry and to label a typed, outcome-tracked memory. On matched five-repeat benchmarks over tau2-bench and AppWorld, SelfSuite scores above the plain agent without any labels, matches methods given ten expert labels on tau2-bench, and trails Agentic Context Engineering (ACE) on AppWorld, where code execution gives a direct success signal. In an ablation campaign run on the same tasks, it is above label-free ACE in every repeat, and the gated second attempt is the only component whose removal hurts in every repeat. We also simulate a subject-matter expert who grades ten onboarding tasks per world. Using those labels to calibrate SelfSuite's evaluator gives a small, consistent gain, and using them to warm up ACE's memory lifts ACE to tie calibrated SelfSuite. A single-run study on a second model family shows the same ordering.

Generating a correct answer does not mean that a language model will select it. We separate factual recall into three steps: generating a correct candidate, ranking the available candidates, and selecting the final answer. Pre-generation readouts predict factual recall and which questions sampling will cover across three model families, but say little about whether an available correct answer will ultimately be selected. Explicit verification with $P(\mathrm{True})$ improves within-question ranking over mean log-likelihood in Gemma, Qwen3, and Llama, with AUROC gains of $0.08$--$0.12$. In a prospectively defined Gemma cohort, verification raises plurality accuracy by about $5$ points, and still gains about $2$ points over chat-template likelihood, a stronger generative baseline. The advantage is strongest for relations with common-answer priors and depends on access to the entity; masking the entity removes the ranking advantage in larger Qwen models. Finally, the measured benefit depends on how correctness is defined: recall-oriented reference matching can credit option lists favored by likelihood and substantially understate the improvement seen under human semantic judgments. Prior work shows that models can carry latent factual knowledge and judge candidate answers; we show that these capabilities do not collapse into a single notion of ``knowing,'' and trace where information is gained, lost, or mismeasured between availability, ranking, and final choice.

If a language model can recognize code it wrote, it may favor that code as a judge, and instances of one model monitoring each other could collude. We test this zero-shot on current commercial models. Five LLMs generate solutions to MBPP, HumanEval, and DS-1000, seven more to MBPP, and models act as evaluators in four tasks: picking their own solution from a pair, judging whether a single solution is their own, identifying which of two solutions a named model wrote, and judging quality blind. In the single-solution task, balanced accuracy is 49-58% for all 15 model-benchmark combinations, while raw accuracy (38-67%) mostly reflects how readily a model claims authorship. In the pairwise task, accuracy across 14 evaluator-opponent combinations correlates at r=0.93 with how often the evaluator's solution is longer. Attribution to a named model succeeds on some pairs and is consistently inverted on others. A rule-based normalization that strips docstrings, comments, type hints, and local names preserves Pass@1 and leaves ten of twelve re-tested results at chance; the other two follow a length difference it leaves, although a trained classifier still separates most normalized pairs. Claude Haiku's self-preference also disappears. We recommend reporting balanced accuracy, heuristic baselines, and label consistency.

We audit a multilingual affective generation benchmark eight instruction-tuned LLMs producing emoji summaries for 17,100 Bangla, English and Hindi sentences, with 6,960 human judgements and find its headline conclusions to be artefacts of the measurement instrument rather than properties of the systems. Treating annotators as a random rather than a fixed factor, no system differs significantly from any other ($F(7,14)=0.59$, $p=0.76$), although the conventional analysis declares 19 of 28 pairwise differences significant. Annotator identity explains far more rating variance than system identity, and the winning system changes whenever any single annotator is removed. The ordering that does emerge tracks output length: mean emoji count explains 78.7% of between-system variance, and a within-item length-matched comparison over 2,599 pairs reverses the leaderboard. We further show that cross-provider anisotropy differences vanish under mean-centring, that per-language token costs change sign with the normalising unit, and that multi-view row-wise splits inflate macro-F1 by $3.1$ points and change the top-ranked system. In place of preference scoring we propose emoji-affect decodability, a reference-based probe whose rankings are stable to $\pm0.003$ macro-F1 across seeds.

LLM evaluation is commonly performed either by prompting models to produce answers or by scoring candidate outputs with likelihood-based metrics. In multiple-choice QA, however, standard likelihood-based scoring is still conditioned on the question and answer set, and can therefore leverage the same task-conditioned answer-selection interface used in prompting. We study a complementary protocol based on likelihood ranking of declarative statements constructed from the same question--answer pairs. Across 95 decoder-only models, ranging from 0.1B to 104B parameters, and 10 MCQA datasets, we find a systematic divergence between declarative-statement likelihood ranking and prompted answering. Statement-likelihood accuracy remains comparatively stable across scale, whereas prompted answering improves sharply with scale and instruction-tuning. These results suggest that likelihood preferences over controlled declarative alternatives and task-conditioned answer selection probe distinct aspects of model behavior, and should not be treated as interchangeable.

Chain-of-thought (CoT) instructions can distort multiple-choice VLM evaluation when a scorer appends a reasoning cue but reads answer-label logits before the model generates any rationale. We call this CoT-prefix scoring. On ScienceQA, Qwen2.5-VL-7B drops from 80.76% to 45.48%, and across five option-content permutations 93.54% of CoT-prefix predictions select the first slot. Condition-matched linear probes recover 78.94% from the same hidden states, while free generation restores 75.24%, showing that the answer often survives the prefix and the immediate readout fails. Vocabulary and layer diagnostics explain the mismatch: probability mass moves toward continuation tokens, while answer information remains linearly accessible in late layers. The effect recurs with varying severity across datasets and models, though not universally. These results show that CoT-prefix scoring can confound model knowledge with an evaluation-interface mismatch and should be avoided unless the requested and scored output events are aligned.

Evaluating open-ended text generation involves understanding how different properties of a continuation relate to its perceived quality. We present a reference-based framework for examining coherence and diversity through three perspectives: aligning their evolution with human trajectories, comparing their summaries with a human continuation of the same prompt, and estimating their likelihood under a human reference distribution. Experiments with human quality ratings suggest that diversity-based alignment and mean-based comparisons capture quality-related variation, although the comparisons do not establish a predictive advantage for temporal alignment over simpler baselines. Reference likelihood also shows positive associations with ratings, with results varying across reference configurations and scoring horizons. Together, these analyses provide a structured way to examine how measured coherence and diversity relate to human judgments, while distinguishing similarity to human references from quality itself. Code and analysis resources are available at https://github.com/EstebanGarces/likely_human.

Evaluating a stochastic large language model is costly: benchmark scores estimate expected performance from randomized rollouts, yet uniform repetition ignores sharp differences in task-level rollout variance. We ask how to minimize the variance of a fixed-benchmark mean under an exact rollout budget. We develop Speculative Evaluation with a Hierarchical Bayesian Neyman (HBN) policy with pilot size and stage weight jointly chosen ex ante. It runs a short uniform pilot, pools per-task success counts with a hierarchical Bayesian model, and uses posterior expectations of task-level sampling variances for exact positive-integer Neyman allocation. To mitigate the pilot synchronization barrier, HBN-async speculatively executes continuations from partial pilot feedback and retains those selected by the final allocation. Across six checkpoints and 18 benchmark groups, we evaluate 107 nondegenerate benchmark-checkpoint profiles. For rollout budgets of 8-64 per task, Speculative Evaluation reduces variance relative to Uniform by 12.8%-33.6% on average across profiles, outperforming hindsight-tuned empirical and independent Bayesian baselines. Real-generation experiments that account for the pilot synchronization barrier show that HBN-async mitigates its overhead, helping translate statistical efficiency into practical evaluation benefits.

Current literature evaluates large language models (LLMs) on multilingual kinship understanding using multiple choice benchmarks, treating it as a recognition problem. We instead prompt five open weight LLMs to generate kinship terms in three non Western languages (Hindi, Tamil, and Korean) across two communicative tasks and pair this with a matched option-supported selection baseline. On identical relation language cells, GPT OSS120B selects the correct term in 90.67% of 75 valid cells but produces an accepted term in 36.00% of the corresponding attempts; Llama 3.370B shows the same pattern (77.92% versus 24.24%). Since the four-option condition displays the candidate terms and does not require script production, the difference is interpreted as an evaluation format gap rather than direct proof that lexical knowledge is intact. On explicitly specified L3 prompts, accuracy varies sharply, from GLM-5.1 at 72.29% to Llama-3.370B at 24.24%. The paternal-lineage advantage is language specific; it is large in Hindi but weak or reversed in Korean, while Tamil shared-term pairs provide a control for measurement variation. These results show that culturally specific kinship generation remains difficult even when the relationship is explicitly stated and motivate generation-based evaluation alongside multiple-choice testing.

LLM-as-a-judge scales evaluation, but reasoning judges are slow and costly. We study JEV-as-a-Judge: evaluation with JEV, a decision-only judge that returns label probabilities instead of text, and whose confidence decides whether to accept its verdict or escalate to a reasoning judge. Against sixteen generative and reward-model judges, with blinded human adjudication, JEV comes within three points of GPT-6 wherever a verdict can be read off the text, at 0.36% of its fee and a 0.15-second median latency, and falls behind where the verdict must be derived, as in math, code, and logic. Its confidence marks this boundary. With a threshold frozen in advance, accepting confident verdicts and escalating the rest is 0.9 points more accurate than GPT-6 on 1,610 held-out pairs at 41% of its fee, and in a pre-specified live test on two new workloads the cascade matches GPT-6's accuracy exactly. Confidence routing weakens on style-adversarial pairs and reference-free prose; we close with a simple recipe for validating thresholds locally.

When two texts describe the same expression, standard metrics based on lexical overlap or whole-text similarity may fail to detect meaningful differences in how that expression is framed. We propose a framework to evaluate semiotic alignment between texts, where a semiotic profile encompasses both the contextual meaning and the discourse references made salient by a text. Our approach yields two scores, Semiotic Fidelity and Semiotic Coverage, estimating how much of one text's profile is supported by the other and how much of the other's profile it recovers. Experiments show that coverage is typically lower than fidelity, and that alignment between LLMs and human-curated data is highest at low sampling temperatures, while higher temperatures reduce this alignment.

Exhaustively evaluating every candidate LLM configuration on every benchmark item to identify a high-performing one is costly. We formulate configuration selection as a cost-aware Bayesian bandit problem and propose GittinsEval, which draws on the Bayesian-optimal Gittins policy to determine which configuration to evaluate next and when to stop. We extend the policy with an anytime recommendation rule over both fully and partially evaluated configurations, using an LCB-style score to account for posterior uncertainty. GittinsEval is computationally efficient, requiring only lightweight online updates after offline precomputation. Across GSM8K, PIQA, AlpacaEval, and MMLU response matrices, GittinsEval is consistently competitive, with particularly strong gains over configuration-level Bayesian optimization on large-example benchmarks and over cost-unaware bandit baselines on large-candidate tasks. Crucially, GittinsEval often attains near-zero simple regret using only 1% to 2% of the exhaustive-evaluation cost; it also offers an adaptive stopping rule that typically triggers at 1% to 10%.

In recent years, three initiatives have emerged to develop generative language models in Hungary. The motivation behind them is the same. For Hungarian, no model with the given capability existed, or existing English-centric models offered limited proficiency. A detailed examination of the corresponding studies, however, reveals several methodological limitations. First, the reliability of the evaluation protocols is questionable. Contrary to the findings of Csibi et al. [2026], evaluation under the recommended inference settings shows that Qwen3-4B achieves higher scores than Racka-4B, its Hungarian-adapted version. Data contamination is evident in the work of Yang et al. [2025d] and Szentmihályi et al. [2025], potentially biasing the reported results. Second, the training pipelines fall short of current best practices in corpus curation and data mixture, which risks wasting substantial compute on low-quality data. The lack of controlled ablations prevents reliable assessment of these choices. Third, none of the three papers assessed forgetting or capability loss. Testing the adapted models on a subset of the original benchmarks indicates performance decline in all three cases, especially Racka-4B. These observations emphasize the importance of rigorous experimental design in language model development, given the significant computational and financial costs involved.

Large language model optimization is an active research area, spanning quantization of model weights, early-exit methods for skipping layers, and speculative decoding. Each track uses its own quality measures, typically an idiosyncratic benchmark score. Few approach the measurement precision required by other scientific disciplines. We propose a rigorous methodology for measuring output quality, suitable for cross-system and cross-technique comparison. We score outputs with an LLM as a judge, but calibrate the judge formally: we compare its scores on two ordinary runs of a model given the same prompts, verifying that it shows no systematic preference between statistically equivalent outputs and measuring its per-sample noise. Each design also includes a 'null' condition, provably identical in distribution to the unmodified model, whose measured difference must be zero. With this one instrument we measure several acceleration techniques on the same prompts, so their quality costs can be compared. Perceived quality proves highly dependent on the domain of discourse. A 4-bit model was indistinguishable from its 16-bit original down to our design's +/-0.3-point resolution, in English prose and Chinese alike. At 3-bit precision the same prompts lost 0.5 points in English prose, 0.9 in Chinese, and 1.1 on multi-step math; early exit that cost 0.7 points on prose cost 2.5 on math, cutting correctly solved problems from 19 of 27 to 6. The pattern held for models from Alibaba and from Meta, but not its magnitude: the same quantizer cost Meta's model 1.8 points where it cost Alibaba's 0.7. A model's certainty about a token predicts how likely it is to differ from the full model's choice, but not how much that difference affects judged quality, so acceptance rules relying on certainty cannot distinguish errors that matter from errors that don't.

Grounded language-model systems are often evaluated by final answer accuracy, yet a correct answer can be unsupported, drawn from the wrong source, or produced when evidence is insufficient or contradictory. We introduce EviScope, a paired counterfactual benchmark that holds the question fixed while adding, removing, distracting, or contradicting its evidence. EviScope-v1.1 contains 40 four-condition quartets with repaired counterfactual claims and span-level support labels for automatic evaluation. Across 960 gold-blind generations from Qwen2.5-7B, Llama 3.1 8B, and Gemini 3.5 Flash, paired metrics expose model-dependent grounding behavior that answer accuracy hides. On two local open models, an explicit evidence-action gate underperforms vanilla RAG on QCS: 0.15 vs. 0.50 for Qwen and 0.10 vs. 0.375 for Llama. Gemini reaches 0.944 joint success under both prompts, yet still answers 5% of conflict cases after contradiction insertion. EviScope therefore distinguishes unsupported answering, conflict blindness, and wrong non-answer actions rather than scoring answers alone.