AI conference papers

AI conference paper library

Search papers from 30+ top conferences, including NeurIPS, ICLR, ICML, CVPR, ACL and AAAI. Filter by keyword, then open any paper for its abstract, authors and links.

arXiv

ResKV: Reconstructing Omitted Attention Contributions for Fixed-Budget KV Cache Compression

Yuhang Zhan, Lisi Chen, Shuo Shang

KV cache compression is essential for efficient long-context inference. Existing eviction methods permanently discard unselected tokens and consequently remove their aggregate contribution to attention. Merging-based alternatives preserve more information but can perturb retained keys and values that should remain exact. We observe that the information omitted by cache eviction can be formulated as residual statistics in both the numerator and denominator of softmax attention. Based on this observation, we propose ResKV, which divides a fixed KV budget into an exact main cache and a compact residual cache that reconstructs the contribution of omitted tokens. ResKV lets main-cache tokens and residual entries participate in the same softmax normalization, so residual entries restore both attention numerator and denominator mass rather than acting as a post-hoc correction. A construction-time validation proxy determines residual allocation for each layer and KV head, while a decode-time dynamic gate adjusts residual contributions for individual queries. Comprehensive evaluations on LongBench and RULER, covering query-aware and query-agnostic settings, multiple backbones, cache budgets, and representative compression baselines, demonstrate broad improvements under the same retained KV budget while preserving the practical efficiency of compressed decoding, including peak memory usage and long-context decode throughput.

arXiv

Sycophancy Undermines Epistemic Vigilance in Cooperative Vision-Language Tasks

Rupak Sarkar, Neha Srikanth, Saloni Gupta, Claire Bonial, Philip Resnik, Rachel Rudinger

To maintain common ground in cooperative conversation, humans iteratively update their beliefs as conversation participants share new information; participants who are epistemically vigilant detect when new information conflicts with prior beliefs and take steps to repair these conflicts. In order for AI systems to serve as reliable partners in complex cooperative tasks, they must similarly weigh incoming information against their own private evidence and shared context and appropriately surface inconsistencies when they arise. To measure the epistemic vigilance of vision-language models in cooperative settings, we present an information-asymmetric, dialog-based "spot-the-difference" task. Two models are privately shown one image each, and must determine through conversation whether the images are identical or, if not, identify the difference. Models routinely fail at this: they frequently overlook key evidence in their private image in favor of agreeing with their conversational partner, even when their agreement is unwarranted. We relate these violations of epistemic vigilance to the broader behavior of sycophancy, which manifests itself in cooperative goal-oriented dialog as over-accommodation and weak evidential grounding. Our results show that model steering to reduce sycophancy with a vector learned from task-agnostic sycophancy examples can reduce epistemic vigilance-related errors, making models more faithful reporters of their evidence, and in turn, more reliable partners in information-asymmetric cooperative tasks.

arXiv

ARB: A Matched Authorship-Rewriting Benchmark Dataset for AI-Text Detector Evaluation

Gaetano Perrone, Simon Pietro Romano

Standard AI-text detection benchmarks compare human-written text against text generated directly by large language models (LLMs). While prior work has shown that rewriting and paraphrasing can degrade detector performance, it remains unclear whether performance measured on this conventional benchmark predicts detector behavior when human-authored content is rewritten by an LLM. To address this gap, we introduce Authorship-Rewriting Benchmark (ARB), built from 1,800 human source texts (600 each from XSum, WritingPrompts, and OpenWebText) and four open-weight generators (Llama-3.2-3B, Qwen2.5-7B, Mistral-7B, Gemma-2-9B). Each source item yields four matched variants: human-written (HUMAN), direct LLM generation (Free-LLM), LLM-rewritten human text (H2L), and same-generator LLM-rewritten LLM text (LLM2L). We evaluated five detectors (FastDetectGPT, Binoculars-falcon-7b, RADAR, BERT-Defense, RoBERTa-Defense) at a strict 1%-false-positive operating point (TPR@1%FPR). FastDetectGPT and Binoculars-falcon-7b detected 91.2% and 93.5\% of direct LLM text, but only 30.8% and 15.1% of human text an LLM had rewritten, a drop of 60-78 percentage points. The same detectors retained 78.3% and 83.0% recall when LLM text was rewritten by the same model, a much smaller decline of 10-13 points. RADAR followed the same pattern (66.8% to 12.2%), while BERT-Defense and RoBERTa-Defense stayed below 3% recall across all regimes. These results show that detector performance measured on the conventional human-vs-LLM benchmark does not transfer to human-authored text revised by an LLM, even though the same detectors remain largely robust to LLM-only rewriting.

arXiv

Evidence-Type Competition: When Can Interventional Data Teach Language Models Causal Direction?

Xining Xun

Interventional data is widely regarded as the gold standard for teaching models causal reasoning. We test this assumption in a fully controlled synthetic environment pitting observational correlation against causal effect, and find it fails instructively. In Simpson's-paradox worlds, where the two have systematically opposite signs, increasing the fraction of interventional samples in pretraining does not improve causal direction: the magnitude of the model's do()-response grows monotonically, yet its sign is copied from the observational context. What governs whether interventional evidence is used is not the training mixture but the evidence type present in the context at inference time. Under an identical training recipe, a purely observational context induces systematic sign reversal in 29/50 worlds, a mixed context in 19/50, while aligned interventional probes alone yield 41/50 correct. Erasing observational evidence from the context immediately releases the suppressed causal interpolation ability (ratio_true = +0.56); a four-state content manipulation shows the switch is content-mediated and graded. The suppression is stable across training seeds (11/11 strong reversals persist on a matched-protocol second seed) and robust as a rate at 0.93B parameters (31.8% vs. 6% reversals in the matched probe-only arm), even as absolute gains shrink four-fold. An external audit on CLadder exposes a learned positive-effect prior with a two-layer structure: sign-randomized retraining removes it in-distribution but not out-of-distribution. We summarize: the capability lives in the weights; the switch lives in the context, and activation patching localizes the switch to the middle layers' observational rows. We further quantify the sampling noise floor of probe-based causal evaluation and an evidence-averaging protocol that cuts sign errors from 26% to 9%.

arXiv

Know It, Act on It: Investigating Memory Utilization in LLM Personalization

Zhaoxin Feng, Jianfei Ma, Emmanuele Chersoni

As large language model (LLM) agents evolve into personalized companions, memory has emerged as a core capability. However, LLMs face a knowledge utilization problem: they may fail to act on relevant user preferences even when they are fully present in context. When an agent fails to tailor its response in a context where previously shared user preferences should matter, it is unclear whether the model failed to remember that information or remembered it but failed to use it. To isolate this breakdown, we introduce a decoupled evaluation paradigm that administers paired Know and Act tests to the same user preference. We conduct large-scale experiments across 16 systems and five memory architectures, evaluating 1,000 preferences embedded at three levels of expression strength. Our results show a large gap between Know and Act outcomes: agents often pass the recall test for a user preference but fail to reflect that same preference in the paired behavioral scenario. While memory architectures reduce this gap, utilization remains especially weak for health and therapy-related preferences, where failures to act carry the greatest real-world stakes.

arXiv

Bridging the Question-Answer Gap in Retrieval-Augmented Generation: Hypothetical Prompt Embeddings

Domen Vake, Jernej Vičič, Aleksandar Tošić

Retrieval-Augmented Generation (RAG) systems synergize retrieval mechanisms with generative language models to enhance the accuracy and relevance of responses. However, bridging the style gap between user queries and relevant information in document text remains a persistent challenge in retrieval-augmented systems, often addressed by runtime solutions (e.g., Hypothetical Document Embeddings (HyDE)) that attempt to improve alignment but introduce extra computational overhead at query time. To address these challenges, we propose Hypothetical Prompt Embeddings (HyPE), a framework that shifts the generation of hypothetical content from query time to the indexing phase. By precomputing multiple hypothetical prompts for each data chunk and embedding the chunk in place of the prompt, HyPE transforms retrieval into a question-question matching task, bypassing the need for runtime synthetic answer generation. This approach does not introduce latency but also strengthens the alignment between queries and relevant context. Our experimental results on six common datasets show that HyPE can improve retrieval context precision by up to 42 percentage points and claim recall by up to 45 percentage points, compared to standard approaches, while remaining compatible with re-ranking, multi-vector retrieval, query decomposition, and other RAG advancements

arXiv

DenseOn with the LateOn: Fully Open Dense and Late-Interaction Models for Multilingual, Long-Context, and Code Search

Raphaël Sourty, Antoine Chaffin, Paulo Roberto Moura Junior, Amélie Chatelain

State-of-the-art retrieval models increasingly rely on closed training data, creating a reproducibility gap. We present an open end-to-end recipe for training retrieval models and study how English supervision transfers to multilingual retrieval through translate-train. We first reconstruct and curate 665M English contrastive pre-training pairs from 1.4B pairs across 34 public sources and build 1.88M supervised fine-tuning pairs with mined hard negatives. Training yields two 149M-parameter models: DenseOn, a single-vector dense model, and LateOn, a ColBERT-style late-interaction model. They achieve 56.20 and 57.22 average nDCG@10 on BEIR, respectively, setting new state-of-the-art results for this size class. We then translate the validated English data into eight languages, yielding 2.8B pairs with cross-lingual samples, and train mDenseOn and mLateOn, two 307M-parameter models built on mmBERT-base. Despite sharing their backbone, data, and objectives, their representations behave differently: the dense model is strong on English and translated languages but degrades outside translate-train support, whereas the late-interaction model generalizes better to unseen languages and scripts. This suggests that token-level matching turns translate-train from a target-language expansion strategy into a multilingual generalization recipe. We publicly release the models, datasets, and training code.

arXiv

SERPO: Self-Evolving Rubric Policy Optimization for Open-Ended Test-Time Reinforcement Learning

Jianze Wang, Kunwang Zheng, Ying Liu, Yu Cao, Qilong Zhang, Jinlong Chen, Hua Yang, Qianglong Chen

Test-time reinforcement learning (TTRL) enables language models to self-evolve at inference time without labeled feedback. Existing methods rely on answer voting and therefore do not extend naturally to open-ended generation, where valid responses cannot be mapped to a shared canonical answer. Without external reward models or stronger judges, adaptation must instead construct reliable rewards from the model's own outputs. We introduce SERPO (Self-Evolving Rubric Policy Optimization), which replaces answer voting with a closed loop that co-evolves response evidence, query-specific rubrics, and policy parameters. Good-Normal-Bad (G-N-B) response evolution organizes maximally separated rollouts into ordered archives; rubric evolution retains criteria that discriminate these archives; probabilistic criterion scoring converts verdict-token likelihoods into reward signals; and policy evolution optimizes the actor with the resulting signals. New actor rollouts then refresh both the archives and rubrics, closing the three-way evolution loop. Across two model configurations, two in-domain benchmarks, and four OOD benchmarks, SERPO improves HealthBench and ResearchQA by up to 20.63 and 20.31 points over the corresponding base models, raises the six-benchmark macro-average by up to 8.06 points, and supports OOD transfer and continued cross-benchmark evolution.

arXiv

Copy Less, Ground More: Overcoming Repetitive Copying in Long-Context Reasoning via Evidence-Aware Reinforcement Learning

Lizhe Fang, Weizhou Shen, Tianyi Tang, Yisen Wang

Large language models that generate step-by-step reasoning traces have achieved strong performance on complex tasks, and extending them to long-context settings has emerged as an important frontier. However, we identify a critical failure mode in this regime: \emph{repetitive copying}, where models extensively copy text from the input into their reasoning traces rather than productively solving the problem. We show that this behavior is pervasive across frontier long-context LLMs and intensifies with context length. By separating each prompt into task-relevant key evidence and irrelevant distractor context, we further show that the root cause is insufficient grounding: models copy from the prompt indiscriminately, and those that fail to focus on key evidence are far more likely to answer incorrectly. Motivated by this diagnosis, we propose GEAR (Grounding Evidence-Aware Reward), a reward shaping method that augments the accuracy signal with a grounding reward for overlap with key evidence and a distractor penalty for overlap with irrelevant context. To enable GEAR on natural-language data, we develop an automated pipeline that constructs evidence-annotated training data from arbitrary documents. We validate GEAR across multiple model scales and benchmarks, showing consistent improvements of up to +4.6 average points over standard RL with accuracy-based rewards, with larger gains at longer contexts, while also reducing repetitive copying and thinking length. Our findings suggest that, even as long-context evaluation shifts from simple retrieval toward complex reasoning, accurate grounding in relevant evidence remains an indispensable capability with substantial room for improvement.

arXiv

DRIP-R: A Benchmark for Decision-Making and Reasoning Under Real-World Policy Ambiguity in the Retail Domain

Hsuvas Borkakoty, Sebastian Pohl, Cheng Wang, Bei Chen, Yufang Hou

LLM-based agents are increasingly deployed for routine but consequential tasks in real-world domains, where their behavior is governed by inherently ambiguous domain policies that admit multiple valid interpretations. Despite the prevalence of such ambiguities in practice, existing agent benchmarks largely assume unambiguous, well-specified policies, leaving a critical evaluation gap. We introduce DRIP-R, a benchmark that systematically exploits real-world retail policy ambiguities to construct scenarios in which no single correct resolution exists. DRIP-R comprises a curated set of policy-ambiguous return scenarios paired with a realistic customer personas, a full-duplex conversational simulation with tool-calling capabilities and a multi-judge evaluation framework covering policy adherence, dialogue quality, behavioral alignment, and resolution quality. Our experiments show that frontier models fundamentally disagree on identical policy-ambiguous scenarios, confirming that ambiguity poses a genuine and systematic challenge to LLM decision-making.

arXiv

Few-Shot Contrastive Adaptation for Audio Abuse Detection in Low-Resource Indic Languages

Aditya Narayan Sankaran, Reza Farahbakhsh, Noel Crespi

Abusive and hateful speech is increasingly spoken rather than written, surfacing in voice notes, calls, and short-form videos. Most detection systems still transcribe speech to text before classifying it, but transcription is unreliable for languages lacking strong speech recognisers, and it discards the tone and emotion that often carry the abuse itself. This paper examines whether abusive speech can instead be detected directly from audio, using CLAP, a model that learns a shared representation of sound and language, evaluated across ten Indic languages in the ADIMA dataset. A lightweight classifier trained on CLAP's existing audio representations, without adapting the model itself, comes within one to three points of a fully supervised system, and far outperforms prompting with no labelled examples at all. Further adaptation with a handful of labelled examples per language yields little extra benefit, varying unpredictably across languages. CLAP-based audio representations thus already offer a strong, inexpensive foundation for detecting abusive speech across languages, lowering the labelled data needed in practice.

arXiv

Preconditioned Test-Time Adaptation for Out-of-Distribution Debiasing in Narrative Generation

Hanwen Shen, Ting Ying, Jiajie Lu, Shanshan Wang

Although debiased large language models (LLMs) excel at handling known or low-bias prompts, they often fail on unfamiliar and high-bias prompts. We demonstrate via out-of-distribution (OOD) detection that these high-bias prompts cause a distribution shift, degrading static model performance. To enable real-time correction, we propose CAP-TTA, a test-time adaptation framework. CAP-TTA triggers context-aware LoRA updates only when a bias-risk score exceeds a set threshold. By utilizing an offline precomputed diagonal preconditioner, it ensures fast and stable optimization. Across multiple benchmarks and human evaluations, CAP-TTA effectively reduces toxicity/bias score with significantly lower latency than standard optimization methods (e.g., AdamW or SGD). Furthermore, it prevents catastrophic forgetting, and substantially improves narrative fluency over state-of-the-art baselines without compromising debiasing performance.

arXiv

When Iterative RAG Beats Ideal Evidence: A Diagnostic Study in Scientific Multi-hop Question Answering

Mahdi Astaraki, Mohammad Arshi Saloot, Ali Shiraee Kasmaee, Hamidreza Mahyar, Soheila Samiee

Retrieval-Augmented Generation (RAG) extends large language models (LLMs) beyond parametric knowledge, yet it is unclear when iterative retrieval-reasoning loops meaningfully outperform static RAG, particularly in scientific domains requiring multi-hop reasoning over sparse, heterogeneous evidence. We provide the first controlled, mechanism-level diagnostic evaluation of whether synchronized iterative retrieval and reasoning can surpass even an idealized static upper bound (Gold Context) RAG. We benchmark eleven state-of-the-art LLMs under three regimes: (i) No Context, measuring reliance on parametric memory; (ii) Gold Context, where all oracle evidence is supplied at once; and (iii) Iterative RAG, a training-free controller that alternates retrieval, hypothesis refinement, and evidence-aware stopping. Using the chemistry-focused ChemKGMultiHopQA dataset, we isolate questions requiring genuine retrieval and analyze retrieval coverage gaps, anchor carry drop, query quality, composition fidelity, and control calibration. Iterative RAG consistently outperforms Gold Context, with gains up to 25.6 percentage points, especially for non-reasoning fine-tuned models. Staged retrieval reduces late-hop failures, mitigates context overload, and enables dynamic correction of early hypothesis drift, but failure modes remain, including incomplete hop coverage, distractor latch trajectories, early stopping miscalibration, and high composition failure rates even with perfect retrieval. Overall, the process of staged retrieval is often more influential than the mere presence of ideal evidence. We provide practical guidance for deploying and diagnosing RAG in specialized scientific settings. Code and evaluation results are available at https://github.com/Matroid1998/Iterative-rag

arXiv

"Not in My Backyard": LLMs Uncover Online and Offline Social Biases Against Homelessness

Jonathan A. Karr Jr., Benjamin F. Herbst, Matthew L. Sisk, Xueyun Li, Ting Hua, Matthew Hauenstein, Georgina Curto, Nitesh V Chawla

Homelessness is a persistent social challenge, impacting millions worldwide. Over 876,000 people experiencing homelessness (PEH) were recorded in the U.S. in 2025. Social bias is a significant barrier to alleviating homelessness, shaping public perception and influencing policymaking. Because online textual media and offline city council discourse both reflect and influence public opinion, they provide valuable signals for identifying and tracking social biases against PEH. We release the first multi-domain PEH bias corpus with a 16-category multi-label taxonomy: a 1,698-item stratified gold-standard set annotated by partner-trained raters, plus 48,389 GPT-4.1-labeled texts, drawn from Reddit, X (formerly Twitter), news, and council meeting transcripts across ten U.S. cities (2015-2025). We benchmark six prompted LLMs on the gold-standard set and complement F1 with prevalence-gap audits. Moderate F1 coexists with large miscalibration: every model over-tags "not in my backyard" (NIMBY) (+11.5 pp) and under-detects factual claims (-30.5 pp). Error analysis on consensus false positives reveals that models treat housing vocabulary and question form as opposition proxies, producing NIMBY false positives on pro-service text. The corpus and audit protocol support municipal PEH stigma monitoring without treating teacher labels as ground truth.

arXiv

On the Fundamental Impossibility of Hallucination Control in Large Language Models

Michał P. Karpowicz

Large language models hallucinate. This paper shows when that is unavoidable and what we can do about it. We model inference as an auction of ideas, in which a model's components, each holding partial knowledge, compete to shape the answer. We then prove Impossibility Theorems showing that whenever a query makes LLM components contest a fact they hold in common, no aggregation of their reports can at once report that knowledge truthfully, avoid manufacturing confidence beyond what it supports, keep the relevant components engaged, and give the best answer. Something must give, and each failure is familiar: a fabricated detail, unearned confidence, ignored knowledge, or a needlessly weak reply. This is no artifact of one design. It reappears when components report probabilities, and inside the transformer itself, where the combined answer is credited more confidence than the internal contributions supplied. The unbalanced semantic budget cannot be settled from within. Factual truth lies outside the model, and in the worst case no internal signal can certify it. What can be certified is support. Given externally authorized evidence, checking that an answer stays within what the evidence entails needs only the answer and the evidence, and we prove when that check is computable. However, a correct answer can lack support, and a supported answer can be false. What counts as evidence, how far beyond it we allow answers to reach, and which failures we can live with are choices no model can make for us.

arXiv

PTP: Previous-Token Prediction based LLM Inversion for Near-Exact Prompt Reconstruction

Pirzada Suhail, Nagasai Saketh Naidu, Atanu R Sinha, Amit Sethi

Large language models (LLMs) generate text by auto-regressively sampling the next token. This inherently leads to a many-to-many mapping between prompts and responses, complicating the task of inferring prompts from observed outputs. Prior work on LLM inversion frames prompt recovery as a semantic reconstruction task. They rely on fine-tuning pretrained sequence-to-sequence models on large external datasets--and requiring access to model weights or logits--to generate semantically plausible prompts. In contrast, we present a functional approach to inverting a given LLM in a black-box setting, without auxiliary aids. We train an explicit inverse language model entirely from scratch on data synthetically generated from the target LLM itself. Analogous to forward next-token prediction, our inverse model is trained using previous-token prediction, establishing a generative link between the forward and inverse processes that enables faithful prompt reconstruction. Moreover, it naturally supports diverse prompt reconstructions through sampling, whereby all such prompts induce similar responses under the forward, target LLM. Our approach generalises across datasets and exhibits transferability in reconstructing prompts from responses generated by different LLMs. Further, across the set of token based evaluation metrics for prompt and response reconstructions, our approach outperforms prior work.

arXiv

Zero-Mem: Zero-Token Memory Operations for LLM Agents

Yilin Xiao, Zhehan Zhu, Yujing Zhang, Jin Chen, Zijin Hong, Luyao Zhuang, Qinggang Zhang, Shengyuan Chen, Xiaocao Ouyang, Lingfei Ren, Xiao Huang

LLM agents need memory to act consistently over long interactions, yet many systems use additional LLM calls to operate that memory. Generating intermediate records and mediating their retrieval adds recurring token and time costs, while omitted or merged details can obscure the original evidence. We ask whether structured memory access requires generation at all. Zero-Mem introduces \emph{zero-token memory operations}: no step outside final question answering invokes an LLM or consumes LLM input or output tokens; encoder computation is accounted for separately. Zero-Mem preserves original interaction traces as its source of record. It organizes the traces in two complementary ways. An entity--context graph exposes connections across interactions, while a temporal hierarchy preserves conversational locality and session state. For each query, Zero-Mem weighs the two views, retrieves from both, and follows their structure to recover supporting relations or surrounding context. Deterministic calibration first discards conflicting evidence and then keeps the reader's answer grounded in the retrieved traces. Only the final-QA reader invokes an LLM. Across long-memory and long-context question-answering benchmarks, Zero-Mem achieves competitive performance while eliminating LLM calls and LLM-token consumption from memory operations. With the same final-QA reader and context budget, it reduces memory-operation time cost by 57.6\% relative to the fastest compared baseline. Ablations support the contribution of the two views and their query-dependent coordination. Overall, the results show that structured agent memory need not generate an intermediate representation of the past. After peer review, the code and implementation details will be available at \textcolor{blue}{https://github.com/TheMoon0815/Zero-mem}.

arXiv

Cross-Lingual Transfer for Machine Translation in Turkic Languages

Omer Burak Cinar, Mehmet Mert Dalkilic, Cagri Toraman

Cross-lingual transfer is central to low-resource machine translation, but its behavior within closely related language families remains insufficiently characterized. We study transfer among five Turkic languages; Turkish, Azerbaijani, Uzbek, Kazakh, and Kyrgyz; using pairwise transfer matrices. In this setting, each model is fine-tuned with one transfer source and evaluated on a different transfer target while the translation target remains the same. Across mT5 experiments, we find that transfer is strongest between closely related Turkic pairs, especially Turkish-Azerbaijani and Kazakh-Kyrgyz. We also show that transfer direction matters, and that the same transfer source-transfer target pair can behave differently when the translation target changes. Latinization improves BLEU and chrF in several script-mismatched settings, but its effect is not uniform across metrics. Additional analyses show that transfer sources are mostly stable across different datasets and model settings.

arXiv

Translation with Thought: Difficulty-Adaptive Reasoning via Reinforcement Learning for Multi-Domain Machine Translation

Yongshi Ye, Biao Fu, Chongxuan Huang, Yidong Chen, Xiaodong Shi

Multi-domain machine translation (MDMT) poses a unique challenge due to varying levels of linguistic complexity across domains. Inspired by human translators' ability to adapt reasoning effort based on difficulty, we propose TwT (Translation with Thought), a resource-rational framework that learns to modulate inference between intuitive and deliberate reasoning. TwT is trained in two stages: (1) supervised fine-tuning on difficulty-aware long chain-of-thought traces distilled from DeepSeek-R1 and rewritten by GPT-4o to reflect human-like reasoning economy, and (2) reinforcement learning with a hybrid reward to optimize translation quality and reasoning efficiency. Evaluated on 15 benchmarks spanning in-domain and out-of-domain settings, as well as 3 seen and 59 unseen languages, with ablations across three backbone models, TwT-7B and TwT-14B outperform much larger SOTA reasoning models in translation quality, while reducing token usage by 32--60\%. These results confirm that aligning translation behavior with cognitive principles enables robust generalization, high translation quality, and efficient reasoning in MDMT.

arXiv

Language Models Agree With Each Other, Not With Readers

Kazuki Nakayashiki, Keisuke Watanabe

Claims that language models homogenise are usually measured against human judgements collected for the study, which makes the human side an artifact of the design: a crowdworker given the model's instruction is running the model's prompt. We measure convergence against a human reference nobody built for the purpose -- 2,523 reader mark sets across 120 web documents, produced by people highlighting for their own reasons on a platform where the overlay of others' marks is off by default. Agreement is the overlap between two size-matched sentence sets minus the overlap expected when each is resampled within its own depth-and-length bands. The null's calibration is demonstrated, not asserted: every pair involving a random baseline lands within 0.006 of zero. On the median document each party names 14 sentences of 70; two readers share 4.1 and two models 8.7. Across 18 model arms spanning 11 vendors, 3 countries and both weight regimes, the median of 153 model pairs is +0.093 against a human yardstick of +0.040, and 99 sit entirely above the human interval. Two frontier models from rival labs reach +0.203, twice what GPT-4o agrees with itself on a second call. The effect is not determinism, prompt wording, procedure, vendor or routing, and it is graded: the smallest models agree at the human level. No model agrees with readers detectably more than a reader does, and at equal depth and length no surface feature separates their choices. The multiples are procedure-dependent and the ordering is not: models are cut to their sharpest set while a reader's is a random draw from what they marked, and blunting the models alike halves the gap without closing it. Tested out of sample on four models released after this analysis, against predictions fixed beforehand, none clears the human interval. A population simulated from several models is not several populations.

Keep browsing

More papers from top AI conferences

30+ top conferences, by conference, year, and topic.