paperlist.ai

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

Credal Large Language Models for Semantic Commitment under Uncertainty

Shireen Kudukkil Manchingal, Sofiia Nikolenko, Fabio Cuzzolin

Large language models (LLMs) often produce fluent but incorrect answers with unwarranted confidence. A central limitation is that standard LLMs represent uncertainty through a single predictive distribution, conflating epistemic ignorance with genuine ambiguity. We introduce Credal Large Language Models (CLLMs): an ensemble of LoRA adapters induces a credal set whose lower and upper probabilities expose the spread of plausible predictive distributions rather than collapsing to a single softmax output. From this representation we derive two complementary commitment scores. Credal Token Commitment (CTC) is a token-space score that combines lower-bound support, credal width, and intersection entropy, computed without additional generation. Semantic Commitment Consistency (SCC) extends commitment to semantic space using sampled completions, with SCC-Gap measuring the mismatch between token-level and semantic-level support. We evaluate hallucination detection, calibration, selective prediction, and reasoning on Gemma-2-9B, Llama-3.1-8B, and Qwen2.5-7B across OpenBookQA, CoQA, TriviaQA, and ARC-Challenge. CLLM is the best method on QA accuracy at competitive expected calibration error, and CTC tracks the best hallucination AUROC within 1.5 pp on most settings without additional generation. On selective prediction at 80% coverage, CLLM with SCC reaches 99.0% accuracy on OpenBookQA, and on ARC-Challenge CLLM with Csem confidence achieves <= 0.6% ECE across the three backbones.

arXiv

A Multi-Domain and Multi-Task Generative Framework with Explicit Task and Domain Conditioning for Cross-Domain Event Extraction

Siting Liang, Omar Adjali, Daniel Sonntag

Event extraction aims to identify event triggers, classify event types, and extract arguments to construct structured event representations. Despite strong in-domain performance, developing models that generalize robustly across domains remains challenging due to variations in contextual expressions and event schemas. Prior unified and multi-task approaches improve in-domain accuracy but exhibit limited flexibility when applied to unseen domains. Even large language model-based methods that provide full event ontologies at inference time often underperform compared to smaller, task-specific fine-tuned models. We propose a unified multi-domain and multi-task training framework that models heterogeneous event schemas within a single model. Our approach introduces domain conditioning signals, jointly with task-specific prompts, enabling dynamic adaptation to dataset-specific schemas without requiring complete event label sets at inference time. The framework supports both pipeline and end-to-end extraction settings, facilitating efficient task- and domain-level transfer. Experiments on diverse event extraction benchmarks demonstrate that our method achieves competitive performance, strong cross-domain generalization, and practical scalability, while preserving domain-specific precision.

arXiv

Aligning Biomedical Texts and Knowledge Graphs: A Systematic Comparison of Lightweight Alignment Strategies

Artem Bisliouk, Elizaveta Nosova, Heiko Paulheim, Andreea Iana, Rita T. Sousa

Biomedical knowledge exists in two complementary but distinct forms: unstructured scientific literature and structured knowledge graphs (KGs). Aligning them is essential for knowledge grounding, evidence retrieval, and KG completion, yet existing methods do not explicitly align free-text evidence with KG triples. We present a unified framework for systematically studying design choices for aligning biomedical text and KGs. With a text encoder and a KG embedding model both frozen, we learn only a lightweight projection between their spaces via a contrastive objective. This enables a fair comparison across six design dimensions: text encoder, KG embedding model, projection head, triple composition, training direction, and hard-negatives sampling. We construct CTD-Align, a corpus of over 22K one-to-one tripledocument pairs linking chemical-gene interactions from the Comparative Toxicogenomics Database to supporting PubMed passages. We evaluate alignment on it in two retrieval settings: document-to-triple and triple-to-document. We find that the triple composition and the training direction (i.e., shared retrieval space) have the greatest impact, whereas the text encoder and hard-negatives sampling matter little. Overall, simple choices win: projecting text into the KG space with a linear head over concatenated subject, predicate, and object embeddings performs best. These findings establish lightweight contrastive alignment as an effective, practical foundation for bridging biomedical text and KGs.

arXiv

Cognitive Profiling of LRMs' Reasoning Traces Using Bloom's Taxonomy

Maria-Eleni Zoumpoulidi, Georgios Paraskevopoulos, Alexandros Potamianos

Large Reasoning Models (LRMs) have revolutionized reasoning in LLMs, and the increasing public availability of reasoning traces creates valuable opportunities to study model behavior not only at the surface level but also at the granularity of individual reasoning steps. However, understanding the types of thinking employed during reasoning - which offers critical insights into models' reasoning patterns and enables actionable applications - remains underexplored. To address this gap, we introduce a framework for automatic annotation of reasoning steps through the lens of Bloom's Taxonomy, which classifies thinking into six cognitive levels, such as Remembering, Applying and Evaluating. Using this framework, we perform a large-scale analysis across models and datasets, revealing both similarities and differences in thinking patterns across models and tasks. Moreover, we demonstrate that thinking-type information derived from reasoning traces correlates with correctness, paving the way for improved reasoning. Our findings establish a fine-grained framework for analyzing thinking patterns in LRMs and provide actionable insights for enhancing reasoning quality.

arXiv

CaRGo-T: Causal Reasoning Graph-of-Thought improves Multimodal Humor Comprehension

Abhilash Nandy, Rahul Seetharaman, Aman Bansal, Rounak Saha, Manav Nitin Kapadnis, Millon Madhur Das, Pawan Goyal, Niloy Ganguly

Large-scale vision-language models (VLMs) have demonstrated remarkable versatility across a wide range of multimodal tasks. However, understanding humor remains challenging because humorous content often depends on subtle interactions among entities, events, context, and implicit relationships across image and text modalities. These interactions can involve complex chains of reasoning that are difficult to capture through conventional prompting or linear chain-of-thought reasoning. In this work, we propose CaRGo-T (Causal Reasoning Graph-of-Thought), a reasoning framework that represents the causal and contextual relationships underlying multimodal humor as a lightweight graph-based reasoning structure. The graph is serialized into a code-based representation generated by a VLM, which can subsequently be interpreted by the same or a different VLM to produce the final prediction in zero-shot or in-context learning settings. We evaluate CaRGo-T on humor understanding and humor detection across four datasets spanning diverse forms of comedic content, including satire, sarcasm, and memes. Experiments with state-of-the-art commercial and open-source VLMs show that CaRGo-T consistently improves performance over existing reasoning-based baselines, achieving gains of approximately 1-20% on humor understanding and 1-3% on humor detection. Further analysis using mutual information indicates that the reasoning representations produced by CaRGo-T contain more information relevant to the target output than those generated by baseline reasoning approaches. Code is available at https://github.com/abhi1nandy2/CaRGo-T.

arXiv

Accelerating Diffusion Language Models via Structured Suffix Modeling

Zifeng Cheng, Keda Li, Zhiwei Jiang, Cong Wang, Fei Shen, Qing Gu

Diffusion Language Models (DLMs) exhibit strong parallel decoding capabilities by denoising multiple tokens in a single generation step. However, this parallelism comes with substantial computational overhead, as each step requires interactions with all suffix tokens. Existing methods typically reduce this cost by retaining only a local suffix window as a substitute for the full suffix. Despite their effectiveness, these methods overlook the structural heterogeneity across suffix regions and re-initialize suffix tokens with identical representations at each timestep. To this end, we propose a structured suffix modeling method for efficient DLM inference. Specifically, we divide the suffix into three regions, i.e., the local, middle, and tail regions, and retain different numbers of suffix tokens in each region according to their structural roles. Moreover, we incorporate the decoding results from the previous step into the suffix token representations at the current step, allowing them to carry evolving denoising information across generation steps. Notably, our method is training-free and orthogonal to several existing acceleration techniques, such as parallel decoding strategies and KV cache. Empirical results across multiple benchmarks on three DLMs demonstrate that our method can further accelerate DLM inference and improve performance in most cases. In particular, in long-sequence inference, our method achieves up to a \(72.81\times\) speedup when combined with other acceleration techniques. Our code is available at https://github.com/zifengcheng/SSM.

arXiv

Activation-Weighted Seeded Residual Coding for Low-Bit LLM Weight Repair

Zehao Liu, Chuangchuang Fang, Yang Ren

Low-bit weight quantization saves storage but leaves errors that degrade language-model quality. We introduce Activation-Weighted Seeded Residual Coding (AWSRC), a compact repair codec for an existing quantization backbone. Given a reconstructed weight $W_0$, AWSRC encodes the residual $W-W_0$ using deterministic seed-generated bases. The sidecar stores seed selectors, low-bit coefficients, and scales rather than an explicit codebook. Activation statistics prioritize errors that affect layer outputs. On Qwen2.5-3B-Instruct, adding 0.162 scope-bits/weight to an INT4 RTN backbone closes 88.2%, 78.9%, and 71.3% of the matched PPL, KL, and accuracy gaps to BF16. Repairing a matched strong low-bit backbone also improves all measured quality metrics. With a matched 49.25 MB sidecar, about 0.8% of the BF16 model-weight payload, AWSRC gives the best perplexity and mean task accuracy among sparse, low-rank, and vector-quantized codecs.

arXiv

Statistical Machine Translation Systems of English-Pnar Language Pair : Some Insights of the Emperical Study

Edawanbiang Dhar Surmila Thokchom, Thoudam Doren Singh

Pnar, an Austroasiatic language spoken by approximately 0.4 million people in the Jaintia Hills of Meghalaya, lacks the digital corpora and natural language processing (NLP) resources. This paper presents the first machine translation study for the English and Pnar language pair. Using articles collected from the Wyrta newspaper, we built a parallel corpus comprising of 10,234 sentences and trained phrase-based statistical machine translation (SMT) systems the models using 9,563 parallel corpora under three configurations for each direction using Moses, GIZA++ , KenLM, varying lexicalized reordering and minimum error rate training (MERT) tuning. The models are evaluated on a held out test set of 371 sentences, the best performing system achieves a BLEU score of 14.97 (chrF2: 33.42, TER: 77.60) for Pnar to English and 11.16 (chrF2: 31.38, TER: 93.51) for English to Pnar, establishing the first quantitative benchmark for this language pair. Lexicalized reordering improves translation quality by 3.73 BLEU points for Pnar to English, reflecting the structural shift from the source language's SOV word order to the target language's SVO order, whereas MERT tuning degrades BLEU performance under low resource conditions. Finally, we analyze the remaining translation errors, including morphological out of vocabulary (OOV) words, long-distance reordering and Khasi code mixing and discuss future directions toward neural and multilingual machine translation for Pnar.

arXiv

Definitional Sensitivity in Media Bias Detection: A Multi-Definition Dataset and Benchmark

Martin Wessel, Timo Spinde, Jürgen Pfeffer, Gianluca Demartini

Media bias detection relies on definitions and examples that specify what counts as bias, yet these specifications often vary across datasets or remain implicit, even when given the same name. Such variation makes it unclear whether models trained for the same bias category learn the same construct or different phenomena, a problem largely overlooked in prior work. We examine how definition choice affects bias annotation in a between-subjects experiment with 354 participants and a parallel evaluation with four LLMs. Participants and models rate six news articles across four bias categories using definitions that vary in conceptual framing and elaboration. Across 8,496 human and 28,800 LLM ratings, we find that the conceptual target of a definition drives annotation divergence, while construct-preserving elaboration does not: conceptual framing significantly shifts annotations for humans and does so even more strongly for LLMs. We discuss implications for construct specification in annotation protocols and prompt-based measurement, and consider how definitional sensitivity may propagate to downstream classification beyond media bias. We also release MUDD, the Multi-Definition Bias Detection Dataset.

arXiv

AgentWeave: Routing Before Reasoning for Efficient Function Calling in Tool-Rich Language Models

Saurav Singla, Aarav Singla, Advik Gupta, Parnika Gupta

Large language models increasingly operate over large collections of tools, functions, APIs, and specialized agents. As the candidate action space grows, a function-calling model must process more schemas, consume more prompt tokens, and distinguish among increasingly similar or irrelevant alternatives. We study a complementary systems strategy: reduce the candidate set before language-model inference while leaving the downstream model unchanged. We introduce AgentWeave, a deterministic pre-inference routing layer that constructs a bounded model-visible action space using eligibility, requirement, capability, and routing signals. We evaluate AgentWeave with a frozen BFCL-derived routing-pressure protocol using the public MadeAgents/Hammer2.1-1.5b model. On 48 fresh BFCL V4 multiple-function tasks, AgentWeave achieves 6/48 (12.5%) native BFCL successes, whereas all-tools, deterministic random top-8, and semantic top-8 baselines each achieve 0/48. The paired success difference is +12.5 percentage points with a 10,000-resample paired bootstrap 95% confidence interval of +4.17 to +22.92 points and exact McNemar p=0.03125. Relative to all-tools exposure, AgentWeave presents 70.18% fewer tools, uses 61.70% fewer input tokens, and exhibits 50.95% lower mean local-model latency. The result is deliberately narrow: this is a BFCL-derived routing-pressure study rather than an official full BFCL leaderboard score, and absolute task success remains low. The evidence nevertheless shows that candidate-space construction can materially affect a fixed model's function-calling behavior and motivates evaluating routing as a distinct stage before model reasoning.

arXiv

Signal or Noise? A Benchmark Study of Agent Skills in Web Development

Ziyue Yang, Fan Ding

Agent Skills are reusable procedural modules that are increasingly injected into coding-agent sessions to encode framework conventions, anti-patterns, and reusable tools. However, because each injected Skill expands the prompt of every query, an effective Skill benchmark must determine not only whether an agent can solve a task, but whether the Skill should have been injected at all. We introduce WebDev-Skills-Bench and use it for a controlled empirical study of 31 public WebDev Skills on 50 Web-Bench projects and 1,000 ordered tasks. The benchmark compares four matched conditions, including a length-matched irrelevant control and leave-one-out component ablations. To isolate Skill effects from prompt-length artifacts, we place only SKILL.md in the prompt while mounting auxiliary files into the agent workspace. Across four models, target Skill injection reduces mean Pass@2 by 1.3% to 4.2%, lowers task completion depth, and increases token cost by 72% to 394%, with gains in only 17% to 36% of Skill-project pairs. Length-matched controls reveal two failure modes: some models are length-distracted, where an equally long irrelevant Skill reproduces most of the loss, while others are content-misled, where prompt length is neutral but Skill content still lowers Pass@2 by 1.1% to 1.4%. Further analysis shows that losses concentrate on easy early tasks, Skill rankings transfer weakly across models, and anti-pattern rules outperform example-heavy content within helpful Skills. These findings recast a matched Skill as a hypothesis about a particular Skill-project-model triple rather than a portable asset, reframing injection as a per-deployment routing decision and making length-matched controls and per-model audits a minimum standard for Agent-Skill evaluation.

arXiv

Cultural Moment Benchmark: Evaluating Video Cultural Reasoning and Grounding in Southeast Asia

Burak Satar, Zhixin Ma, Cheng Yu-Tong, Huy Hoang Tran, Phuong Anh Nguyen, Chong-Wah Ngo

Cultural understanding in video means more than recognizing what is visible; it requires grasping the symbolic and temporal significance of cultural concepts. We decompose this into three abilities: naming what a concept symbolizes, visually recognizing it on video, and locating its sub-events in time. Existing video-cultural benchmarks tend to test what is seen, collapsing these three abilities into a single score that hides the bottleneck. We introduce the Cultural Moment Benchmark (CMB): 306 expert-curated concepts from seven countries in Southeast Asia across five categories. We evaluate each concept through three stages, one per ability. Given a description, Stage 1 (S1) selects from four candidate concept names, Stage 2 (S2) selects from four candidate video moments, and Stage 3 (S3) predicts the start and end times of the moment in a video. To keep each stage focused on a distinct ability, we use three design choices: semantic-similarity distractors (S1, S2), unlabeled video moments (S2), and free-form localization on a different example video (S3). Across six vision-language models, failure modes vary by ability and modality. i) Even the strongest closed-source models score below 30% when all three stages must be correct; ii) The three abilities do not fully cascade: naming a concept correctly helps half the models recognize it on video, but recognizing it has little effect on locating the sub-event in time; iii) Audio is complementary, redundant, or distracting depending on the concept, more often distracting in non-Latin-script countries; removing both audio and subtitles hurts Games and Music the most. Our 14-rater human study shows that even Expert raters score below chance on concepts from a neighboring country, indicating that CMB requires country-specific cultural knowledge. CMB acts as a diagnostic harness, attributing failures to a specific ability or modality.

arXiv

Beyond Verdicts: A Graph-Based Analysis of Human and LLM Reasoning in Scientific Fact-Checking

Abdul Ghafoor, Muhammad Arslan Manzoor, Yufang Hou

Misinformation that cites legitimate papers can be especially harmful when it distorts what those studies actually report. While existing automatic fact-checking systems based on large language models (LLMs) can assess whether a model assigns an Incorrect verdict and can gen- erate explanations for that decision, they typi- cally do not indicate whether the model follows the same reasoning path as human experts or arrives at the verdict through a different but still valid path. In this work, we introduce a graph- based framework (typed reasoning graph) for comparing human and LLM reasoning paths in scientific fact-checking. Building on prior work on fallacious reasoning in biomedical misinformation, MISSCIPLUS (Glockner et al., 2025), we model each explanation as a rea- soning graph that links the false claim to the relevant study context, study findings, fallacy- supporting premises, and fallacy labels. This representation enables one-to-one alignment of human and LLM reasoning at the level of fallacy-specific sub-graphs. For non-human- aligned LLM paths, we validate grounding in the cited study, relevance to the claim, and suf- ficiency for the verdict. Using 84 false claims from MISSCIPLUS, we evaluate GPT-5, Claude Opus 4.7, and Qwen3-32B across prompt and evidence settings. Results show distinct perfor- mance dimensions: Qwen3-32B has the lowest verdict failure rate, GPT-5 the highest human alignment, and Claude Opus 4.7 weak verdict prediction but often valid reasoning in success- ful cases

arXiv

AutoSaddler: Automatic Harness Optimization with Durable Updates from Agent Execution Traces

Sungho Park, Wonjoong Kim, Rongyuan Tan, Jue Zhang, Wook-Shin Han, Pengfei Gao, Chanyoung Park, Yongqiang Yao, Rao Fu, Elsie Nallipogu, Qingwei Lin, Saravan Rajmohan, Dongmei Zhang

LLM agents remain unreliable on long-horizon tasks, where small local failures can compound over extended interactions and lead to overall task failure. Although external harnesses can substantially improve robustness, harness design remains a manual and expensive process that requires searching over a large space of prompts, tool configurations, and control logic. We propose AutoSaddler, an automatic harness optimization framework that formulates harness improvement as an offline learning problem and iteratively updates the harness using failure signals from mini-batches. AutoSaddler combines failure-trace diagnosis, structured patch generation that treats the harness as code, and validation-based update selection. Experiments on GAIA2, SWE-Bench Pro, and Terminal-Bench 2.0 show that AutoSaddler substantially improves agent performance over the corresponding base harnesses, achieving gains of 9.0, 9.6, and 10.0 percentage points, respectively. Ablation studies further suggest that effective harness optimization benefits from three ingredients: deep debugging rather than shallow reflection, targeted modifications rather than unconstrained editing, and generalization-aware selection rather than trajectory-specific repair. Together, these results suggest that automatic harness optimization is a promising path toward more performant and reliable agent systems.

arXiv

The Multilingual FrameNet Corpus

Beatrice Fiumanò, Nicolas Lazzari, Simone Paolo Ponzetto, Valentina Presutti

This paper introduces the Multilingual FrameNet Corpus (mFNC), a novel resource that extends the English Berkeley FrameNet corpus by collecting and harmonizing existing language-specific corpora across nine additional languages: Brazilian Portuguese, Chinese, Dutch, French, German, Italian, Korean, Latvian and Swedish. By training models that rely on different architectures on the mFNC, we consistently outperform existing state-of-the-art Frame Semantic Parsers in both multilingual and cross-lingual settings, underscoring the importance of multilingual training data. The mFNC and our trained FSP models are openly available at https://github.com/beatrice-f/mFNC.

arXiv

ST$^2$U: Stateful Test-Time Unlearning via Restricted Knowledge Boundary Control

Xunlei Chen, Qinghui Gong, Ruini Xue, Yaodong Hu, Tian Lan, Wenhong Tian

Controlling restricted knowledge in large language models is essential for model alignment and safe deployment. Test-time unlearning avoids costly retraining and parameter updates by intervening only during inference. However, existing activation-editing methods apply isolated pointwise corrections, overlooking how autoregressive generation continually reconstructs hidden states from the prompt, cache, and generated prefix. Consequently, later states may return to restricted knowledge regions after a locally successful correction, causing restricted knowledge re-entry. In this work, we propose Stateful Test-Time Unlearning via restricted knowledge boundary control (ST$^2$U), which formulates test-time unlearning as trajectory-wide boundary control. ST$^2$U first models restricted knowledge boundaries in low-dimensional invertible coordinates while leaving orthogonal non-target components unchanged. During inference, ST$^2$U monitors risk along the trajectory, applies minimal boundary corrections with contextual anchoring, and propagates historical correction states across tokens to mitigate knowledge re-entry. This trajectory-wide control enables more persistent forgetting while preserving non-target capabilities and limiting inference overhead. Across three benchmarks and three model families, ST$^2$U delivers the strongest overall balance, combining best or second-best retention with competitive forgetting and substantially less restricted-knowledge re-entry than test-time baselines (13.76%-19.84% versus 46.50%-59.10%).

arXiv

Meta-Moderator: Empowering Multi-Agent Debate with Meta-Cognition

Wentao Hu, Zhuoyue Wan, Jinhao Shen, Chen Jason Zhang, Xiaoyong Wei, Qing Li

Multi-agent debate can improve large language model reasoning by eliciting diverse hypotheses and critiques, yet its performance is often constrained by weak moderation. Common pipelines rely on fixed budgets, agreement-based stopping, or untrained judges, leading to redundant deliberation and unreliable evidence aggregation. We cast moderation as a meta-cognitive process, monitoring debate utility, controlling deliberation, and adjudicating a final answer, and introduce Meta-Moderator, a learnable framework that dynamically regulates debate and decides when to finalize an answer. Meta-Moderator is trained independently of the debaters via outcome-driven policy optimization, making debate regulation an explicit capability rather than an incidental effect of prompting. Across five benchmarks, Meta-Moderator outperforms widely used decision layers and transfers across tasks and system configurations. Further analyses show that it allocates debate more selectively and reduces mis-aggregation after informative hypotheses appear.

arXiv

Beyond Surface Cues: Disentangling Sociocultural Signals in Multilingual LLMs

Yuanjun Feng, Tanzhou Liu, Stefan Feuerriegel, Yash Raj Shrestha

Multilingual LLM outputs can vary across sociocultural contexts. However, evidence of cultural grounding can be misleading: identity labels may be inferred from explicit or indirect textual cues, while names and wording can reveal the source language. Treating all these signals as evidence of cultural grounding may obscure potential biases. We present a human-validated, multi-agent audit that separates three questions: whether outputs reproduce social biases, whether identity groups are represented differently, and whether outputs reflect cross-cultural patterns. The study analyzes 89,253 outputs from 12 LLMs in English, French, and Chinese, spanning 18 occupations and three task conditions. We find that bias representation varies systematically across languages and tasks. Removing direct identity cues sharply reduces identity-label prediction in English and Chinese, but has a much smaller effect in French. Across all language-genre settings, the cultural context associated with the source language receives the highest average relevance score, with moderate agreement between automated and human ratings. However, the ability to identify the source language drops substantially after translation and again after masking names. Without these controls, multilingual audits may mistake surface cues for cultural understanding, leading to misleading conclusions about cross-cultural variation and bias. Our audit offers a practical framework for separating such shortcuts from more meaningful cross-cultural patterns.

arXiv

Unlearning Is Not Just Erasing: Temporal Decoupling via Generation Inequality

Xunlei Chen, Qirui Ye, Yuang Li, Yi Gong, Zhaokun Wang, Wenyi Li, Shiyao Guo, Jinyu Guo

Large language models (LLMs) require effective unlearning to address privacy regulations and safety concerns. However, achieving precise forgetting without compromising general utility remains challenging. Existing sequence- and token-level methods penalize target outputs without modeling their context-dependent retrieval paths, which can disrupt linguistic structure or suppress benign knowledge. We present ADU, a fine-grained, training-based framework that shifts unlearning from token erasure to contextual attention-pathway decoupling. Exploiting the functional distinction between local and global attention heads, ADU identifies preplan positions that retrieve persistent sensitive anchors and fixes their candidate paths under the original model. It then trains attention-projection adapters to suppress attention mass along these paths while preserving local-attention structure and retain-set language modeling. Post-training activation exchange tests whether the modified attention-output module transmits the learned forgetting effect. ADU achieves the strongest aggregate performance among evaluated baselines on the TOFU and WMDP benchmarks, including a Forget Quality of (0.93) on TOFU. It preserves 87--98% of model utility (92.9% on average versus 81.9% for baselines) while reducing side effects in benign contexts.

arXiv

PatchWrite: One Line, Not One Section -- Compile-Gated, Validity-Preserving Editing for AI-Drafted Manuscripts

Weiwei Yang

Automated manuscript pipelines often regenerate an entire section to repair a local defect, allowing unrelated metrics and citations to change even when the resulting PDF still builds. PatchWrite instead constrains how candidate edits become committed manuscript states: it reuses bounded EDIT N M editing and rollback, but tightens compilation acceptance with fatal-log checks and adds evidence locks that require every cited key and experimental numeric token to be attested by a reference registry or experimental log. Candidates that fail either check are rejected and the previous HEAD is retained. On a 24-manuscript x 8-fault oracle stress test (768 jobs, evenly split between compile-breaking and content-only faults), whole-slot rewriting mutated an unrelated "12-layer" line in every case (0/192 preserved; numeric Jaccard 0.6667), whereas PatchWrite preserved it in 192/192 cases. Removing the compile gate reduced acceptance to 0, while removing the evidence gate allowed a hallucinated citation to pass. The same pattern held across all eight faults. To test the protocol with generation rather than oracle edits, we reran the 192 jobs with the writer model proposing the edits. The model's candidates were accepted in 75% of cases; nearly all rejections came from one reproducible failure mode in which the model attempted to delete a line using an empty replacement unsupported by the current grammar. Every accepted candidate passed both gates, and 93.75% fixed the injected fault; the remaining cases involved a technically valid but sentence-inappropriate citation and one markup-changing near-miss. In a blind evaluation of sixteen PDF pairs, both raters preferred PatchWrite for preserving lab-grounded facts (C1 Likert 5.0 vs. 2.0), while rating prose quality nearly identically. Logs from 193 in-product drafting tasks show the same classes of failures occurring in practice.

Keep browsing

More papers from top AI conferences

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