paperlist.ai

AI 顶会论文

AI 顶会论文库

检索 30+ 个顶会的论文，包括 NeurIPS、ICLR、ICML、CVPR、ACL、AAAI 等。输入关键词筛选，打开任意一篇查看摘要、作者和链接。

arXiv

When LLM Meets Tree Search: A Systematic View of Inference as Search in Large Language Models

Jiaqi Wei, Xiang Zhang, Yuejin Yang, Wenxuan Huang, Juntai Cao, Sheng Xu, Xiang Zhuang, Zhangyang Gao, Muhammad Abdul-Mageed, Laks VS Lakshmanan, Chenyu You, Wanli Ouyang, Siqi Sun

As pretraining scaling laws approach saturation, Test-Time Scaling (TTS) has emerged as an important direction for improving reasoning by allocating inference-time compute to a fixed model prior. Viewed at a high level, TTS reframes inference as search over a space of partial reasoning states. While Chain-of-Thought (CoT) exposes intermediate steps, common instantiations rely on single-trajectory decoding, limiting recovery from early errors and exploration. This survey systematizes recent progress in tree-search-based reasoning, viewing inference as instance-specific optimization rather than decoding. We trace the evolution from uninformed search to Monte Carlo Tree Search (MCTS), highlighting how sampling-based control supports principled exploration-exploitation trade-offs. To unify a fragmented literature, we introduce a Unified Design Space spanning search topology, evaluation signals, and control dynamics, and advocate a standardized compute-reporting abstraction to make compute-accuracy trade-offs explicit and comparable.

arXiv

Quantitative Evidence Mining for Plausibility-Aware Biomedical AI

Negin Sadat Babaiha, Stefan Geissler, Marie-Christine Simon, Martin Hofmann-Apitius, Marc Jacobs

Biomedical artificial intelligence (AI) systems increasingly extract, organize, and reuse scientific claims from literature, clinical trials, and regulatory documents. But automatic extraction alone does not make a claim reliable evidence: a claim becomes useful only when it can be traced to its source, linked to the quantitative details that support it, and read within its biomedical context and uncertainty. This matters as large language models (LLMs) and increasingly autonomous systems drive evidence synthesis, knowledge graph (KG) construction, and decision support. Many text-mining and LLM pipelines remain relation-centric: they capture entities and relations such as Drug--TREATS--Disease, but drop the dose, effect size, population, comparator, uncertainty, and conditions under which a claim holds. Such relations can look actionable yet remain hard to verify, compare, or reuse. In this perspective, we argue for a shift toward quantitative evidence mining---extracting values, units, measured entities and properties, context, uncertainty, provenance, and plausibility as structured evidence units that populate evidence-aware KGs and can be checked for source grounding, unit consistency, completeness, and biological plausibility. We outline a framework for plausibility-aware AI that treats extracted claims not as final answers but as auditable evidence objects, making clear what was measured, how much it changed, in which setting, with what uncertainty, and from which source. The central risk is not only incorrect extraction, but claims that look like evidence while lacking the structure needed to trust them.

arXiv

Using Grounded Theory for Agent Behavior Analysis at Scale

Zhuoran Lu, Yangyang Yu, Zhuoyan Li, Yibo Meng, Nan Jiang, Chengxi Zang, Jie Gao, Ziang Xiao

Understanding agent behavior requires methods that scale to thousands of trajectories and surface new patterns in long, often unfamiliar tasks where pre-built classifiers fall short. We propose to bring grounded theory into agent trajectory analysis: a six-decade-old qualitative method from the social sciences, with a principled saturation criterion and an auditable trail from data to theory. We propose AutoTraceGT (Automated Trace analysis through Grounded Theory), the first multi-agent pipeline that automates grounded theory on agent trajectories. It iteratively performs open, axial, and theoretical coding until saturation, producing a behavioral taxonomy tailored to each task. Across six trajectory corpora, AutoTraceGT produces codebooks that recover 73-91 percent of the failure modes in human-annotated taxonomies and surface additional patterns that those taxonomies miss. The emergent theoretical narrative aligns with prior expert accounts. Used as a deductive feature space, the codebook outperforms zero-shot and few-shot LLM baselines on downstream failure prediction. These results suggest Grounded Theory offers a scalable analytic tool for ML researchers and agent developers studying what agents actually do.

arXiv

Kathleen Remembers: Length-Invariant One-Shot Recall Without Attention

George Fountzoulas

Recurrent, attention-free sequence models share a structural weakness: a fading state cannot perform exact recall of something seen once, far in the past. We add to the Kathleen trunk a second memory layer -- a "notebook": a fixed-key holographic (HRR) associative store with a learned local write gate, a self-gating raw read, and write-triggered forgetting -- 25K parameters that attach to the logits of any trunk. (1) Mechanism: on a controlled needle-in-haystack task the notebook reaches 80-82% one-shot recall at 4x the training length, where the bare trunk scores ~4% and a parameter-matched attention head scores 100% inside its training length and 0% beyond it. Addressing is length-invariant by construction; the untrained memory alone recalls at 90% accuracy identically at 512, 2048 and 4096 bytes. Because the store is a linear superposition, two capabilities follow from arithmetic alone: selective unlearning (one subtraction erases one fact to chance, retained facts unharmed) and per-token attribution (counterfactual erasure names the source fact of every correct byte, 100% provenance). (2) Real text: on WikiText-2 bytes the notebook improves prediction of repeated rare words by +0.15-0.27 bits/byte, the gain growing with the distance between mentions and holding zero-shot at 4x training length; write-triggered forgetting eliminates memory pollution at 8x length (first-mention cost +0.33 -> -0.004). (3) Scope and scale: a parameter-matched attention head does generalize on natural-text repetition, so the notebook's claim is exact recall at O(L); on a WikiText-103 ladder (8 to 512 MB) the zero-shot repeat gain rises monotonically. All experiments are pre-registered, seeds reported, and reproducible on a single free-tier GPU.

arXiv

Beyond Consensus: Downward Bias and Role Asymmetry in Multi-Agent LLM Judges for Subjective Evaluation

Minsoo Song, Chanwoo Kim, Sugyeong Eo, Chanjun Park

Multi-Agent Debate (MAD) has been widely adopted to improve LLM-based evaluation by prompting multiple agents to negotiate and reach a consensus. However, for subjective rubric-based scoring, inter-agent agreement does not guarantee alignment with human judgments. In this paper, we compare a single-judge baseline against a consensus-based MAD protocol on subjective evaluation tasks and design three ablations to isolate the impact of role prompting, multi-round interaction, and explicit score sharing. Evaluations across six LLMs show that the single-judge baseline achieves the strongest human alignment on average across six judge models, whereas MAD shows degradation in human alignment on both tasks. Our ablations demonstrate that this performance drop stems primarily from asymmetric role prompting rather than the interaction itself. Specifically, assigning a strict judge role introduces a systematic downward bias that the consensus process fails to correct. The central finding is that this bias reflects strict-stance dominance beyond averaging: the consensus score falls well beyond the arithmetic midpoint of the standalone strict and lenient conditions, rather than averaging them out. Removing role asymmetry (Symmetric MAD) largely recovers baseline performance, while masking peer scores widens inter-agent disagreement on average and worsens average human alignment. These findings demonstrate that multi-agent consensus can enforce artificial agreement at the expense of true human alignment, revealing a structural limitation in consensus-style, role-specialized MAD protocols for subjective scoring.

arXiv

Auditing MCQA Benchmarks through Probability Landscapes

Minsoo Song, Chanjun Park

As Large Language Models rapidly advance, performance on standard multiple-choice question answering (MCQA) benchmarks is reaching saturation. While the community has responded by developing increasingly difficult datasets, validating question quality and filtering flawed items remains a labor-intensive process. To provide a scalable diagnostic approach, we propose a two-component probabilistic framework for auditing MCQA benchmarks using model output distributions. First, for benchmark-level analysis, we characterize the probability landscape using the top prediction probability ($P_{top1}$) and normalized residual entropy ($H_{norm}$), summarized globally by Mean Pairwise Distance (MPD). Second, for item-level diagnostics, we introduce noise injection to reduce meaningful distractor competition, enabling us to flag candidate items for targeted human review and categorize residual failure patterns. Across four MCQA benchmarks, our landscape analysis reveals benchmark-level differences in model confidence and residual option competition. Concurrently, our noise-injection method flags potentially actionable item-level issues, showing alignment with expert error annotations from MMLU-Redux. These results suggest that our probability-based framework provides a lightweight audit lens for comparing macro-level benchmark structure and prioritizing individual items for targeted human review.

arXiv

Will the User Ever Know? Covert Indirect Prompt Injection Attacks on Tool-Using LLM Agents

Yunseok Lee, Yunji Kim, Woojin Lee

As LLM agents take real-world actions through tools, indirect prompt injection (IPI) has emerged as a serious threat. The standard metric, Attack Success Rate (ASR), counts whether an injection succeeds but ignores what the user notices in the agent's final response. Looking at successful injection traces, we find two distinct outcomes: the agent executes the injection while returning an otherwise normal response, or reports the injected action in its final response, giving the user a chance to notice. We call these covert and overt successes. From the user's perspective, we decompose ASR into the Covert Success Rate (CSR), counting successes leaving no trace in the final response, and the Overt Success Rate (OSR), counting successes the user can detect. To understand what drives the gap, we analyze successful trajectories and find that the agent's behavior after the injection separates covert from overt: covert traces hand control back to the user task before ending, while overt traces end at the attack itself. This split follows from the ReAct format, where the final response summarizes the most recent action. Building on this observation, we propose ICoA (Induced Covert Attack), an IPI attack designed to induce covert outcomes by steering the agent back to the user task after executing the injection. Across four target models on AgentDojo, ICoA achieves the highest CSR, with gains of 3.79-12.01 percentage points over the strongest baseline.

arXiv

Do Small Models Use the Law You Give Them? Measuring Context Use on a Bilingual Bangladesh Legal Benchmark

Moniruzzaman Mahadi, Abrar Mohammed Tanzim Alam, Sayma Siddika Monalisa, Mir Mohammad Asif Abdullah, Swakkhar Shatabda, Md Adnan Arefeen

Fine-tuning can improve legal question-answering accuracy without improving how models use law supplied in context. We study this distinction in bilingual Bangladeshi legal QA, where observed errors can arise from answer scoring, retrieval, or failure to use relevant law. We construct a hierarchy-preserving statutory corpus, 2,165 reviewed bilingual fine-tuning examples, and a 150-item supplied-law control. We evaluate six instruction-tuned models: Llama-3.2-1B, Llama-3.2-3B, Qwen3.5-0.8B, Qwen3.5-2B, Qwen3.5-4B, and Gemma-4-E2B, with three LoRA seeds per model. To separate effects, we combine constrained option-letter scoring, cyclic option rotation, and controlled removal of the governing provision. On 398 Bar Council outputs, an exact-line parser attributes an accuracy gain of 50.0\% to the Qwen3.5-2B seed-42 adapter, whereas option scoring yields only $3.0\%$. For Gemma-4-E2B, the two scoring methods favor different systems. When the governing provision is guaranteed to be present, five of six reference models improve by $14.7\%-19.3\%$ under the four-order criterion. Removing that provision reduces accuracy by $8.0\%-15.3\%$ for models and by $13.8\%-14.9\%$ points for their adapters. However, difference-in differences estimates show no increase in reliance on the governing provision after fine-tuning. Results show that legal adaptation claims require separating scorer, retriever, and model effects. Our Code and data are available at https://anonymous.4open.science/r/bangladesh-legal-qa-11E3

arXiv

Sequential Trajectories and Simultaneous Blending: Multi-Emotion Modeling for Instruction-Following TTS

Yan Zhou, Yun Hong, Yang Feng

Natural-language instructions enable flexible control of synthesized speech, yet emotional TTS systems primarily model a single utterance-level affect, leaving multi-emotion control underexplored. We study two complementary multi-emotion TTS tasks: emotion trajectory, which spans several ordered affective stages, and emotion blending, in which multiple emotions coexist throughout an utterance. These tasks expose a supervision mismatch: supervised fine-tuning (SFT) does not explicitly evaluate emotion features, while single-emotion rewards provide neither structure-aware feedback for trajectory completion nor pair-aware feedback for blending. We introduce HybridEmo, a post-training framework that initializes both tasks with SFT and then aligns the speech-token policy through Group Relative Policy Optimization using a sample-aware hybrid reward. For trajectory samples, segment-aligned consistency combines average and weakest-stage evidence to preserve the correctness and completeness of prescribed stages. For blending samples, a GMM-based reward combines frame-level support from the union of target-emotion anchors in an offline emotion space with an utterance-level weaker-target margin. Both branches share an ASR reward and are routed within a unified policy. On MultiEmo-Test, HybridEmo significantly improves trajectory correctness and blending intensity, without a noticeable degradation in speaker similarity. Human evaluation prefers HybridEmo to CosyVoice 3 and EmoVoice-0.5B, with nearly balanced preferences against Qwen3-TTS.

arXiv

Ignorance or Incompetence? Constructing Knowledge-Gated, Verifiable Tasks for LLM Agents

Hanlin Tian, Minhao Li, Yu Mi, Sihan Zhu, Zhao Yang, Yuxiang Wang, Hongquan Zhu, Qiufei Hu

Professional agent tasks often depend on conventions that are absent from public corpora, yet benchmarks rarely control whether an agent has access to those conventions. We introduce a knowledge-gated task-construction protocol that separates a task instruction from a compact artefact containing private conventions, reference tables, and utility operators. Construction-time provenance, byte-identical task instructions across the provided- and withheld-artefact conditions, leak audits, and executable witnesses make dependence on the artefact explicit and testable. Across fifteen calibration tasks, one frontier agent configuration achieves a 68.0% pass rate with the artefact and 0% without it; on one task, a plausible but incorrect artefact also yields 0% across five trials. Deterministic solvers and rule corpora provide exact ground truth for structured tasks, while named criterion-level rubrics support outputs that cannot be checked by a single executable oracle. A configuration-relative calibration screen retains seven tasks satisfying our five-trial empirical knowledge-gating screen. These experiments validate the behavior of the construction protocol; they do not establish that the retained tasks improve post-training. We publicly release part of the task suite and supporting tooling at https://github.com/DatagridsAI/Knowledge-Gated-Task-Construction.

arXiv

On the Design of Qwen3.8-Next Architecture: Evaluation, Efficiency, and Training Stability

Zihan Qiu, Zekun Wang, Xiao Li, Yanpeng Li, Yang Xu, Yixuan Wang, Huaqing Zhang, Rui Men, Bochao Mao, Chengruidong Zhang, Fan Zhou, Hao Luo, Haofeng Huang, Haoran Lian, Haoyan Huang, Hongqing Chen, Jianwei Zhang, Jing Xu, Junjie Wang, Langshi Chen, Liangyu Wang, Linlang Jiang, Man Yuan, Minmin Sun, Peng Jin, Siqi Zhang, Siyu Wang, Xingzhang Ren, Yakai Wang, Yi Zhang, Yiming Dong, Yizhong Cao, Yubo Ma, Yunfei Mao, Bo Zheng, Dayiheng Liu

We describe the architecture and ablations of Qwen3.8-Flash-Next, a sparse mixture-of-experts model with 125B parameters, 6B activated per token, and additional 51B parameters of n-gram embedding tables held off the accelerator. On fourteen pre-training benchmarks the model leads the 397B-A17B predecessor on eight and trails it on the rest by at most 2.6 points, at 1/3 the activated parameters, 1/3 the training tokens, and roughly 1/9 the training FLOPs. Token mixing uses a layer-wise hybrid of Gated DeltaNet (GDN) and global attention, with one full-attention layer in every four; at continued-pretraining time those full-attention layers are replaced by Qwen Sparse Attention (QSA), which scores context at micro-block granularity with a compressed lightweight indexer. The residual stream is widened to four branches and read through an elementwise gate, a design we call the Gated Residual (GR). Capacity is added outside the backbone by a single n-gram embedding layer whose tables are prefetched from host memory. We evaluate every candidate change along three axes: loss together with downstream benchmarks; the cost of the change in training, prefill and decode; and its effect on the optimal hyperparameters and training stability. Loss and downstream accuracy do not always move together: enlarging the n-gram vocabulary lowers loss monotonically while downstream accuracy saturates. The architecture and the Muon optimizer together shift the optimal learning rate and batch size upwards, render batch-size warmup unnecessary, and substantially improve stability under stress tests. Loss, benchmarks, efficiency and stability form one design problem. Solved jointly, they yield a recipe that is simultaneously more efficient, more capable and more stable.

arXiv

Beyond Token-Level Guidance: Inference-Time Alignment of Specialized LLMs via Cross-Family Representation Steering

Jin Gan, Xin Li, Jun Luo

Large language models (LLMs) finetuned for specialized domains represent crucial high-impact applications. Inference-time alignment improves safety degraded from specialization finetuning without requiring substantial computational resources, complementing finetuning-based methods with an easy-to-use, plug-and-play solution. However, existing inference-time methods fail to reliably improve safety without disrupting domain capability. We identify the root cause as complementary expertise orthogonality: specialized base models and general-domain guidance models have orthogonal competencies, making the guidance signal unreliable for specialized generation. This primarily manifests as stop token interference, where the guidance model's tendency toward continuation overrides the base model's decision to stop, burying correct answers under guidance-induced continuation. To address this problem, we propose CREST, an inference-time alignment method that steers base model hidden representations using safety directions extracted from a guidance model of any family, avoiding token-level structural limitations entirely. CREST improves safety where specialization has weakened it while preserving both domain-specific capability and the safety of already well-aligned models, outperforming baselines by up to 22.2\% on safety benchmarks. Our code is available at: https://github.com/DecayingSeart/CREST.

arXiv

Context Staircase: Signature-Aligned Dynamics of Token Embeddings under Small Initialization

Junjie Yao, Liangkai Hang, Zhi-Qin John Xu

Token embeddings are the basic representational units that connect discrete tokens with continuous computation in language models. Although modern language models learn embeddings from random initialization through gradient-based training, the dynamical mechanism by which meaningful embedding structures emerge remains unclear. In this work, we identify that the evolving embedding structures are closely related to token-conditioned label and contextual distributions, which we formalize as probability signatures. We observe a progressive learning process, which we term Context Staircase: embeddings learn the low-order statistic signatures of the data before the high-order ones. More specifically, we observe that early in training they align with the simplest, context-free signature linking a token to its label, and as training proceeds, they progressively reflect signatures involving more and more context tokens. We then analyze the gradient flow of embeddings under small initialization to explain this phenomenon, deriving embedding evolution equations for feed-forward and self-attention architectures. We further extend these observations to real language-model training. Finally, we show that these embedding structures play an important role in both task learning and the incorporation of semantic structure into the embedding space. Overall, our results provide a dynamic explanation of how data statistics and architecture jointly shape token embeddings in language models, and reveal an implicit bias in the space of data statistics: training proceeds from simpler, low-order statistical relations toward increasingly complex, context-dependent ones.

arXiv

Lazy Grounding: Attacking Search Agents with Factual Evidence

Yulin Zhang, Yukun Huang, Sanxing Chen, Tianyi Lin, Ziang Yang, Xunjian Yin, Bhuwan Dhingra

Search agents mitigate hallucination by grounding their answers in retrieved web results. However, retrieval-based approaches also introduce an attack surface: agents may cite misinformation from poisoned search corpora containing false or malicious documents. We demonstrate that, in some cases, search agents' reasoning and responses may be steered by completely factual but distracting information. We refer to this failure as lazy grounding. We expose lazy grounding by injecting nearby evidence from answer-changing rewrites of benchmark questions into the search corpora. Each document contains factual evidence that supports a neighboring rewritten question but is retrieved for the original question. Across 12 model-benchmark pairs, the attack causes the accuracy of search agents' responses to drop by 5.9 points on average and by up to 17.3 points, while inducing nearby-answer adoption in every setting. The effect is even stronger when nearby evidence appears later or is more answer-shaped. Our results show that robust search agents must defend against not only misinformation but also the misapplication of factual evidence. The code is publicly available at https://github.com/frankyzha/lazy-grounding.

arXiv

AIA$^{2}$: Attribute-Agnostic Imbalance Augmentation for Subgroup Robustness

Hanshu Rao, Guangzeng Han, Xiaolei Huang

Attributes describing data content and context can induce diverse imbalance patterns that go beyond label imbalance alone. However, existing studies primarily address label imbalance while overlooking data attributes, such as topics and demographics, which can induce meaningful subgroup structure while causing model degradation on underrepresented subgroups. We propose Attribute-Agnostic Imbalance Augmentation (AIA$^{2}$), a framework for improving model robustness under varying subgroup imbalances without explicit subgroup annotations. AIA$^{2}$ automatically discovers varying imbalances via latent semantic distributions, obtains slices with both learning difficulty and subgroup imbalance deficits, and deploys a large language model (LLM) for subgroup-aware imbalance augmentation. We have evaluated AIA$^{2}$ on 5 popular corpora with rich domains and their attribute values, covering social issues and diverse topics. Results show improved performance on the lowest-performing subgroups and consistent gains over competitive baselines. Ablation studies confirm complementary contributions from each component, and additional analyses show that AIA$^{2}$ provides a practical and consistent way to improve worst-group robustness under data subgroup imbalance. Code is available at https://github.com/trust-nlp/AIA2-Subgroup-Robustness.

arXiv

Read the Room, Read the Image: Understanding Indirect Speech Acts in Multimodal Visual Contexts

Jaehee Kim, Ji Hoon Chung, Seoyoon Park, Unsol Kim, Kyungwon Park, Ji Hak Kim, Yi-Jun Chen, Hansaem Kim

Indirect speech acts (ISAs) require pragmatic reasoning over context, as directive intent can- not be inferred from surface form alone. Prior text-based studies and existing multimodal benchmarks largely overlook this requirement, focusing instead on explicitly encoded context or perceptual recognition, and thus underex- plore context-dependent pragmatic understand- ing, particularly in high-context languages such as Korean. We introduce READI, a multimodal benchmark for evaluating ISA understanding through integrated reasoning over visual con- text and dialogue. READI models graded in- directness grounded in pragmatic theory and formulates the task as vision-based pragmatic question answering (V-PQA), supporting cross- lingual evaluation in English and Korean. Ex- periments show that even state-of-the-art multi- modal models struggle with visually grounded indirect speech acts, with performance declin- ing as indirectness increases, underscoring the need for benchmarks that explicitly target con- textual pragmatic reasoning.

arXiv

Using Prosody to Predict Syntactic Structure

Junghyun Min, Alex Warstadt, Tamar I. Regev, Tiago Pimentel, Ethan Gotlieb Wilcox

While it is well-established that prosody carries crucial cues for syntactic structure, the degree and nature of correspondence between these two domains remains contested. We investigate the syntax-prosody interface through an information-theoretic lens, quantifying the interaction between prosodic features and syntactic representations as their mutual information. We provide a general-purpose framework for estimating this quantity over large speech-text corpora using multimodal language models. Our framework is structure-agnostic and modular, insofar as it can be used to measure the contributions of individual prosodic features or components of structure. We evaluate the syntax-prosody relationship for two features (word duration and inter-word pauses) across two domains--read audiobooks and spontaneous conversations--both in English. Our results demonstrate that prosody contains measurable syntactic information, with prosodic features reducing syntactic uncertainty in spontaneous conversations by up to 10.2%. Our findings offer new empirical support for several theoretical accounts of the syntax-prosody interface.

arXiv

Stratified Consistency Distillation for Natural Language Formalization

Zhichao Hou, Ferhat Erata, Joe Lilien, MohamadAli Torkamani

Neurosymbolic reasoning has shown promising success in addressing complex reasoning tasks by combining large language models (LLMs) and symbolic solvers. While this approach shows promise, a fundamental challenge remains: improving the accuracy of translations from natural language to logical formulas. Current methods predominantly rely on prompt engineering, which is difficult to scale across different domains and input formats. Drawing inspiration from the success of fine-tuning in other model adaptation and alignment applications, we propose a fine-tuning-based Stratified Consistency Distillation approach: (1) We generate K logical translations per input using a frontier LLM and cluster them by semantic equivalence (2) Based on the entropy level, we apply majority voting (low entropy), LLM-as-a-Judge (medium entropy), or unification/abstention (high entropy), and (3) fine-tune a smaller model using the selected pseudo-labels. Our experiments show significant and consistent improvements in both Pass@K and our novel Equivalent Logical Similarity metrics, demonstrating the potential of advancing logical translation through consistency distillation.

arXiv

Beyond Surface Forms: Symbolic Edits as a Test for Logical Reasoning with LLMs

Ramya Keerthy Thatikonda, Wray Buntine, Ehsan Shareghi

Logical reasoning with large language models (LLMs) is a critical capability, as it reflects a system's ability to correctly deduce hypotheses from a given context using faithful deductive processes. However, LLM reasoning has often been shown to be sensitive to small surface-level variations in problem formulation, raising questions about whether models truly follow the underlying logical structure. Studying this behavior is challenging because the symbolic components of logical problems, such as operators and predicates, are difficult to systematically manipulate in natural language. We introduce a tool-driven framework for generating controlled, label-preserving edits to logical reasoning problems. Our method operates on symbolic representations of first-order logic and constraint satisfaction problem tasks, enabling targeted modifications to logical operators and other structural components before translating them back into natural language. Using this framework, we evaluate various LLMs under cumulative and individual operator edits and analyze their behavior in response to these changes. Our quantitative and qualitative analyses show that LLM reasoning behavior under controlled operator edits is inconsistent, regardless of model size or family: models sometimes adapt correctly to structural changes but often fail to track their logical consequences. The results from this automated stress test enable an evaluation of language models across different dimensions and help measure the reliability of their reasoning.

arXiv

PaperBanana-Interact: Scientific Diagram Refinement with Multi-Turn Human Feedback

Xueqing Wu, Ashwin Balasubramanian, Bingxuan Li, Dawei Zhu, Kai-Wei Chang, Yale Song, Yiwen Song, Rui Meng, Tomas Pfister, Nanyun Peng

Recent efforts have aimed to automate scientific diagram generation from paper content (Lin et al., 2026; Zhu et al., 2026a). However, fully satisfying an author's visual and communicative preferences in a single turn is challenging: in our formative user study (N = 14), all participants requested further revisions after viewing an initial draft, and 86% of them rated the refined diagrams as more satisfactory. Despite the clear demand, the multi-turn workflow remains largely underexplored. To bridge this gap, we present MTPaperBananaBench, a benchmark for multi-turn diagram generation containing 292 images annotated with 3,518 user requirements. To reduce expensive human studies and enable scalable benchmarking, we construct a user simulator that, at each turn, identifies unsatisfied requirements and converts k of them into natural language feedback. Evaluating both requirement satisfaction and overall diagram quality reveals two key failure modes shared across baseline multiturn systems: (1) quality drift, where diagram quality progressively declines over turns, and (2) forgetting, where previously implemented features are lost in subsequent turns. To address these issues, we introduce PaperBanana-Interact, a multi-agent system that refines diagrams via an internal critique-and-refine loop. PaperBanana-Interact consistently improves rather than degrades diagram quality across turns, outperforming baselines by 11.9-18.6 points in quality score and reducing forgetting by 3.7-6.2 points.

继续浏览

更多 AI 顶会论文

30+ 个顶会，按会议、年份和主题浏览。

浏览全部会议

进入任一会议和年份，查看完整论文列表。

全部会议 →

搜索论文

按标题、作者或关键词，在 21 万+ 篇论文里搜索。

搜索论文 →