Reinforcement learning with verifiable rewards (RLVR) typically improves pass@1 but can fall behind its base model at large sampling budgets $k$, commonly interpreted as sharpening existing solutions. Because adjacent pass@$k$ estimates reuse generations, pointwise error bars cannot establish a crossover. We formulate paired inference on the RL-minus-base difference curve: a prompt-level multiplier bootstrap yields a one-sided dominance test and simultaneous confidence bands requiring both an early gain and a later loss. Among nine public RLVR checkpoints at 3k-16k token limits, only one narrowly meets this criterion before multiplicity adjustment; asymmetric truncation can create, mask, or reverse crossovers. On 1,060 deduplicated competition problems at 32k tokens, two 1.5B models exhibit certified crossovers around $k\approx10$-$60$, while two 7B models show no supported loss. A paired variance decomposition shows why adding prompts improves power more than deeper sampling. Prompts with identical base success probability diverge after RLVR, with dispersion replicating across independent generation halves and violating Lipschitz slope bounds for reasonable deterministic maps. Modeling RLVR as a Markov kernel, Jensen's inequality shows that conditional dispersion depresses pass@$k$ for $k\ge2$. Once dispersion is modeled, the conditional-mean log-odds slope is below one in every pair: RLVR improves hard prompts on average, while late-$k$ loss comes from a 2%-3% minority of base-solved prompts collapsing onto a repeated modal wrong answer and resisting prefix rescue up to 4,096 tokens. In controlled runs from one base model on 3,000 new problems, all four GRPO arms fall behind near $k\approx200$ and remain behind through $k=4096$, while SFT stays significantly ahead throughout.
Zian Liu, Xiaoxi Meng, Viet Vo, Siqi Ma, Chao Chen, Dongxi Liu, Jun Zhang, Shigang Liu 等 10 位作者
Binary similarity detection is a critical technique that has been applied in many real-world scenarios where source code is not available, e.g., bug search, malware analysis, and code plagiarism detection. Existing works are ineffective in detecting similar binaries in cases where different compiling optimizations, compilers, source code versions, or obfuscation are deployed. We observe that all the cases do not change a binary's key code behaviors although they significantly modify its syntax and structure. With this key observation, we extract a set of key instructions from a binary to capture its key code behaviors. By detecting the similarity between two binaries' key instructions, we can address well the ineffectiveness limitation of existing works. Specifically, we translate each extracted key instruction into a self-defined key expression, generating a key-semantics graph based on the binary's control flow. Each node in the key-semantics graph denotes a key instruction, and the node attribute is the key expression. To quantify the similarity between two given key-semantics graphs, we first serialize each graph into a sequence of key expressions by topological sort. Then, we tokenize and concatenate key expressions to generate token lists. We calculate the locality-sensitive hash value for all token lists and quantify their similarity. %We implement a prototype, called SemDiff, consisting of two modules: graph generation and graph diffing. The first module generates a pair of key-semantics graphs and the second module diffs the graphs. Our evaluation results show that overall, SemDiff outperforms state-of-the-art tools when detecting the similarity of binaries generated from different optimization levels, compilers, and obfuscations. SemDiff is also effective for library version search and finding similar vulnerabilities in firmware.
Large language model agents have enabled software development through natural-language interaction, but extending this paradigm to end-to-end game development remains challenging. Building a playable game requires coordinating code, scenes, assets, animations, physics, and configurations, while allowing developers to validate their integrated behavior through interactive testing. We introduce VibeGame, a framework for creating complete 2D game projects from natural-language prompts and optional reference images. Its AI-native game engine represents project state as schema-validated structured text and provides frame-synchronous runtime control, enabling agents to inspect game states, execute actions, and verify behavior at their own pace. Built upon this interface, we design an Adversarial Agent Team of eight specialized agents to separate implementation from review, so that problems are found and fixed through multi-round loops. For each conducted and accepted project, we further introduce a training-free self-evolution mechanism to distill it into reusable skeletons, modules, and collaboration contracts for future development. We evaluate prompt-assets-to-game (PA2G), where agents are given an asset library, and prompt-to-game (P2G), where agents generate their own art. On a GameCraft-Bench subset, VibeGame in P2G consistently outperforms a range of agent harnesses in either P2G or PA2G with the same underlying model, including GPT-6.1-Sol.
AI 速览：将多模态智能体的开放式推理改为从预设自然语言候选集中选择，以提升推理效率并保持任务表现。
Multimodal agents commonly generate free-form reasoning before each action. For small models, limited model capacity can result in lengthy reasoning that provides little useful guidance for action generation while incurring substantial inference cost. To address this challenge, we introduce Selection-based Structured Reasoning (SSR), a framework that reformulates reasoning as selection instead of open-ended generation. SSR represents recurring high-level reasoning as pre-specified, reusable natural-language candidates. At each turn, the model selects from these reasoning candidates based on their likelihoods given the current context, without requiring an auxiliary task head. Using pre-specified reasoning traces enables parallel scoring, where teacher-forced prefilling computes token likelihoods concurrently within and across candidates using a shared context KV cache. We evaluate SSR on seven multimodal search benchmarks using 2B and 4B models. Across multiple reinforcement learning objectives and supervised fine-tuning, SSR delivers significant efficiency gains without sacrificing task performance. SSR achieves an average success rate competitive with leading search agents of the same scale, while reducing per-turn reasoning latency by over 90% and total per-question model inference latency by 28-54%. Project page: https://zfy0314.github.io/ssr-webpage/.
The opacity of massive pretraining corpora in Large Language Models (LLMs) raises significant privacy and copyright concerns, making pretraining data detection a critical challenge.Existing state-of-the-art methods typically rely on token likelihoods, yet they often overlook the gap between the target token and the model’s top-1 prediction, as well as local correlations between adjacent tokens.In this work, we propose Gap-K%, a novel pretraining data detection method grounded in the optimization dynamics of LLM pretraining. By analyzing the next-token prediction objective, we observe that discrepancies between the model’s top-1 prediction and the target token induce strong gradient signals, which are explicitly penalized during training.Motivated by this, Gap-K% leverages the log probability gap between the top-1 predicted token and the target token, incorporating a sliding window strategy to capture local correlations and mitigate token-level fluctuations. Extensive experiments on the WikiMIA and MIMIR benchmarks demonstrate that Gap-K% achieves state-of-the-art performance, consistently outperforming prior baselines across various model sizes and input lengths.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Shenglai Zeng Michigan State University, Tianqi Zheng Amazon, Chuan Tian Amazon, Dante Everaert Amazon, Yau-Shian Wang, Yupin Huang Amazon, Michael J. Morais, Rohit Patki Amazon 等 13 位作者
Personalizing large language models (LLMs) to individual users requires incorporating extensive interaction histories and profiles, but input token constraints make this impractical due to high inference latency and API costs. Existing approaches rely on heuristic methods such as selecting recent interactions or prompting summarization models to compress user profiles. However, these methods treat context as a monolithic whole and fail to consider how LLMs internally process and prioritize different profile components. We investigate whether LLMs’ attention patterns can effectively identify important personalization signals for intelligent context compression. Through preliminary studies on representative personalization tasks, we discover that (a) LLMs’ attention patterns naturally reveal important signals, and (b) fine-tuning enhances LLMs’ ability to distinguish between relevant and irrelevant information. Based on these insights, we propose Attn-GS, an attention-guided context compression framework that leverages attention feedback from a marking model to mark important personalization sentences, then guides a compression model to generate task-relevant, high-quality compressed user contexts. Extensive experiments demonstrate that Attn-GS significantly outperforms various baselines across different tasks, token limits, and settings, achieving performance close to using full context while reducing token usage by 50 times.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Siyuan Gan, Jiaheng Liu Nanjing University, Boyan Wang nanjing university, Tianpei Yang Nanjing University, Runqing Miao China Mobile Communications Company Limited Research Institute, Yuyao Zhang China Mobile Communications Company Limited Research Institute, Fanyu Meng JIUTIAN Research, Junlan Feng 等 11 位作者
Large reasoning models (LRMs) have attracted much attention due to their exceptional performance. However, their performance mainly stems from thinking, a long Chain of Thought (CoT), which significantly increase computational overhead. To address this overthinking problem, existing work focuses on using reinforcement learning (RL) to train hybrid reasoning models that automatically decide whether to engage in thinking or not based on the complexity of the query. Unfortunately, using RL will suffer the the reward hacking problem, e.g., the model engages in thinking but is judged as not doing so, resulting in incorrect rewards.To mitigate this problem, existing works either employ supervised fine-tuning (SFT), which incurs high computational costs, or enforce uniform token limits on non-thinking responses, which yields limited mitigation of the problem.In this paper, we propose Thinking-Based Non-Thinking (TNT). It does not employ SFT, and sets different maximum token usage for responses not using thinking across various queries by leveraging information from the solution component of the responses using thinking. Experiments on five mathematical benchmarks demonstrate that TNT reduces token usage by around 50\\%$ compared to DeepSeek-R1-Distill-Qwen-1.5B/7B and DeepScaleR-1.5B, while significantly improving accuracy. In fact, TNT achieves the optimal trade-off between accuracy and efficiency among all tested methods. Additionally, the probability of reward hacking problem in TNT’s responses, which are classified as not using thinking, remains below $10\\%$ across all tested datasets.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Kerem Zaman Department of Computer Science, University of North Carolina at Chapel Hill, Shashank Srivastava University of North Carolina at Chapel Hill
Recent work, using the Biasing Features metric, labels a CoT as unfaithful if it omits a prompt-injected hint that affected the prediction. We argue this metric adopts a narrow notion of faithfulness and confuses unfaithfulness with incompleteness, the lossy compression needed to turn distributed transformer computation into a linear natural language narrative. On multi-hop reasoning tasks with instruct-tuned and reasoning models, many CoTs flagged as unfaithful by Biasing Features are judged faithful by other metrics, exceeding 50% in some models. With a new faithful@k metric, we show that larger inference-time budgets greatly increase hint verbalization (up to 90% in some settings), suggesting much apparent unfaithfulness is due to tight token limits. Using Causal Mediation Analysis, we further show that even non-verbalized hints can causally mediate prediction changes through the CoT. We therefore caution against relying solely on hint-based evaluations and advocate a broader interpretability toolkit, including causal mediation and corruption-based metrics. We do not claim all CoTs are faithful, only that the absence of hint words alone does not prove unfaithfulness.
尚无 AI 解读，展示原文摘要
cs.CL
Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表
Autoregressive (AR) visual generation relies on tokenizers to map images to and from discrete sequences. However, tokenizers are trained to reconstruct clean images from ground-truth tokens, while AR generators are optimized only for token likelihood. This misalignment leads to generated token sequences that may decode into low-quality images, without direct supervision from the pixel space. We propose VA-\boldsymbol \pi , a lightweight post-training framework that directly optimizes AR models with a principled pixel-space objective. VA-\pi formulates the generator-tokenizer alignment as a variational optimization, deriving an evidence lower bound (ELBO) that unifies pixel reconstruction and autoregressive modeling. To optimize under the discrete token space, VA-\pi introduces a reinforcement-based alignment strategy that treats the AR generator as a policy, uses pixel-space reconstruction quality as its intrinsic reward. The reward is measured by how well the predicted token sequences can reconstruct the original image under teacher forcing, giving the model direct pixel-level guidance without expensive free-running sampling.The regularization term of the ELBO serves as a natural regularizer, maintaining distributional consistency. VA-\pi enables rapid adaptation of existing AR generators, without neither tokenizer retraining nor external reward models. With only 1% ImageNet-1K data and 25 minutes of tuning, it reduces FID from 14.36 to 7.65 and improves IS from 86.55 to 116.70 on LlamaGen-XXL, while also yielding notable gains in the text-to-image task on GenEval for both visual generation model (LlamaGen: from 0.306 to 0.339) and unified multi-modal model (Janus-Pro: from 0.725 to 0.744).
尚无 AI 解读，展示原文摘要
cs.CV
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2026 · 正式发表
Autoregressive (AR) visual generators model images as sequences of discrete tokens and are trained with next token likelihood. This strict causality supervision optimizes each step only by its immediate next token, which diminishes global coherence and slows convergence. We ask whether foresight, training signals that originate from later tokens, can help AR visual generation. We conduct a series of controlled diagnostics along the injection level, foresight layout, and foresight source axes, unveiling a key insight: aligning foresight to AR models' internal representation on the 2D image grids improves causality modeling.We formulate this insight with Mirai (meaning "future" in Japanese), a general framework that injects future information into AR training with no architecture change and no extra inference overhead: Mirai-E uses explicit foresight from multiple future positions of unidirectional representations, whereas Mirai-I leverages implicit foresight from matched bidirectional representations.Extensive experiments show that Mirai significantly accelerates convergence and improves generation quality. For instance, Mirai can speed up LlamaGen-B's convergence by up to 10xand reduce the generation FID from 5.34 to 4.34 on the ImageNet class-condition image generation benchmark.Our study highlights that visual autoregressive models need foresight.
尚无 AI 解读，展示原文摘要
cs.CV
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition · 2026 · 正式发表