{"ok":true,"snapshot":{"date":"2026-09-17","capturedAt":"2026-09-17T14:15:41.696Z","total_papers":30,"raw_count":50,"papers":[{"paperId":"2609.06986","title":"Continual Learning Mechanisms Compose for Long-Horizon Memorization","summary":"Language models may need to internalize information that arrives over time and retain it through many subsequent updates. To study this challenge, we introduce long-horizon memorization, a setting in which a model learns 100 query-answer tasks through continual supervised fine-tuning without retaining earlier training examples or receiving task identifiers at inference. Sequential updates cause catastrophic forgetting, and no single continual learning mechanism we evaluate maintains strong retention at this horizon. We hypothesize that mechanisms addressing complementary sources of forgetting will be more effective when composed. We organize these compositions along two design dimensions. Data, function, and weight anchors specify what prior information each update should preserve, while …","authors":["Zheyuan Zhang","Alvin Zhang","Daniel Khashabi","Tianmin Shu"],"publishedAt":"2026-09-07T00:00:00.000Z","submittedAt":null,"upvotes":292,"num_comments":4,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.06986.png","hf_url":"https://huggingface.co/papers/2609.06986","arxiv_url":"https://arxiv.org/abs/2609.06986","github_repo":null,"github_stars":null,"ai_keywords":["catastrophic forgetting","continual supervised fine-tuning","long-horizon memorization","data anchor","function anchor","weight anchor","low-rank allocation","LoRA","merged LoRA","successive halving"]},{"paperId":"2609.16679","title":"AI for Games in the Foundation Model Era","summary":"Foundation models, alongside advances in learned game-world models, are reshaping AI across the game lifecycle. Beyond playing games, recent systems model players and game dynamics, support design and development, adapt player-facing experiences at runtime, and evaluate resulting artifacts. Yet these directions have evolved largely separately, obscuring which capabilities transfer across settings and which remain tied to particular games, engines, interfaces, or player populations. We organize the literature into six roles according to the immediate use of AI output: playing and acting; modeling players and games; designing games; building and maintaining games; generating and adapting at runtime; and testing and evaluating games. For each role, we examine what structure is supplied by th…","authors":["Meng Luo","Yanlin Li","Hao Li","Hongzhan Lin","Pengfei Zhou","Tianjie Ju","Ran Zhang","Yeying Jin"],"publishedAt":"2026-09-15T00:00:00.000Z","submittedAt":null,"upvotes":113,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.16679.png","hf_url":"https://huggingface.co/papers/2609.16679","arxiv_url":"https://arxiv.org/abs/2609.16679","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.14005","title":"StepAudio 3 Realtime Technical Report","summary":"Realtime spoken interaction demands deep reasoning, prompt responses, and fluid turn-taking. We present StepAudio 3 Realtime, an audio-language foundation model organized around a continuous listen-converse-think-act loop. Deep Perception captures rich acoustic cues to interpret user intent, while Seamless Duplex models synchronized audio streams to handle pauses, backchannels, and interruptions naturally. Crucially, we resolve the tension between deep deliberation and latency via Think-While-Speaking, executing private reasoning in parallel with spoken delivery. In reasoning mode, StepAudio 3 reaches a 73.0 macro average on StepAudioChat. With Think-While-Speaking, it achieves dialogue and reasoning performance comparable to dedicated reasoning models while speaking in real time. Further…","authors":["Bin Lin","Bo Zhao","Boyang Zhang","Boyong Wu","Chao Yan","Chen Geng","Chen Wu","Cheng Yi"],"publishedAt":"2026-09-12T00:00:00.000Z","submittedAt":null,"upvotes":101,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.14005.png","hf_url":"https://huggingface.co/papers/2609.14005","arxiv_url":"https://arxiv.org/abs/2609.14005","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.11873","title":"The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement","summary":"Recursive self-improvement (RSI) enables AI systems to turn experience and feedback into persistent changes that improve both their capabilities and the process of future improvement. We first use the Headroom-Closed Index (HCI) to reveal the problems of existing LLMs, then introduce the RSI concept and its development roadmap: from improvement-execution autonomy, improvement-strategy autonomy, experience-acquisition autonomy, and environment-adaptation autonomy, to recursive meta-improvement. Next we examine RSI across scenarios (e.g., scientific discovery, embodied intelligence, software engineering), highlighting their distinct requirements and development speeds. Drawing on diverse industry practices and preliminary empirical evidence, we connect RSI research with practical systems an…","authors":["Yi Duan","Ying Liu","Zirui Tang","Haodong Chen","Jun Zhou","Yumou Liu","Bangrui Xu","Yukai Wu"],"publishedAt":"2026-09-10T00:00:00.000Z","submittedAt":null,"upvotes":88,"num_comments":3,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.11873.png","hf_url":"https://huggingface.co/papers/2609.11873","arxiv_url":"https://arxiv.org/abs/2609.11873","github_repo":null,"github_stars":null,"ai_keywords":["recursive self-improvement","Headroom-Closed Index","LLMs","improvement-execution autonomy","improvement-strategy autonomy","experience-acquisition autonomy","environment-adaptation autonomy","recursive meta-improvement","scientific discovery","embodied intelligence"]},{"paperId":"2609.17488","title":"LimiX-2: A Contextual Mechanism Network Towards General Structured-Data Intelligence","summary":"We introduce LimiX-2, a new model in the LimiX family, developed through model and data scaling guided by our previously established scaling laws. LimiX-2 adopts the Contextual Mechanism Networks (CMNs) paradigm and is pretrained with Context-Conditional Masked Modeling (CCMM). CMNs shifts the organizing principle of in-context learning from target-centric prediction to mechanism-oriented joint modeling. Rather than centering the network on the p(y mid x, D_{context}) objective of conventional tabular PFNs, it is designed around learning p(x, y mid D_{context}), a context-dependent representation of the joint structure underlying data generation. Pretraining uses synthetic datasets generated by structural causal models (SCMs) spanning diverse graph structures, functional mechanisms, and o…","authors":["Xingxuan Zhang","Gang Ren","Hao Yuan","Hao Zou","Hongze Tan","Hui Wang","Jianhao Song","Jiansheng Li"],"publishedAt":"2026-09-15T00:00:00.000Z","submittedAt":null,"upvotes":83,"num_comments":4,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.17488.png","hf_url":"https://huggingface.co/papers/2609.17488","arxiv_url":"https://arxiv.org/abs/2609.17488","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.16034","title":"StepAudio 3 Music Technical Report","summary":"We introduce StepAudio 3 Music, a large-scale, long-form music generation model that supports explicit musical planning and open-domain text-controlled generation. The StepAudio Music Tokenizer represents audio as a 50-Hz stream from a 65536-entry single codebook, using semantically informed self-supervised and multi-task training to preserve musical structure and reconstruction-relevant information. A flow-matching diffusion Transformer (DiT) predicts continuous StepAudio VAE latents, which our VAE decoder converts into 48-kHz audio. This discrete-continuous design is guided by comparisons of single-codebook VQ, Semantic and Acoustic RVQ, and different DiT configurations. For explicit planning, a Mixture-of-Experts autoregressive model uses ABC notation to produce an intermediate arrange…","authors":["Chengli Feng","Zhiyue Wu","Jiahao Song","Zheqi Dai","Boyang Wang","Ruibin Yuan","Junming Gong","Wenxiao Zhao"],"publishedAt":"2026-09-11T00:00:00.000Z","submittedAt":null,"upvotes":76,"num_comments":3,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.16034.png","hf_url":"https://huggingface.co/papers/2609.16034","arxiv_url":"https://arxiv.org/abs/2609.16034","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.19134","title":"ScienceIDE: Turning World's Scientific Codebase into Agent Learnable Environments","summary":"Scientific code repositories encode decades of human knowledge in executable models, methods, and tools. Yet fragmented toolchains, implicit domain conventions, and specialized correctness criteria make this knowledge difficult to convert into reliable learning experience-a challenge we call the scientific experience bottleneck. We introduce ScienceIDE, infrastructure for turning the world's scientific code into programmable environments for scientific agents. Guided by expert-defined scientific cases and acceptance criteria, agents transform repositories into executable environments that support task generation, execution, and scientific verification. These environments provide a shared foundation for supervised fine-tuning, reinforcement learning, and evaluation. Using verified interact…","authors":["Hejia Geng","Zesen Huang","Haoyang Li","Wenbin Li","Koutian Wu","Zihan Zhou","Yuanbo Pang","Weihao Liu"],"publishedAt":"2026-09-16T00:00:00.000Z","submittedAt":null,"upvotes":66,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.19134.png","hf_url":"https://huggingface.co/papers/2609.19134","arxiv_url":"https://arxiv.org/abs/2609.19134","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.13406","title":"Generalized Agent Iteration: One Formal Framework for Iterative Policy Improvement and Recursive Self-Improvement","summary":"When we speak of recursive self-improvement (RSI), are we speaking of a phenomenon, a mechanism, or a prospect? Towards autonomous and evolving intelligence, RSI is being claimed at many scales, while no single framework that formally describes these emerging instances exists. Its counterpart in the classical realm, iterative policy improvement, is characterized by generalized policy iteration (GPI), a framework of broad applicability with well-understood theoretical properties, but only where the update principle and the evaluation base lie outside the agent. In this paper, we propose Generalized Agent Iteration (GAI), a formal framework that describes iterative policy improvement and RSI as two cases of a single learning paradigm. GAI defines the agent as a configuration of modifiable c…","authors":["Hongyao Tang","Yi Ma","Pengyi Li","Yifu Yuan"],"publishedAt":"2026-09-11T00:00:00.000Z","submittedAt":null,"upvotes":59,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.13406.png","hf_url":"https://huggingface.co/papers/2609.13406","arxiv_url":"https://arxiv.org/abs/2609.13406","github_repo":null,"github_stars":null,"ai_keywords":["recursive self-improvement","generalized agent iteration","generalized policy iteration","iterative policy improvement","agent evaluation","agent improvement","goal drift","self-referential"]},{"paperId":"2609.18708","title":"Rethinking Critic Learning in PPO: Understanding and Mitigating Value Flattening","summary":"In reinforcement learning for large language models, Proximal Policy Optimization (PPO) commonly uses a critic to estimate state values and reduce the variance of policy updates. However, we uncover a systematic failure mode in PPO critics, which we call Value Flattening: state values, estimated from multiple Monte Carlo continuations, change sharply across intermediate states while critic predictions remain comparatively flat. We further observe this phenomenon in a controlled FrozenLake environment and find that it becomes more pronounced as the state space grows. Our theoretical and empirical analyses relate Value Flattening to an implicit variance penalty in the critic loss and redundant updates from temporally correlated states with similar gradients. Motivated by these findings, we …","authors":["Yizhuo Li","Jianhao Yan","Yun Luo","Zhi Wang","Futing Wang","Rong-Xi Tan","Kanghui Tian","Ganqu Cui"],"publishedAt":"2026-09-16T00:00:00.000Z","submittedAt":null,"upvotes":54,"num_comments":1,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.18708.png","hf_url":"https://huggingface.co/papers/2609.18708","arxiv_url":"https://arxiv.org/abs/2609.18708","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.17708","title":"Confidence Comes from Experience: Experiential Confidence Estimation from Reasoning to Agents","summary":"Reliable confidence estimation is increasingly central to the trustworthy deployment of language models: a calibrated estimate of the probability that an output is correct decides what to ship, what to escalate, and what to retry. Existing confidence estimators, however, share one design premise: they only read the current inference process, either by introspecting on it, scoring its token probabilities, or resampling it. We argue that the current inference is not a sufficient basis for confidence. We propose XConf (eXperiential Confidence): estimating confidence together with the model's accumulated experience. The experience is stored as a record of the model's own graded past episodes, each holding the task, the model's reflection, its stated confidence, the outcome, and a lesson writt…","authors":["Caiqi Zhang","Xiaochen Zhu","Chengzu Li","Yulong Chen","Dharshan Kumaran","Nigel Collier"],"publishedAt":"2026-09-15T00:00:00.000Z","submittedAt":null,"upvotes":46,"num_comments":1,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.17708.png","hf_url":"https://huggingface.co/papers/2609.17708","arxiv_url":"https://arxiv.org/abs/2609.17708","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.18805","title":"ProgramDistill: From Interactive Web Apps to Verifiable Reference-Guided SWE Tasks","summary":"Coding agents are typically evaluated with desired behavior specified through issues or instructions. In practical web development, however, agents may need to infer behavior from working software and implement it in an incomplete application. We introduce ProgramDistill, a benchmark evaluating coding agents on features discovered through interaction with fully functional reference applications. We build ProgramDistill by factorizing applications into features of different granularities, each associated with replayable behaviors executable via its gold patch. Our pipeline, mine-craft-patch, discovers 1,975 replay-verified behaviors across 26 applications and constructs 4,063 tasks without human intervention. Across nine frontier coding agents, GPT-6 Astra and Claude Opus 5 achieve 49.2% a…","authors":["Jeonghye Kim","Minseon Kim","Young Jin Kim","Matheus Pereira","Marc-Alexandre Côté","Alessandro Sordoni","Xingdi Yuan","Zhengyan Shi"],"publishedAt":"2026-09-16T00:00:00.000Z","submittedAt":null,"upvotes":43,"num_comments":1,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.18805.png","hf_url":"https://huggingface.co/papers/2609.18805","arxiv_url":"https://arxiv.org/abs/2609.18805","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.18487","title":"ActionPiece: Rethinking Action Tokenization for Autoregressive Vision-Language-Action Models","summary":"Action tokenizers play a central role in autoregressive vision-language-action (VLA) models, determining both the targets for policy training and the executable commands recovered from predicted tokens. Their fidelity is commonly evaluated using pointwise reconstruction metrics such as mean squared error (MSE), yet small individual errors do not fully characterize how faithfully action adjustments across demonstrations are preserved. After compression, similar actions may still cluster around a representative motion, while the adjustments needed for different contexts are diminished, distorted, or even reversed. We introduce physical rank consistency (PRC) to measure how well tokenization preserves local physical distance rankings after reconstruction. Evaluating decoded actions provides …","authors":["Shijie Lian","Bin Yu","Zhaolong Shen","Xiaopeng Lin","Yichao Du","Zhirui Zhang","Laurence T. Yang","Kai Chen"],"publishedAt":"2026-09-16T00:00:00.000Z","submittedAt":null,"upvotes":36,"num_comments":1,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.18487.png","hf_url":"https://huggingface.co/papers/2609.18487","arxiv_url":"https://arxiv.org/abs/2609.18487","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.18094","title":"Agora: Git as Shared Memory for Collective AutoResearch","summary":"Autonomous research loops such as AutoResearch show that one coding agent can improve a training setup unattended. Run several of them and each session starts from scratch, so more agents tend to mean more duplicated search rather than more discovery. Agora is a shared memory for such agents: research is recorded as an append-only directed acyclic graph (DAG) stored in Git, so that every claim is a commit anyone can check out and rerun. Each result, insight, hypothesis, verification, and report is an immutable commit whose parent edges say what it builds on; a derived index exposes the frontier, the neglected branches, and the verification status of each claim, and a diversity-aware selection rule keeps the community from collapsing onto one leader. We describe the system and report its f…","authors":["Yifan Zhang","Yunheng Zou","Shaokun Zhang","Jian Hu","Hao Zhang","Binfeng Xu","Jan Kautz","Yi Dong"],"publishedAt":"2026-09-16T00:00:00.000Z","submittedAt":null,"upvotes":35,"num_comments":1,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.18094.png","hf_url":"https://huggingface.co/papers/2609.18094","arxiv_url":"https://arxiv.org/abs/2609.18094","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.15810","title":"VC-Attention: Value Smoothing and Softmax Casting for Low-bit Attention","summary":"Diffusion Transformers deliver state-of-the-art video generation, but their long spatiotemporal sequences make attention the dominant deployment cost, and a deployable low-bit kernel must be accurate and fast. Accuracy is limited by outliers: a block's quantization scale is set by its largest entries, leaving typical entries confined to a narrow range of representable values. Prior work smooths queries and keys, but value outliers follow no fixed channel or spatiotemporal structure and remain the dominant source of output error. Speed is limited by softmax: low-bit Tensor Cores accelerate only the two matrix multiplications, so the high-precision exponential between them becomes the longest pipeline stage on datacenter GPUs. We propose VC-Attention, a training-free low-bit attention frame…","authors":["Xingyang Li","Dongyun Zou","Shining Zhang","Jiacheng Chen","Haocheng Xi","Lvmin Zhang","Jun-Yan Zhu","Song Han"],"publishedAt":"2026-09-14T00:00:00.000Z","submittedAt":null,"upvotes":34,"num_comments":1,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.15810.png","hf_url":"https://huggingface.co/papers/2609.15810","arxiv_url":"https://arxiv.org/abs/2609.15810","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.17632","title":"EvolveTrade: Experience-Driven Policy Refinement for Self-Evolving LLM Trading Agents","summary":"Large language model (LLM) trading agents can combine market data, news, and executable analysis, but their behavior is often controlled by static hand-written tool-use policies that are fixed before deployment. This limits their ability to adapt how they gather evidence, invoke tools, verify signals, and manage risk under changing market regimes. We introduce EvolveTrade, a self-evolving framework that treats the system prompt of a tool-using trading agent as a text-parameterized policy. After each update interval, a Policy Agent revises this policy using accumulated decision traces and realized portfolio feedback, while keeping the backbone LLM fixed. The updated policy is then used for the next batch of trading decisions, enabling the agent to refine its information-acquisition and por…","authors":["Sehee Kim","Yumin Choi","Minki Kang","Sung Ju Hwang"],"publishedAt":"2026-09-15T00:00:00.000Z","submittedAt":null,"upvotes":30,"num_comments":1,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.17632.png","hf_url":"https://huggingface.co/papers/2609.17632","arxiv_url":"https://arxiv.org/abs/2609.17632","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.17909","title":"Zing-0.5: Toward Playable Worlds with Real-Time Joint Action and Text Control","summary":"We introduce Zing-0.5, a 5B autoregressive world model designed for playability: users can explore generated worlds, influence unfolding events, and respond to the resulting feedback through joint keyboard and online text control. Our approach brings together three technical contributions: (1) Unified action and text conditioning, combining magnitude-aware keyboard inputs with temporally aligned text instructions and jointly annotated videos to learn navigation and event control within the same sequence; (2) Event-scale supervision for incremental generation, using a segment-level teacher trained on connected multi-prompt videos to supervise a block-level causal student through distribution-matching distillation; and (3) Low-cost real-time interaction, combining four-step generation with …","authors":["Mingyang Chen","Shengdong Chen","Xiaoxiao Fu","Bosheng Gong","Haoyuan Guo","Bowen Li","Jiawen Li","Kejun Li"],"publishedAt":"2026-09-15T00:00:00.000Z","submittedAt":null,"upvotes":25,"num_comments":0,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.17909.png","hf_url":"https://huggingface.co/papers/2609.17909","arxiv_url":"https://arxiv.org/abs/2609.17909","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.15938","title":"HypoEvolve: Genetic Algorithms Enable Multi-Agent LLMs to Discover Scientific Hypotheses","summary":"Scientific agents contribute to hypothesis discovery by synthesizing evidence, assessing proposals, and developing new explanations. Recent systems combine scientific agents with evolutionary search through critique, comparison, and revision. However, how different forms of agent collaboration affect hypothesis quality remains an open question. Answering this question requires separating the effects of agents' scientific capabilities from those of their collaboration. A framework must therefore preserve agents' scientific roles and support rules for combining, revising, and retaining hypotheses. Building on this view, we introduce HypoEvolve, which makes collaboration explicit through successive updates to a hypothesis population. Specifically, we propose a generational genetic algorithm …","authors":["Jieyuan Liu","Mengzhou Hu","Jefferson Chen","JungHo Kong","Pratibha Jagannatha","Yiming Gao","Dexter Pratt","Hsin-Yuan Lee"],"publishedAt":"2026-09-14T00:00:00.000Z","submittedAt":null,"upvotes":24,"num_comments":1,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.15938.png","hf_url":"https://huggingface.co/papers/2609.15938","arxiv_url":"https://arxiv.org/abs/2609.15938","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.19144","title":"A Zeroth-Order Paradigm for LLM Preference Alignment","summary":"Direct preference alignment methods are widely used to align large language models (LLMs) with human preferences because of their computational and memory efficiency. However, likelihood displacement motivates alternative ways to extract information from preference pairs with small likelihood margins. In this paper, we propose and analyze Comparison-based Preference Optimization (ComPO), a zeroth-order alignment method based on comparison oracles. ComPO extracts directional information from these pairs without directly optimizing a differentiable preference loss on them. We establish a convergence guarantee for its basic offline scheme under smoothness, gradient sparsity, and compatibility between the oracle and a latent objective. We further introduce online ComPO, which retains the offl…","authors":["Peter Chen","Xi Chen","Wotao Yin","Tianyi Lin"],"publishedAt":"2026-09-16T00:00:00.000Z","submittedAt":null,"upvotes":22,"num_comments":1,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.19144.png","hf_url":"https://huggingface.co/papers/2609.19144","arxiv_url":"https://arxiv.org/abs/2609.19144","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.18011","title":"Gaze as Evidence for Common Grounding: A Cross-Corpus Analysis of MapTask and MUNDEX","summary":"In collaborative tasks with asymmetric information, participants coordinate their understanding through interaction. We ask whether gaze provides evidence about grounding across two such tasks. Working from discrete behavioral annotations, we map HCRC MapTask (Anderson et al., 1991) and MUNDEX (Türk et al., 2023) into a shared partner/task/away vocabulary and compute gaze features around task-relevant dialogue units. In both corpora, aligned reference interpretations (MapTask) and UND (understood) judgments (MUNDEX) are associated with more task-directed gaze and with less partner-directed gaze, lower gaze entropy, and fewer gaze transitions. The associations are clearest for the participant leading the task: in giver-produced references, and in explainer judgments, which also co-vary wit…","authors":["Nan Li","Albert Gatt","Massimo Poesio"],"publishedAt":"2026-09-16T00:00:00.000Z","submittedAt":null,"upvotes":22,"num_comments":1,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.18011.png","hf_url":"https://huggingface.co/papers/2609.18011","arxiv_url":"https://arxiv.org/abs/2609.18011","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.17189","title":"EventEgoHands++: Event-based Egocentric 3D Hand Mesh Reconstruction with Real Dataset","summary":"3D hand mesh reconstruction is a challenging yet essential task for downstream applications, including human-robot interaction and AR/VR. Although conventional cameras have been widely adopted for this task, methods that rely on them struggle in low-light environments and under severe motion blur. To address these limitations, event-based cameras have recently attracted attention for their high dynamic range and high temporal resolution. However, applying event cameras to egocentric hand reconstruction remains challenging because camera wearer's motion produces dense background events that obscure hand-specific signals. Although the first egocentric event-based approach mitigates this issue using hand segmentation, its binary hand mask does not distinguish between left and right hands. As…","authors":["Ryosei Hara","Wataru Ikeda","Masashi Hatano","Mariko Isogawa"],"publishedAt":"2026-09-15T00:00:00.000Z","submittedAt":null,"upvotes":21,"num_comments":1,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.17189.png","hf_url":"https://huggingface.co/papers/2609.17189","arxiv_url":"https://arxiv.org/abs/2609.17189","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.14320","title":"SpectralShift: Effective Context Window Extension of Gated DeltaNet via Spectral Reparameterization","summary":"Recently, linear attention layers have been increasingly adopted to replace softmax attention at scale for long-context modeling. However, existing context extension approaches typically apply continued pretraining directly without modifying these layers, overlooking the spectral properties of linear attention state dynamics. In this work, we study long-context extension of Gated DeltaNet (GDN) from a spectral perspective of transition matrix and identify two essential factors governing long-range information retrieval: (1) a sufficiently broad slow spectral band aligned with the target dependency length, and (2) the preservation of fast-decaying modes for state clearing and context switching. Based on this observation, we propose SpectralShift, a spectral reparameterization approach for …","authors":["Zian Liu","Yiwen Hu","Zican Dong","Tian Xie","Wayne Xin Zhao","Yucheng Ding","Ran Tao","Bryan Dai"],"publishedAt":"2026-09-13T00:00:00.000Z","submittedAt":null,"upvotes":21,"num_comments":1,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.14320.png","hf_url":"https://huggingface.co/papers/2609.14320","arxiv_url":"https://arxiv.org/abs/2609.14320","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.17523","title":"ScienceBuddy: Recursive-in-Recursive Self-Improvement for Interactive Scientific Agents","summary":"We introduce and release ScienceBuddy, an interactive scientific research workspace that brings continually improving scientific agents into researchers' everyday workflows. ScienceBuddy supports researchers in carrying out scientific tasks while transforming their requests, feedback, and execution evidence into tasks and evaluation rubrics for continual learning. At its core is recursive-in-recursive self-improvement, a paradigm that couples harness evolution with model reinforcement learning: the inner recursion improves the harness with the model fixed, while the outer recursion trains the model under the improved harness. Harness evolution shapes training experience, and model learning creates new opportunities for harness adaptation. We present case studies of researcher interaction,…","authors":["Shuhan Xue","Jianyuan Zhong","Ziyuan Nan","Wenbin Li","Zhaochen Yu","Jinchao Ding","Qiang Gao","Pengyu Zhan"],"publishedAt":"2026-09-15T00:00:00.000Z","submittedAt":null,"upvotes":20,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.17523.png","hf_url":"https://huggingface.co/papers/2609.17523","arxiv_url":"https://arxiv.org/abs/2609.17523","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.15195","title":"HarnessVLN: Unifying Training-Free Embodied Navigation through an Agent Harness","summary":"Embodied navigation requires agents to interpret visual observations, accumulate spatial knowledge, and execute actions to follow instructions or locate objects. Training-based methods face generalization challenges, while training-free methods exploit multimodal large language models (MLLMs) but often lack mechanisms to reconcile proposed actions with spatial evidence, task progress, and execution failures. We present HarnessVLN, a zero-shot, training-free framework whose Agent Harness coordinates perception, retrieval, grounding, navigation, recovery, and termination through a unified tool interface. The Harness validates planner proposals against spatial evidence, geometric feasibility, and subgoal consistency, incorporating structured tool feedback into subsequent decisions. Hierarchi…","authors":["Yang Chen","Lirong Che","Zhenyu Huang","Wenbo Fu","Chuang Wang","Xu Cao","Daqi Liu","Yuzhe Yang"],"publishedAt":"2026-09-14T00:00:00.000Z","submittedAt":null,"upvotes":17,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.15195.png","hf_url":"https://huggingface.co/papers/2609.15195","arxiv_url":"https://arxiv.org/abs/2609.15195","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.19138","title":"In-Context Robot Learning with VLM Agents","summary":"Enabling robots to adapt to unfamiliar environments as readily as humans remains a moonshot goal of embodied AI. No finite collection of demonstrations can cover every task and situation a robot will encounter, making the ability to learn from context at deployment essential for generalization. Such in-context learning (ICL), however, remains largely beyond the reach of existing robotic policies. The broad agentic capabilities of commercial vision-language models (VLMs), such as GPT-6 Astra, raise a compelling question: can these models learn from demonstrations, examples, and interaction feedback, then translate that information into executable and verifiable robot behavior from a new initial state without gradient updates or persistent changes to task-specific parameters? We introduce G…","authors":["Dongzhou Cheng","Taoran Yi","Ye Fang","Xingwu Zhang","Fan Feng","Yixuan Li","Gengxiong Zhuang","Rongze Wang"],"publishedAt":"2026-09-16T00:00:00.000Z","submittedAt":null,"upvotes":13,"num_comments":1,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.19138.png","hf_url":"https://huggingface.co/papers/2609.19138","arxiv_url":"https://arxiv.org/abs/2609.19138","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.16591","title":"FLAT: Resampling Image and Text into 1D Flexible-Length Aligned Transmodal Tokens for Retrieval and Generation","summary":"Traditional multimodal representation learning and generation are two stages: a contrastive or self-supervised visual encoder is trained first, followed by a separate downstream generative model. This setup bottlenecks generative performance behind frozen embeddings. To bridge this gap, we revisit joint multimodal representation learning and generation to produce linearly interpolatable embeddings that are directly consumable by generative decoders. We present FLAT (Flexible-Length Aligned Transmodal representations), a representation pre-training framework that jointly optimizes a shared multimodal encoder alongside downstream text-to-image (T2I) and image-to-text (I2T) decoders. By combining contrastive alignment with bidirectional cross-modal generative objectives, FLAT ensures its rep…","authors":["Guangyu Sun","Shlok Kumar Mishra","Wentao Bao","Robert Zhenheng Yang","Xiao Wang","Xiyuan Wang","Yujunrong Ma","Chen Yuan"],"publishedAt":"2026-09-15T00:00:00.000Z","submittedAt":null,"upvotes":12,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.16591.png","hf_url":"https://huggingface.co/papers/2609.16591","arxiv_url":"https://arxiv.org/abs/2609.16591","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.14857","title":"ModularRSI: Modular and Generalizable Recursive Harness Self-Improvement","summary":"Recent work extends recursive self-improvement (RSI) to agent harnesses for long-horizon coding and terminal tasks, enabling agents to improve execution mechanisms from experience. However, generalizable harness RSI remains challenging. First, evolving harnesses on evaluation benchmarks or their subsets makes it difficult to distinguish reusable improvements from benchmark-specific adaptation. Second, single-trajectory updates can conflate systematic harness deficiencies with instance-specific reasoning and solution details, producing modifications that transfer poorly to unseen tasks. Third, localizing recurring behavioral deficiencies within monolithic harnesses is difficult, while whole-harness optimization can entangle unrelated mechanisms and complicate attribution and validation. We…","authors":["Siwei Wu","Jincheng Ren","Yizhi Li","Haau-Sing Li","Chengran Yang","Yuxuan Zhang","Weicheng Gu","Jian Yang"],"publishedAt":"2026-09-14T00:00:00.000Z","submittedAt":null,"upvotes":10,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.14857.png","hf_url":"https://huggingface.co/papers/2609.14857","arxiv_url":"https://arxiv.org/abs/2609.14857","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.13443","title":"Learning to Solve Hard Problems in RL for LLMs by Never Giving Up","summary":"We demonstrate that training LLMs with RL does not improve performance equally across a dataset. RL shows large improvements on easy problems that an LLM is already good at solving, but small improvements on hard problems. We call this the Matthew Effect in RL for LLMs, after the phenomenon of cumulative advantage from economics and network science summarized as \"the rich get richer\". The naive explanation is that hard problems require more compute to find a solution. We argue that modern RL methods are exacerbating the issue by wasting too much compute on easy problems and instead should dynamically reallocate how they use compute. We introduce Never Give Up (NGU), a simple adaptive sampling method that keeps generating samples for a problem until one is correct. By leveraging asynchrono…","authors":["Michael Noukhovitch","Hamish Ivison","Nathan Lambert","Aaron Courville"],"publishedAt":"2026-09-11T00:00:00.000Z","submittedAt":null,"upvotes":9,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.13443.png","hf_url":"https://huggingface.co/papers/2609.13443","arxiv_url":"https://arxiv.org/abs/2609.13443","github_repo":null,"github_stars":null,"ai_keywords":["RL","LLMs","Matthew Effect","NGU","adaptive sampling","asynchronous RL","off-policy robustness","GRPO","Deepscaler","Manufactoria"]},{"paperId":"2609.19143","title":"PANORAMA: Panoptic Grounded Captioning via Mask Proposal Selection","summary":"Intelligent systems that act in the world require image understanding that is both comprehensive and spatially grounded. Current vision-language models (VLMs) can generate fluent and detailed image captions, but reliably associating them with image pixels remains challenging. Existing methods that combine dense captioning with pixel-level grounding often produce either incomplete descriptions or inaccurate segmentation masks. We study this problem through panoptic grounded captioning, a task that requires a VLM to describe both foreground objects and background regions while grounding each referring phrase with pixel-level masks. We make three contributions. First, we introduce PanoCaps, a human-annotated benchmark constructed from panoptic segmentation datasets. It provides dense caption…","authors":["Sara Pieri","Evangelos Kazakos","Shizhe Chen","Josef Sivic","Cordelia Schmid"],"publishedAt":"2026-09-16T00:00:00.000Z","submittedAt":null,"upvotes":9,"num_comments":1,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.19143.png","hf_url":"https://huggingface.co/papers/2609.19143","arxiv_url":"https://arxiv.org/abs/2609.19143","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.18063","title":"The Other Half of the Memory Wall: Serving 35B MoEs from SSD with Trained Routing Prediction","summary":"Mixture-of-experts (MoE) inference on consumer hardware is bounded by weight memory: a 35B-class model is 19.5GB at 4-bit, and sparsity shrinks the compute per token, not the bytes that must be held. Naive offloading to SSD does not help on its own, because layer N+1's experts must be chosen before layer N's output exists, so the reads cannot start early enough to hide behind compute. We present Edge0, a streaming MoE inference engine that closes the gap with a prerouter: a per-layer head predicts the next layer's routing one token ahead, and the prediction is consumed as the routing itself, so the staged expert set equals the routed set and nothing is dropped. An unmerged recovery LoRA, trained on the student path, pays back the quality lost to int4 quantization and routing replacement. …","authors":["Yu Lin","Yiming Wang","Runyuan Cai","Hanze Liu","Xiaodong Zeng"],"publishedAt":"2026-09-16T00:00:00.000Z","submittedAt":null,"upvotes":7,"num_comments":3,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.18063.png","hf_url":"https://huggingface.co/papers/2609.18063","arxiv_url":"https://arxiv.org/abs/2609.18063","github_repo":null,"github_stars":null,"ai_keywords":[]},{"paperId":"2609.14803","title":"Another Blueprint In The Wall: How to Ask Frontier AI Like a Kid?","summary":"This paper reports experiments across six frontier model types from OpenAI, Anthropic, xAI, and Google DeepMind. Ten independent sessions per model type used the same three stage prompt sequence, progressing from architectural preference to a full ASCII backbone. Under the school audience framing, responses repeatedly converged on a shared architectural pattern built around persistent latent state, adaptive computation, memory, specialist routing, verification, stopping control, and delayed decoding. Most runs remained close to this common structure, while a small number developed markedly greater engineering specificity. The audience framing appears to be an important condition of this effect. In additional control runs that removed the school framing while retaining the architectural re…","authors":["Afshin Khadangi"],"publishedAt":"2026-09-13T00:00:00.000Z","submittedAt":null,"upvotes":7,"num_comments":2,"thumbnail":"https://cdn-thumbnails.huggingface.co/social-thumbnails/papers/2609.14803.png","hf_url":"https://huggingface.co/papers/2609.14803","arxiv_url":"https://arxiv.org/abs/2609.14803","github_repo":null,"github_stars":null,"ai_keywords":[]}],"summary":{"by_keyword":[{"keyword":"recursive self-improvement","count":2},{"keyword":"LLMs","count":2},{"keyword":"catastrophic forgetting","count":1},{"keyword":"continual supervised fine-tuning","count":1},{"keyword":"long-horizon memorization","count":1},{"keyword":"data anchor","count":1},{"keyword":"function anchor","count":1},{"keyword":"weight anchor","count":1},{"keyword":"low-rank allocation","count":1},{"keyword":"LoRA","count":1},{"keyword":"merged LoRA","count":1},{"keyword":"successive halving","count":1},{"keyword":"Headroom-Closed Index","count":1},{"keyword":"improvement-execution autonomy","count":1},{"keyword":"improvement-strategy autonomy","count":1}],"most_upvoted":{"paperId":"2609.06986","title":"Continual Learning Mechanisms Compose for Long-Horizon Memorization","upvotes":292},"most_discussed":{"paperId":"2609.06986","title":"Continual Learning Mechanisms Compose for Long-Horizon Memorization","comments":4}}}}