ABSTRACT Automatic Speech Recognition (ASR) systems remain prone to errors that affect downstream applications. In this paper, we propose LIR-ASR, a heuristic optimized iterative correction framework using LLMs, inspired by human auditory perception. LIR-ASR applies a "Listening-Imagining-Refining" strategy, generating phonetic variants and refining them in context. A heuristic optimization with finite state machine (FSM) is introduced to prevent the correction process from being trapped in local optima and rule-based constraints help maintain semantic fidelity. Experiments on both English and Chinese ASR outputs show that LIR-ASR achieves average reductions in CER/WER of up to 1.5 percentage points compared to baselines, demonstrating substantial accuracy gains in transcription.

Scaling up the deployment of autonomous excavators is of great economic and societal importance. Yet it remains a challenging problem, as effective systems must robustly handle unseen worksite conditions and new hardware configurations. Current state-of-the-art approaches rely on highly engineered, task-specific controllers, which require extensive manual tuning for each new scenario. In contrast, recent advances in large-scale pretrained models have shown remarkable adaptability across tasks and embodiments in domains such as manipulation and navigation, but their applicability to heavy construction machinery remains largely unexplored. In this work, we introduce ExT, a unified open-source framework for large-scale demonstration collection, pretraining, and fine-tuning of multitask excavation policies. ExT policies are first trained on large-scale demonstrations collected from a mix of experts, then fine-tuned either with supervised fine-tuning (SFT) or reinforcement learning fine-tuning (RLFT) to specialize to new tasks or operating conditions. Through both simulation and real-world experiments, we show that pretrained ExT policies can execute complete excavation cycles with centimeter-level accuracy, successfully transferring from simulation to real machine with performance comparable to specialized single-task controllers. Furthermore, in simulation, we demonstrate that ExT's fine-tuning pipelines allow rapid adaptation to new tasks, out-of-distribution conditions, and machine configurations, while maintaining strong performance on previously learned tasks. These results highlight the potential of ExT to serve as a foundation for scalable and generalizable autonomous excavation.

The debate around Artificial General Intelligence (AGI) remains open due to two fundamentally different goals: replicating human-like performance versus replicating human-like cognitive processes. We argue that current performance-based definitions are inadequate because they provide no clear, mechanism-focused roadmap for research, and they fail to properly define the qualitative nature of genuine intelligence. Drawing inspiration from the human brain, we propose a new paradigm that shifts the focus from external mimicry to the development of foundational cognitive architectures. We define True Intelligence (TI) as a system characterized by six core components: embodied sensory fusion, core directives, dynamic schemata creation, a highly-interconnected multi-expert architecture, an orchestration layer, and lastly, the unmeasurable quality of Interconnectedness, which we hypothesize results in consciousness and a subjective experience. We propose a practical, five-level taxonomy of AGI based on the number of the first five measurable components a system exhibits. This framework provides a clear path forward with developmental milestones that directly address the challenge of building genuinely intelligent systems. We contend that once a system achieves Level-5 AGI by implementing all five measurable components, the difference between it and TI remains as a purely philosophical debate. For practical purposes - and given theories indicate consciousness is an emergent byproduct of integrated, higher-order cognition - we conclude that a fifth-level AGI is functionally and practically equivalent to TI. This work synthesizes diverse insights from analytical psychology, schema theory, metacognition, modern brain architectures and latest works in AI to provide the first holistic, mechanism-based definition of AGI that offers a clear and actionable path for the research community.

Large language models (LLMs) have achieved great success in medical question answering and clinical decision-making, promoting the efficiency and popularization of the personalized virtual doctor in society. However, the traditional fine-tuning strategies on LLM require the updates of billions of parameters, substantially increasing the training cost, including the training time and utility cost. To enhance the efficiency and effectiveness of the current medical LLMs and explore the boundary of the representation capability of the LLMs on the medical domain, apart from the traditional fine-tuning strategies from the data perspective (i.e., supervised fine-tuning or reinforcement learning from human feedback), we instead craft a novel sparse medical LLM named SparseDoctor armed with contrastive learning enhanced LoRA-MoE (low rank adaptation-mixture of experts) architecture. To this end, the crafted automatic routing mechanism can scientifically allocate the computational resources among different LoRA experts supervised by the contrastive learning. Additionally, we also introduce a novel expert memory queue mechanism to further boost the efficiency of the overall framework and prevent the memory overflow during training. We conduct comprehensive evaluations on three typical medical benchmarks: CMB, CMExam, and CMMLU-Med. Experimental results demonstrate that the proposed LLM can consistently outperform the strong baselines such as the HuatuoGPT series.

We show that language models' activations linearly encode when information was learned during training. Our setup involves creating a model with a known training order by sequentially fine-tuning Llama-3.2-1B on six disjoint but otherwise similar datasets about named entities. We find that the average activations of test samples corresponding to the six training datasets encode the training order: when projected into a 2D subspace, these centroids are arranged exactly in the order of training and lie on a straight line. Further, we show that linear probes can accurately (~90%) distinguish "early" vs. "late" entities, generalizing to entities unseen during the probes' own training. The model can also be fine-tuned to explicitly report an unseen entity's training stage (~80% accuracy). Interestingly, the training-order encoding does not seem attributable to simple differences in activation magnitudes, losses, or model confidence. Our paper demonstrates that models are capable of differentiating information by its acquisition time, and carries significant implications for how they might manage conflicting data and respond to knowledge modifications.

Recent advances in deep-research systems have demonstrated the potential for AI agents to autonomously discover and synthesize knowledge from external sources. In this paper, we introduce WebResearcher, a novel framework for building such agents through two key components: (1) WebResearcher, an iterative deep-research paradigm that reformulates deep research as a Markov Decision Process, where agents periodically consolidate findings into evolving reports while maintaining focused workspaces, overcoming the context suffocation and noise contamination that plague existing mono-contextual approaches; and (2) WebFrontier, a scalable data synthesis engine that generates high-quality training data through tool-augmented complexity escalation, enabling systematic creation of research tasks that bridge the gap between passive knowledge recall and active knowledge construction. Notably, we find that the training data from our paradigm significantly enhances tool-use capabilities even for traditional mono-contextual methods. Furthermore, our paradigm naturally scales through parallel thinking, enabling concurrent multi-agent exploration for more comprehensive conclusions. Extensive experiments across 6 challenging benchmarks demonstrate that WebResearcher achieves state-of-the-art performance, even surpassing frontier proprietary systems.

We present LightVLA, a simple yet effective differentiable token pruning framework for vision-language-action (VLA) models. While VLA models have shown impressive capability in executing real-world robotic tasks, their deployment on resource-constrained platforms is often bottlenecked by the heavy attention-based computation over large sets of visual tokens. LightVLA addresses this challenge through adaptive, performance-driven pruning of visual tokens: It generates dynamic queries to evaluate visual token importance, and adopts Gumbel softmax to enable differentiable token selection. Through fine-tuning, LightVLA learns to preserve the most informative visual tokens while pruning tokens which do not contribute to task execution, thereby improving efficiency and performance simultaneously. Notably, LightVLA requires no heuristic magic numbers and introduces no additional trainable parameters, making it compatible with modern inference frameworks. Experimental results demonstrate that LightVLA outperforms different VLA models and existing token pruning methods across diverse tasks on the LIBERO benchmark, achieving higher success rates with substantially reduced computational overhead. Specifically, LightVLA reduces FLOPs and latency by 59.1% and 38.2% respectively, with a 2.6% improvement in task success rate. Meanwhile, we also investigate the learnable query-based token pruning method LightVLA* with additional trainable parameters, which also achieves satisfactory performance. Our work reveals that as VLA pursues optimal performance, LightVLA spontaneously learns to prune tokens from a performance-driven perspective. To the best of our knowledge, LightVLA is the first work to apply adaptive visual token pruning to VLA tasks with the collateral goals of efficiency and performance, marking a significant step toward more efficient, powerful and practical real-time robotic systems.

Software engineering presents complex, multi-step challenges for Large Language Models (LLMs), requiring reasoning over large codebases and coordinated tool use. The difficulty of these tasks is exemplified by benchmarks like SWE-bench, where current LLMs still struggle to resolve real-world issues. A promising approach to enhance performance is test-time scaling (TTS), but its gains are heavily dependent on the diversity of model outputs. While standard alignment methods such as Direct Preference Optimization (DPO) and Kahneman-Tversky Optimization (KTO) are effective at aligning model outputs with human preferences, this process can come at the cost of reduced diversity, limiting the effectiveness of TTS. Additionally, existing preference optimization algorithms are typically designed for single-turn tasks and do not fully address the complexities of multi-turn reasoning and tool integration required for interactive coding agents. To bridge this gap, we introduce EntroPO, an entropy-enhanced framework that adapts existing preference optimization algorithms to the multi-turn, tool-assisted setting. EntroPO augments the preference objective to explicitly preserve policy entropy and generalizes learning to optimize over multi-turn interactions rather than single-turn responses. We validate EntroPO by fine-tuning a diverse suite of models from different families and sizes (up to 106B parameters). To maximize performance gains from TTS, we further propose a hybrid best-trajectory selection scheme combining a learned verifier model with model free approaches. On the \swebench leaderboard, our approach establishes new state-of-the-art results among open-weight models. A 30B parameter model trained with EntroPO ranks 1st on \lite and 4th on \verified on the open-weight leaderboard, surpassed only by models with over 10x more parameters(\eg$>$350B).

Puns are a form of humorous wordplay that exploits polysemy and phonetic similarity. While LLMs have shown promise in detecting puns, we show in this paper that their understanding often remains shallow, lacking the nuanced grasp typical of human interpretation. By systematically analyzing and reformulating existing pun benchmarks, we demonstrate how subtle changes in puns are sufficient to mislead LLMs. Our contributions include comprehensive and nuanced pun detection benchmarks, human evaluation across recent LLMs, and an analysis of the robustness challenges these models face in processing puns.

Speculative decoding is a powerful way to accelerate autoregressive large language models (LLMs), but directly porting it to vision-language models (VLMs) faces unique systems constraints: the prefill stage is dominated by visual tokens whose count scales with image resolution and video length, inflating both compute and memory--especially the key-value (KV) cache. We study speculative decoding for VLMs and introduce SpecVLM, a practical system that (1) establishes a strong EAGLE-2-style baseline, EagleVLM, delivering 1.5-2.3 To avoid costly offline distillation corpora, we propose an online-logit distillation protocol that trains the draft model with on-the-fly teacher logits and penultimate features using a combined cross-entropy and Smooth L1 objective, eliminating storage and preprocessing while remaining compute-efficient. This protocol reveals a training-time scaling effect: longer online training monotonically increases the draft model's average accepted length, improving speculative efficiency. Empirically, SpecVLM achieves additional acceleration, culminating in 2.5-2.9 end-to-end speedups within 5 epochs across LLaV A and MMMU, consistently over resolutions and task difficulties, while preserving the target model's output distribution (lossless decoding). Autoregressive decoding underpins many high-quality vision-language models (VLMs) such as LLaV A (Liu et al., 2023), GPT -4 (Achiam et al., 2023), and Gemini (Team et al., 2023), which are widely used for image captioning, visual question answering, and multimodal dialogue. While these teacher models produce high-fidelity outputs, their token-by-token decoding is computationally expensive--an issue that is amplified in multimodal settings because the prefill stage (visual encoding projection token injection) can dominate wall-clock time and memory usage (Li et al., 2025b). Higher image resolutions, denser visual tokenizations, and video inputs dramatically increase the number of visual tokens, which in turn inflates both the prefill cost and the KV cache traffic during decoding.