We study the process through which reasoning models trained with reinforcement learning on verifiable rewards (RLVR) can learn to solve new problems. We find that RLVR drives performance in two main ways: (1) by compressing pass@$k$ into pass@1 and (2) via "capability gain" in which models learn to solve new problems that they previously could not solve even at high $k$. We find that while capability gain exists across model scales, learning to solve new problems is primarily driven through self-distillation. We demonstrate these findings across model scales ranging from 0.5B to 72B parameters on >500,000 reasoning problems with prompts and verifiable final answers across math, science, and code domains. We further show that we can significantly improve pass@$k$ rates by leveraging natural language guidance for the model to consider within context while still requiring the model to derive a solution chain from scratch. Based of these insights, we derive $\text{Guide}$ -- a new class of online training algorithms. $\text{Guide}$ adaptively incorporates hints into the model's context on problems for which all rollouts were initially incorrect and adjusts the importance sampling ratio for the "off-policy" trajectories in order to optimize the policy for contexts in which the hints are no longer present. We describe variants of $\text{Guide}$ for GRPO and PPO and empirically show that Guide-GRPO on 7B and 32B parameter models improves generalization over its vanilla counterpart with up to 4$\%$ macro-average improvement across math benchmarks. We include careful ablations to analyze $\text{Guide}$'s components and theoretically analyze Guide's learning efficiency.

This study bridges cognitive science and neural network design by examining whether artificial models exhibit human-like forgetting curves. Drawing upon Ebbinghaus' seminal work on memory decay and principles of spaced repetition, we propose a quantitative framework to measure information retention in neural networks. Our approach computes the recall probability by evaluating the similarity between a network's current hidden state and previously stored prototype representations. This retention metric facilitates the scheduling of review sessions, thereby mitigating catastrophic forgetting during deployment and enhancing training efficiency by prompting targeted reviews. Our experiments with Multi-Layer Perceptrons reveal human-like forgetting curves, with knowledge becoming increasingly robust through scheduled reviews. This alignment between neural network forgetting curves and established human memory models identifies neural networks as an architecture that naturally emulates human memory decay and can inform state-of-the-art continual learning algorithms.

In the age of generative AI and ubiquitous digital tools, human cognition faces a structural paradox: as external aids become more capable, internal memory systems risk atrophy. Drawing on neuroscience and cognitive psychology, this paper examines how heavy reliance on AI systems and discovery-based pedagogies may impair the consolidation of declarative and procedural memory -- systems essential for expertise, critical thinking, and long-term retention. We review how tools like ChatGPT and calculators can short-circuit the retrieval, error correction, and schema-building processes necessary for robust neural encoding. Notably, we highlight striking parallels between deep learning phenomena such as "grokking" and the neuroscience of overlearning and intuition. Empirical studies are discussed showing how premature reliance on AI during learning inhibits proceduralization and intuitive mastery. We argue that effective human-AI interaction depends on strong internal models -- biological "schemata" and neural manifolds -- that enable users to evaluate, refine, and guide AI output. The paper concludes with policy implications for education and workforce training in the age of large language models.

Graph Retrieval Augmented Generation (GraphRAG) has garnered increasing recognition for its potential to enhance large language models (LLMs) by structurally organizing domain-specific corpora and facilitating complex reasoning. However, current evaluations of GraphRAG models predominantly rely on traditional question-answering datasets. Their limited scope in questions and evaluation metrics fails to comprehensively assess the reasoning capacity improvements enabled by GraphRAG models. To address this gap, we introduce GraphRAG-Bench, a large-scale, domain-specific benchmark designed to rigorously evaluate GraphRAG models. Our benchmark offers three key superiorities: \((i)\) Challenging question design. Featuring college-level, domain-specific questions that demand multi-hop reasoning, the benchmark ensures that simple content retrieval is insufficient for problem-solving. For example, some questions require mathematical reasoning or programming. \((ii)\) Diverse task coverage. The dataset includes a broad spectrum of reasoning tasks, multiple-choice, true/false, multi-select, open-ended, and fill-in-the-blank. It spans 16 disciplines in twenty core textbooks. \((iii)\) Holistic evaluation framework. GraphRAG-Bench provides comprehensive assessment across the entire GraphRAG pipeline, including graph construction, knowledge retrieval, and answer generation. Beyond final-answer correctness, it evaluates the logical coherence of the reasoning process. By applying nine contemporary GraphRAG methods to GraphRAG-Bench, we demonstrate its utility in quantifying how graph-based structuring improves model reasoning capabilities. Our analysis reveals critical insights about graph architectures, retrieval efficacy, and reasoning capabilities, offering actionable guidance for the research community.

Humans naturally understand moments in a video by integrating visual and auditory cues. For example, localizing a scene in the video like "A scientist passionately speaks on wildlife conservation as dramatic orchestral music plays, with the audience nodding and applauding" requires simultaneous processing of visual, audio, and speech signals. However, existing models often struggle to effectively fuse and interpret audio information, limiting their capacity for comprehensive video temporal understanding. To address this, we present TriSense, a triple-modality large language model designed for holistic video temporal understanding through the integration of visual, audio, and speech modalities. Central to TriSense is a Query-Based Connector that adaptively reweights modality contributions based on the input query, enabling robust performance under modality dropout and allowing flexible combinations of available inputs. To support TriSense's multimodal capabilities, we introduce TriSense-2M, a high-quality dataset of over 2 million curated samples generated via an automated pipeline powered by fine-tuned LLMs. TriSense-2M includes long-form videos and diverse modality combinations, facilitating broad generalization. Extensive experiments across multiple benchmarks demonstrate the effectiveness of TriSense and its potential to advance multimodal video analysis. Code and dataset will be publicly released.

Current research on long-form context in Large Language Models (LLMs) primarily focuses on the understanding of long-contexts, the Open-ended Long Text Generation (Open-LTG) remains insufficiently explored. Training a long-context generation model requires curation of gold standard reference data, which is typically nonexistent for informative Open-LTG tasks. However, previous methods only utilize general assessments as reward signals, which limits accuracy. To bridge this gap, we introduce ProxyReward, an innovative reinforcement learning (RL) based framework, which includes a dataset and a reward signal computation method. Firstly, ProxyReward Dataset generation is accomplished through simple prompts that enables the model to create automatically, obviating extensive labeled data or significant manual effort. Secondly, ProxyReward Signal offers a targeted evaluation of information comprehensiveness and accuracy for specific questions. The experimental results indicate that our method ProxyReward surpasses even GPT-4-Turbo. It can significantly enhance performance by 20% on the Open-LTG task when training widely used open-source models, while also surpassing the LLM-as-a-Judge approach. Our work presents effective methods to enhance the ability of LLMs to address complex open-ended questions posed by human.

Vietnamese, the 20th most spoken language with over 102 million native speakers, lacks robust resources for key natural language processing tasks such as text segmentation and machine reading comprehension (MRC). To address this gap, we present VSMRC, the Vietnamese Text Segmentation and Multiple-Choice Reading Comprehension Dataset. Sourced from Vietnamese Wikipedia, our dataset includes 15,942 documents for text segmentation and 16,347 synthetic multiple-choice question-answer pairs generated with human quality assurance, ensuring a reliable and diverse resource. Experiments show that mBERT consistently outperforms monolingual models on both tasks, achieving an accuracy of 88.01% on MRC test set and an F1 score of 63.15\% on text segmentation test set. Our analysis reveals that multilingual models excel in NLP tasks for Vietnamese, suggesting potential applications to other under-resourced languages. VSMRC is available at HuggingFace

Stochastic linear bandits are a fundamental model in sequential decision-making, where in each round, an agent selects a vector-valued action and receives a stochastic reward with expectation equal to the inner product between the action and an unknown parameter vector [14]. This setting has been extensively studied in the single-agent case, where the goal is to maximize cumulative reward over time through efficient exploration and exploitation [19, 12, 1]. In this paper, we consider a generalized multi-agent setting, motivated by many practical scenarios such as distributed decision systems [13, 24], personalized adaptive interfaces [16, 11], and multi-device learning [4]--where in each round, a set of m agents must each select an action and receive feedback from a stochastic, potentially heterogeneous reward function. A natural baseline approach is to run a standard single-agent linear bandit algorithm independently for each agent. While straightforward, this approach entirely ignores potential similarities across agents. When the agents' environments are similar, such independent learning leads to redundant exploration and a suboptimal use of data. In contrast, our goal is to develop collaborative learning algorithms that adaptively share information across agents--without knowing in advance how similar or different they are. This raises a fundamental question: How can collaboration reduce regret under unknown heterogeneity? To address the question, we model heterogeneity using a hierarchical/empirical Bayesian framework, where each agent's unknown parameter vector is drawn from an unknown population distribution.

Research into humanoid robots is a rapidly advancing field, with companies around the world striving to produce robots that look and act more like us. But what is it about recreating ourselves in robot form that we find so captivating? Why do humanoid robots both enthral and terrify us? And is our obsession with robotic humans just vanity, or could they play valuable roles in our future society? In this special live recording at Imperial College London as part of the Great Exhibition Road Festival, Claire chatted to Ben Russell (Science Museum), Maryam Banitalebi Dehkordi (University of Hertfordshire) and Petar Kormushev (Imperial College London) about humanoid robotics.

After Julia Dixon graduated from the University of Michigan in 2014, her family and friends asked for her help with the college application process. Dixon was happy to share her recently earned expertise about the world of higher education but soon realized how many parents and students in her community needed help and how hard it was for them to access that support. The ratio of college counselors to students in the US, according to the American School Counselor Association, is one for every 376 students. Many students don't have proper access to a college counselor to help them with admissions or pick which schools and areas of study might suit them best. Hiring a private college counselor is an option, but that can cost thousands of dollars.