Conventional human activity recognition (HAR) relies on classifiers trained to predict discrete activity classes, inherently limiting recognition to activities explicitly present in the training set. Such classifiers would invariably fail, putting zero likelihood, when encountering unseen activities. We propose Open Vocabulary HAR (OV-HAR), a framework that overcomes this limitation by first converting each activity into natural language and breaking it into a sequence of elementary motions. This descriptive text is then encoded into a fixed-size embedding. The model is trained to regress this embedding, which is subsequently decoded back into natural language using a pre-trained embedding inversion model. Unlike other works that rely on auto-regressive large language models (LLMs) at their core, OV-HAR achieves open vocabulary recognition without the computational overhead of such models. The generated text can be transformed into a single activity class using LLM prompt engineering. We have evaluated our approach on different modalities, including vision (pose), IMU, and pressure sensors, demonstrating robust generalization across unseen activities and modalities, offering a fundamentally different paradigm from contemporary classifiers.

Retrieval-Augmented Generation (RAG) systems have recently shown remarkable advancements by integrating retrieval mechanisms into language models, enhancing their ability to produce more accurate and contextually relevant responses. However, the influence of various components and configurations within RAG systems remains underexplored. A comprehensive understanding of these elements is essential for tailoring RAG systems to complex retrieval tasks and ensuring optimal performance across diverse applications. In this paper, we develop several advanced RAG system designs that incorporate query expansion, various novel retrieval strategies, and a novel Contrastive In-Context Learning RAG. Our study systematically investigates key factors, including language model size, prompt design, document chunk size, knowledge base size, retrieval stride, query expansion techniques, Contrastive In-Context Learning knowledge bases, multilingual knowledge bases, and Focus Mode retrieving relevant context at sentence-level. Through extensive experimentation, we provide a detailed analysis of how these factors influence response quality. Our findings offer actionable insights for developing RAG systems, striking a balance between contextual richness and retrieval-generation efficiency, thereby paving the way for more adaptable and high-performing RAG frameworks in diverse real-world scenarios. Our code and implementation details are publicly available.

Large language model (LLM) architectures are often described as functionally hierarchical: Early layers process syntax, middle layers begin to parse semantics, and late layers integrate information. The present work revisits these ideas. This research submits simple texts to an LLM (e.g., "A church and organ") and extracts the resulting activations. Then, for each layer, support vector machines and ridge regressions are fit to predict a text's label and thus examine whether a given layer encodes some information. Analyses using a small model (Llama-3.2-3b; 28 layers) partly bolster the common hierarchical perspective: Item-level semantics are most strongly represented early (layers 2-7), then two-item relations (layers 8-12), and then four-item analogies (layers 10-15). Afterward, the representation of items and simple relations gradually decreases in deeper layers that focus on more global information. However, several findings run counter to a steady hierarchy view: First, although deep layers can represent document-wide abstractions, deep layers also compress information from early portions of the context window without meaningful abstraction. Second, when examining a larger model (Llama-3.3-70b-Instruct), stark fluctuations in abstraction level appear: As depth increases, two-item relations and four-item analogies initially increase in their representation, then markedly decrease, and afterward increase again momentarily. This peculiar pattern consistently emerges across several experiments. Third, another emergent effect of scaling is coordination between the attention mechanisms of adjacent layers. Across multiple experiments using the larger model, adjacent layers fluctuate between what information they each specialize in representing. In sum, an abstraction hierarchy often manifests across layers, but large models also deviate from this structure in curious ways.

Multi-modal language model has made advanced progress in vision and audio, but still faces significant challenges in dealing with complex reasoning tasks in the time series domain. The reasons are twofold. First, labels for multi-modal time series data are coarse and devoid of analysis or reasoning processes. Training with these data cannot improve the model's reasoning capabilities. Second, due to the lack of precise tokenization in processing time series, the representation patterns for temporal and textual information are inconsistent, which hampers the effectiveness of multi-modal alignment. To address these challenges, we propose a multi-modal time series data construction approach and a multi-modal time series language model (TLM), TempoGPT. Specially, we construct multi-modal data for complex reasoning tasks by analyzing the variable-system relationships within a white-box system. Additionally, proposed TempoGPT achieves consistent representation between temporal and textual information by quantizing temporal embeddings, where temporal embeddings are quantized into a series of discrete tokens using a predefined codebook; subsequently, a shared embedding layer processes both temporal and textual tokens. Extensive experiments demonstrate that TempoGPT accurately perceives temporal information, logically infers conclusions, and achieves state-of-the-art in the constructed complex time series reasoning tasks. Moreover, we quantitatively demonstrate the effectiveness of quantizing temporal embeddings in enhancing multi-modal alignment and the reasoning capabilities of TLMs. Code and data are available at https://github.com/zhanghaochuan20/TempoGPT.

Foundation models require fine-tuning to ensure their generative outputs align with intended results for specific tasks. Automating this fine-tuning process is challenging, as it typically needs human feedback that can be expensive to acquire. We present AutoRefine, a method that leverages reinforcement learning for targeted fine-tuning, utilizing direct feedback from measurable performance improvements in specific downstream tasks. We demonstrate the method for a problem arising in algorithmic hiring platforms where linguistic biases influence a recommendation system. In this setting, a generative model seeks to rewrite given job specifications to receive more diverse candidate matches from a recommendation engine which matches jobs to candidates. Our model detects and regulates biases in job descriptions to meet diversity and fairness criteria. The experiments on a public hiring dataset and a real-world hiring platform showcase how large language models can assist in identifying and mitigation biases in the real world.

Data quality is crucial for training Large Language Models (LLMs). Traditional heuristic filters often miss low-quality text or mistakenly remove valuable content. In this paper, we introduce an LLM-based line-level filtering method to enhance training data quality. We use GPT-4o mini to label a 20,000-document sample from FineWeb at the line level, allowing the model to create descriptive labels for low-quality lines. These labels are grouped into nine main categories, and we train a DeBERTa-v3 classifier to scale the filtering to a 10B-token subset of FineWeb. To test the impact of our filtering, we train GPT-2 models on both the original and the filtered datasets. The results show that models trained on the filtered data achieve higher accuracy on the HellaSwag benchmark and reach their performance targets faster, even with up to 25\% less data. This demonstrates that LLM-based line-level filtering can significantly improve data quality and training efficiency for LLMs. We release our quality-annotated dataset, FinerWeb-10BT, and the codebase to support further work in this area.

Process Reward Models (PRMs) emerge as a promising approach for process supervision in mathematical reasoning of Large Language Models (LLMs), which aim to identify and mitigate intermediate errors in the reasoning processes. However, the development of effective PRMs faces significant challenges, particularly in data annotation and evaluation methodologies. In this paper, through extensive experiments, we demonstrate that commonly used Monte Carlo (MC) estimation-based data synthesis for PRMs typically yields inferior performance and generalization compared to LLM-as-a-judge and human annotation methods. MC estimation relies on completion models to evaluate current-step correctness, leading to inaccurate step verification. Furthermore, we identify potential biases in conventional Best-of-N (BoN) evaluation strategies for PRMs: (1) The unreliable policy models generate responses with correct answers but flawed processes, leading to a misalignment between the evaluation criteria of BoN and the PRM objectives of process verification. (2) The tolerance of PRMs of such responses leads to inflated BoN scores. (3) Existing PRMs have a significant proportion of minimum scores concentrated on the final answer steps, revealing the shift from process to outcome-based assessment in BoN Optimized PRMs. To address these challenges, we develop a consensus filtering mechanism that effectively integrates MC estimation with LLM-as-a-judge and advocates a more comprehensive evaluation framework that combines response-level and step-level metrics. Based on the mechanisms, we significantly improve both model performance and data efficiency in the BoN evaluation and the step-wise error identification task. Finally, we release a new state-of-the-art PRM that outperforms existing open-source alternatives and provides practical guidelines for future research in building process supervision models.

The centralization of Large Language Models (LLMs) development has created significant barriers to AI advancement, limiting the democratization of these powerful technologies. This centralization, coupled with the scarcity of high-quality training data and mounting complexity of maintaining comprehensive expertise across rapidly expanding knowledge domains, poses critical challenges to the continued growth of LLMs. While solutions like Retrieval-Augmented Generation (RAG) offer potential remedies, maintaining up-to-date expert knowledge across diverse domains remains a significant challenge, particularly given the exponential growth of specialized information. This paper introduces LLMs Networks (LLM-Net), a blockchain-based framework that democratizes LLMs-as-a-Service through a decentralized network of specialized LLM providers. By leveraging collective computational resources and distributed domain expertise, LLM-Net incorporates fine-tuned expert models for various specific domains, ensuring sustained knowledge growth while maintaining service quality through collaborative prompting mechanisms. The framework's robust design includes blockchain technology for transparent transaction and performance validation, establishing an immutable record of service delivery. Our simulation, built on top of state-of-the-art LLMs such as Claude 3.5 Sonnet, Llama 3.1, Grok-2, and GPT-4o, validates the effectiveness of the reputation-based mechanism in maintaining service quality by selecting high-performing respondents (LLM providers). Thereby it demonstrates the potential of LLM-Net to sustain AI advancement through the integration of decentralized expertise and blockchain-based accountability.

Lifelong learning, also known as continual or incremental learning, is a crucial component for advancing Artificial General Intelligence (AGI) by enabling systems to continuously adapt in dynamic environments. While large language models (LLMs) have demonstrated impressive capabilities in natural language processing, existing LLM agents are typically designed for static systems and lack the ability to adapt over time in response to new challenges. This survey is the first to systematically summarize the potential techniques for incorporating lifelong learning into LLM-based agents. We categorize the core components of these agents into three modules: the perception module for multimodal input integration, the memory module for storing and retrieving evolving knowledge, and the action module for grounded interactions with the dynamic environment. We highlight how these pillars collectively enable continuous adaptation, mitigate catastrophic forgetting, and improve long-term performance. This survey provides a roadmap for researchers and practitioners working to develop lifelong learning capabilities in LLM agents, offering insights into emerging trends, evaluation metrics, and application scenarios. Relevant literature and resources are available at \href{this url}{https://github.com/qianlima-lab/awesome-lifelong-llm-agent}.

The integrity of time series data in smart grids is often compromised by missing values due to sensor failures, transmission errors, or disruptions. Gaps in smart meter data can bias consumption analyses and hinder reliable predictions, causing technical and economic inefficiencies. As smart meter data grows in volume and complexity, conventional techniques struggle with its nonlinear and nonstationary patterns. In this context, Generative Artificial Intelligence offers promising solutions that may outperform traditional statistical methods. In this paper, we evaluate two general-purpose Large Language Models and five Time Series Foundation Models for smart meter data imputation, comparing them with conventional Machine Learning and statistical models. We introduce artificial gaps (30 minutes to one day) into an anonymized public dataset to test inference capabilities. Results show that Time Series Foundation Models, with their contextual understanding and pattern recognition, could significantly enhance imputation accuracy in certain cases. However, the trade-off between computational cost and performance gains remains a critical consideration.