The scaling of Generative AI (GenAI) models into the hundreds of billions of parameters makes low-precision computation indispensable for efficient deployment. We argue that the fundamental solution lies in developing low-precision floating-point formats, which inherently provide numerical stability, memory savings, and hardware efficiency without dequantization overhead. In this paper, we present a theoretical and empirical study of an exponent concentration phenomenon in GenAI weights: exponents consistently exhibit low entropy across architectures and modalities. We show that this arises naturally from $α$-stable distributions induced by stochastic gradient descent, and we prove tight bounds on the entropy of exponents. Our analysis establishes a theoretical compression limit near FP4.67, which motivates the design of a practical FP8 format. Building on these insights, we propose Exponent-Concentrated FP8 (ECF8), a lossless compression framework with entropy-aware encoding and GPU-optimized decoding. Experiments on LLMs and DiTs up to 671B parameters demonstrate up to 26.9% memory savings and 177.1% throughput acceleration, with perfectly lossless computations, i.e., no deviation in model outputs. Our results establish exponent concentration as a statistical law of trained models and open a principled path for lossless low-precision floating-point design in the FP8 era.

The rapid adoption of Large Language Models (LLMs) has driven a growing demand for efficient inference, particularly in latency-sensitive applications such as chatbots and personalized assistants. Unlike traditional deep neural networks, LLM inference proceeds in two distinct phases: the prefill phase, which processes the full input sequence in parallel, and the decode phase, which generates tokens sequentially. These phases exhibit highly diverse compute and memory requirements, which makes accelerator design particularly challenging. Prior works have primarily been optimized for high-batch inference or evaluated only short input context lengths, leaving the low-batch and long context regime, which is critical for interactive applications, largely underexplored. We propose HALO, a heterogeneous memory centric accelerator designed for these unique challenges of prefill and decode phases in low-batch LLM inference. HALO integrates HBM based Compute-in-DRAM (CiD) with an on-chip analog Compute-in-Memory (CiM), co-packaged using 2.5D integration. To further improve the hardware utilization, we introduce a phase-aware mapping strategy that adapts to the distinct demands of the prefill and decode phases. Compute bound operations in the prefill phase are mapped to CiM to exploit its high throughput matrix multiplication capability, while memory-bound operations in the decode phase are executed on CiD to benefit from reduced data movement within DRAM. Additionally, we present an analysis of the performance tradeoffs of LLMs under two architectural extremes: a fully CiD and a fully on-chip analog CiM design to highlight the need for a heterogeneous design. We evaluate HALO on LLaMA-2 7B and Qwen3 8B models. Our experimental results show that LLMs mapped to HALO achieve up to 18x geometric mean speedup over AttAcc, an attention-optimized mapping and 2.5x over CENT, a fully CiD based mapping.

Abstract--Foundation models have revolutionized artificial intelligence, yet their application in recommender systems remains limited by reasoning opacity and knowledge constraints. This paper introduces AgenticRAG, a novel framework that combines tool-augmented foundation models with retrieval-augmented generation for zero-shot explainable recommendations. Our approach integrates external tool invocation, knowledge retrieval, and chain-of-thought reasoning to create autonomous recommendation agents capable of transparent decision-making without task-specific training. Experimental results on three real-world datasets demonstrate that AgenticRAG achieves consistent improvements over state-of-the-art baselines, with NDCG@10 improvements of 0.4% on Amazon Electronics, 0.8% on MovieLens-1M, and 1.6% on Y elp datasets. The framework exhibits superior explainability while maintaining computational efficiency comparable to traditional methods.

Recent advancements in self-improvement for Large Language Models (LLMs) have efficiently enhanced model capabilities without significantly increasing costs, particularly in terms of human effort. While this area is still relatively young, its extension to the multimodal domain holds immense potential for leveraging diverse data sources and developing more general self-improving models. This survey is the first to provide a comprehensive overview of self-improvement in Multimodal LLMs (MLLMs). We provide a structured overview of the current literature and discuss methods from three perspectives: 1) data collection, 2) data organization, and 3) model optimization, to facilitate the further development of self-improvement in MLLMs. We also include commonly used evaluations and downstream applications. Finally, we conclude by outlining open challenges and future research directions.

As students increasingly adopt large language models (LLMs) as learning aids, it is crucial to build models that are adept at handling the nuances of tutoring: they need to identify the core needs of students, be adaptive, provide personalized guidance, and be accurate. To this end, we introduce TutorBench, a dataset and evaluation benchmark designed to rigorously evaluate the core tutoring skills of LLMs. The dataset comprises 1,490 samples curated by human experts, focused on high-school and AP-level curricula. The samples are drawn from three common tutoring tasks: (i) generating adaptive explanations tailored to a student's confusion, (ii) providing actionable feedback on a student's work, and (iii) promoting active learning through effective hint generation. To account for the inherent complexity of tutoring, samples are accompanied by sample-specific rubrics which are used to judge model responses during evaluation. TutorBench uses a reliable and fine-grained automatic evaluation method that uses an LLM-judge and the sample-specific rubrics. We evaluate 16 frontier LLMs on TutorBench and present a detailed analysis of their performance and behavior. Our results show that none of the frontier LLMs achieve a score of greater than $56\%$, showing a large room for improvement. We find that LLMs fall short in exhibiting the full range of tutoring skills needed to guide, diagnose, and support students effectively, with all the frontier models achieving less than a $60\%$ pass rate on rubric criteria related to these skills. We also find that different model families exhibit varied strengths and limitations: the Claude models outperform others in supporting active learning, while they lag behind in the other two use cases. By releasing TutorBench, we provide a comprehensive and unsaturated benchmark to guide the development of the next-generation of AI tutors.

This study presents the development of Geolog-IA, a novel conversational system based on artificial intelligence that responds naturally to questions about geology theses from the Central University of Ecuador. Our proposal uses the Llama 3.1 and Gemini 2.5 language models, which are complemented by a Retrieval Augmented Generation (RAG) architecture and an SQLite database. This strategy allows us to overcome problems such as hallucinations and outdated knowledge. The evaluation of Geolog-IA's performance with the BLEU metric reaches an average of 0.87, indicating high consistency and accuracy in the responses generated. The system offers an intuitive, web-based interface that facilitates interaction and information retrieval for directors, teachers, students, and administrative staff at the institution. This tool can be a key support in education, training, and research and establishes a basis for future applications in other disciplines.

Recent developments have enabled Large Language Models (LLMs) to engage in complex reasoning tasks through deep thinking. However, the capacity of reasoning has not been successfully transferred to non-high-resource languages due to resource constraints, which struggles with multilingual reasoning tasks. To this end, we propose Structured-of-Thought (SoT), a training-free method that improves the performance on multilingual reasoning through a multi-step transformation: Language Thinking Transformation and Structured Knowledge Transformation. The SoT method converts language-specific semantic information into language-agnostic structured representations, enabling the models to understand the query in different languages more sophisticated. Besides, SoT effectively guides LLMs toward more concentrated reasoning to maintain consistent underlying reasoning pathways when handling cross-lingual variations in expression. Experimental results demonstrate that SoT outperforms several strong baselines on multiple multilingual reasoning benchmarks when adapting to various backbones of LLMs. It can also be integrated with other training-free strategies for further improvements. Our code is available at https://github.com/Cherry-qwq/SoT.

As Large Language Models (LLMs) are increasingly deployed in customer-facing applications, a critical yet underexplored question is how users communicate differently with LLM chatbots compared to human agent. In this study, we present empirical evidence that users adopt distinct communication styles when users interact with chatbots versus human agents. Our analysis reveals significant differences in grammatical fluency, politeness, and lexical diversity in user language between the two settings. These findings suggest that models trained exclusively on human-human interaction data may not adequately accommodate the communication style shift that occurs once an LLM chatbot is deployed. To enhance LLM robustness to post-launch communication style changes, we experimented with two strategies: (1) data augmentation during the post-training phase and (2) inference-time user message reformulation. Our results indicate that models trained on stylistically diverse datasets significantly outperform those trained exclusively on original or stylistically uniform datasets, while inference-time reformulation proved less effective. These insights help us to better adapt our models for improved LLM-user interaction experiences.

Building officials, especially those in resource - constrained or rural jurisdictions, struggle with labor - intensive, error - prone, and costly manual reviews of design documents as projects scale in size and complexity. Widespread adoption of Building Information Modeling (BIM) and Large Language Models (LLMs) has created opportunities for automated code review (AC R) solutions . This study proposes a novel agent - driven framework that integrates BIM - based data extraction with automated verification using both re trieval - augmented generation (RAG) and Model Context Protocol (MCP) agent pipelines. The framework employs LLM - enabled agents to extract geometry, schedules, and system attributes from heterogeneous file types, which are then processed for building code checking via two complementary mechanisms: (i) direct API calls to DOE's COMcheck engine, providing deterministic and audit - ready outputs, and (ii) RAG - based reasoning over rule provisions, allowing flexible interpretation where coverage is incomplete or amb iguous . The framework was evaluated through case demonstrations, including automated extraction of geometric attributes (e.g., surface area, tilt, and insulation values), parsing of operational schedules, and design validation for lighting allowances under ASHRAE Standard 90.1 - 2022. Comparative performance tests across multiple large language models showed that Generative Pre - trained Transformer 4 Omni (GPT - 4o) achieved the best balance of efficiency and stability, while smaller models exhibited inconsistenc ies or failure s . Results confirm that MCP agent pipelines perform better than RAG reasoning pipelines on rigor and flexibility in workflows.

Large language models (LLMs) can improve reasoning at inference time through test-time scaling (TTS), where multiple reasoning traces are generated and the best one is selected. Prior work shows that increasing the number of samples K steadily improves accuracy. In this paper, we demonstrate that this trend does not hold indefinitely: at large K, further scaling yields no gains, and certain hard questions remain unsolved regardless of the number of traces. Interestingly, we find that different sampling temperatures solve different subsets of problems, implying that single-temperature scaling explores only part of a model's potential. We therefore propose scaling along the temperature dimension, which enlarges the reasoning boundary of LLMs. Temperature scaling also enables base models to reach performance comparable to reinforcement learning (RL)-trained counterparts, without additional post-training. We further provide a comprehensive analysis of this phenomenon and design a multi-temperature voting method that reduces the overhead of temperature scaling. Overall, our findings suggest that TTS is more powerful than previously thought, and that temperature scaling offers a simple and effective way to unlock the latent potential of base models. Large language models (LLMs) have demonstrated strong reasoning capabilities for complex problems at test time (Wei et al., 2022). As illustrated in Figure 1a, two main approaches have emerged to achieve such reasoning. The first trains models to produce long reasoning traces with self-reflection and correction, often implemented through reinforcement learning (RL) (Guo et al., 2025a; Y ang et al., 2025c). While effective, this approach requires costly and time-consuming training (Liu et al., 2025a). The second, known as test-time scaling (TTS) (Brown et al., 2024; Snell et al., 2025; Zhao et al., 2025), shifts the burden to inference: the model generates multiple reasoning traces in parallel and a verifier selects the most reliable one (Saad-Falcon et al., 2025a).