Large Language Models (LLMs) hold immense promise for revolutionizing financial analysis and decision-making, yet their direct application is often hampered by issues of data hallucination and lack of access to real-time, verifiable financial information. This paper introduces QuantMCP, a novel framework designed to rigorously ground LLMs in financial reality. By leveraging the Model Context Protocol (MCP) for standardized and secure tool invocation, QuantMCP enables LLMs to accurately interface with a diverse array of Python-accessible financial data APIs (e.g., Wind, yfinance). Users can interact via natural language to precisely retrieve up-to-date financial data, thereby overcoming LLM's inherent limitations in factual data recall. More critically, once furnished with this verified, structured data, the LLM's analytical capabilities are unlocked, empowering it to perform sophisticated data interpretation, generate insights, and ultimately support more informed financial decision-making processes. QuantMCP provides a robust, extensible, and secure bridge between conversational AI and the complex world of financial data, aiming to enhance both the reliability and the analytical depth of LLM applications in finance.

Large Language Models (LLMs) have achieved remarkable success in tasks requiring complex reasoning, such as code generation, mathematical problem solving, and algorithmic synthesis -- especially when aided by reasoning tokens and Chain-of-Thought prompting. Yet, a core question remains: do these models truly reason, or do they merely exploit shallow statistical patterns? In this paper, we introduce Chain-of-Code Collapse, where we systematically investigate the robustness of reasoning LLMs by introducing a suite of semantically faithful yet adversarially structured prompt perturbations. Our evaluation -- spanning 700 perturbed code generations derived from LeetCode-style problems -- applies transformations such as storytelling reframing, irrelevant constraint injection, example reordering, and numeric perturbation. We observe that while certain modifications severely degrade performance (with accuracy drops up to -42.1%), others surprisingly improve model accuracy by up to 35.3%, suggesting sensitivity not only to semantics but also to surface-level prompt dynamics. These findings expose the fragility and unpredictability of current reasoning systems, underscoring the need for more principles approaches to reasoning alignments and prompting robustness. We release our perturbation datasets and evaluation framework to promote further research in trustworthy and resilient LLM reasoning.

Machine learning approaches, such as contextual multi-armed bandit (cMAB) algorithms, offer a promising strategy to reduce sedentary behavior by delivering personalized interventions to encourage physical activity. However, cMAB algorithms typically require large participant samples to learn effectively and may overlook key psychological factors that are not explicitly encoded in the model. In this study, we propose a hybrid approach that combines cMAB for selecting intervention types with large language models (LLMs) to personalize message content. We evaluate four intervention types: behavioral self-monitoring, gain-framed, loss-framed, and social comparison, each delivered as a motivational message aimed at increasing motivation for physical activity and daily step count. Message content is further personalized using dynamic contextual factors including daily fluctuations in self-efficacy, social influence, and regulatory focus. Over a seven-day trial, participants receive daily messages assigned by one of four models: cMAB alone, LLM alone, combined cMAB with LLM personalization (cMABxLLM), or equal randomization (RCT). Outcomes include daily step count and message acceptance, assessed via ecological momentary assessments (EMAs). We apply a causal inference framework to evaluate the effects of each model. Our findings offer new insights into the complementary roles of LLM-based personalization and cMAB adaptation in promoting physical activity through personalized behavioral messaging.

Training large language models is generally done via optimization methods on clusters containing tens of thousands of accelerators, communicating over a high-bandwidth interconnect. Scaling up these clusters is expensive and can become impractical, imposing limits on the size of models that can be trained. Several recent studies have proposed training methods that are less communication intensive, avoiding the need for a highly connected compute cluster. These state-of-the-art low communication training methods still employ a synchronization step for model parameters, which, when performed over all model replicas, can become costly on a low-bandwidth network. In this work, we propose a novel optimization method, NoLoCo, that does not explicitly synchronize all model parameters during training and, as a result, does not require any collective communication. NoLoCo implicitly synchronizes model weights via a novel variant of the Nesterov momentum optimizer by partially averaging model weights with a randomly selected other one. We provide both a theoretical convergence analysis for our proposed optimizer as well as empirical results from language model training. We benchmark NoLoCo on a wide range of accelerator counts and model sizes, between 125M to 6.8B parameters. Our method requires significantly less communication overhead than fully sharded data parallel training or even widely used low communication training method, DiLoCo. The synchronization step itself is estimated to be one magnitude faster than the all-reduce used in DiLoCo for few hundred accelerators training over the internet. We also do not have any global blocking communication that reduces accelerator idling time. Compared to DiLoCo, we also observe up to $4\%$ faster convergence rate with wide range of model sizes and accelerator counts.

We introduce Magistral, Mistral's first reasoning model and our own scalable reinforcement learning (RL) pipeline. Instead of relying on existing implementations and RL traces distilled from prior models, we follow a ground up approach, relying solely on our own models and infrastructure. Notably, we demonstrate a stack that enabled us to explore the limits of pure RL training of LLMs, present a simple method to force the reasoning language of the model, and show that RL on text data alone maintains most of the initial checkpoint's capabilities. We find that RL on text maintains or improves multimodal understanding, instruction following and function calling. We present Magistral Medium, trained for reasoning on top of Mistral Medium 3 with RL alone, and we open-source Magistral Small (Apache 2.0) which further includes cold-start data from Magistral Medium.

Online learning has experienced rapid growth due to its flexibility and accessibility. Personalization, adapted to the needs of individual learners, is crucial for enhancing the learning experience, particularly in online settings. A key aspect of personalization is providing learners with answers customized to their specific questions. This paper therefore explores the potential of Large Language Models ( LLMs) to generate personalized answers to learners' questions, thereby enhancing engagement and reducing the workload on educators. To evaluate the effectiveness of LLMs in this context, we conducted a comprehensive study using the StackExchange platform in two distinct areas: language learning and programming. We developed a framework and a dataset for validating automatically generated personalized answers. Subsequently, we generated personalized answers using different strategies, including 0-shot, 1-shot, and few-shot scenarios. The generated answers were evaluated using three methods: 1. BERTScore, 2. LLM evaluation, and 3. human evaluation. Our findings indicated that providing LLMs with examples of desired answers (from the learner or similar learners) can significantly enhance the LLMs' ability to tailor responses to individual learners' needs.

To advance time series forecasting (TSF), various methods have been proposed to improve prediction accuracy, evolving from statistical techniques to data-driven deep learning architectures. Despite their effectiveness, most existing methods still adhere to a fast thinking paradigm-relying on extracting historical patterns and mapping them to future values as their core modeling philosophy, lacking an explicit thinking process that incorporates intermediate time series reasoning. Meanwhile, emerging slow-thinking LLMs (e.g., OpenAI-o1) have shown remarkable multi-step reasoning capabilities, offering an alternative way to overcome these issues. However, prompt engineering alone presents several limitations - including high computational cost, privacy risks, and limited capacity for in-depth domain-specific time series reasoning. To address these limitations, a more promising approach is to train LLMs to develop slow thinking capabilities and acquire strong time series reasoning skills. For this purpose, we propose Time-R1, a two-stage reinforcement fine-tuning framework designed to enhance multi-step reasoning ability of LLMs for time series forecasting. Specifically, the first stage conducts supervised fine-tuning for warmup adaptation, while the second stage employs reinforcement learning to improve the model's generalization ability. Particularly, we design a fine-grained multi-objective reward specifically for time series forecasting, and then introduce GRIP (group-based relative importance for policy optimization), which leverages non-uniform sampling to further encourage and optimize the model's exploration of effective reasoning paths. Experiments demonstrate that Time-R1 significantly improves forecast performance across diverse datasets.

Large language models (LLMs) are widely used to evaluate the quality of LLM generations and responses, but this leads to significant challenges: high API costs, uncertain reliability, inflexible pipelines, and inherent biases. To address these, we introduce PAJAMA (Program-As-a-Judge for Automated Model Assessment), a new alternative that uses LLMs to synthesize executable judging programs instead of directly scoring responses. These synthesized programs can be stored and run locally, costing orders of magnitude less while providing interpretable, and auditable judging logic that can be easily adapted. Program-based judges mitigate biases, improving judgment consistency by 15.83% and reducing biased responses by 23.7% on average compared to a Qwen2.5-14B-based LLM-as-a-judge. When program judgments are distilled into a model, PAJAMA outperforms LLM-as-a-judge on the challenging CHAT-HARD subset of RewardBench, outperforming metrics by 2.19% on Prometheus and 8.67% on the JudgeLM dataset, all at three orders of magnitude lower cost.

A key ingredient in machine learning and statistical pipelines alike is the availability of large amounts of high-quality labeled data. Breakthroughs in computer vision stem from the collection of millions of labeled images [8]; social science research relies on extensively labeled datasets to understand human behavior and opinions [22]. While acquiring unlabeled data (e.g., raw images or texts from the internet) can be relatively inexpensive, acquiring high-quality labels is typically an endeavor that requires significant time and effort from human experts. Given the expense of collecting high-quality labels, an enticing prospect is to use increasingly powerful AI models to predict labels for datasets, bypassing the need for human experts entirely. Indeed, recent works have demonstrated AI models' ability to predict protein structures [17], to evaluate language model responses [39], and even to simulate human experimental subjects [23]. These advances highlight the potential for AI to streamline data annotation, and to produce high-quality labels at a fraction of the cost of human experts. The problem with such an approach is that AI models are not always correct, and in particular come with no guarantees on how well they will label a given dataset. This makes it untenable to use AI-predicted labels as a direct substitute for human labels, particularly in settings where label quality is critical--for instance, in high-stakes applications like medical diagnosis, or when the downstream task is to draw conclusions that inform policy decisions. Motivated by this state of affairs, in this paper we ask: Can we leverage powerful AI models to label data, while still guaranteeing quality?

Faithful evaluation of language model capabilities is crucial for deriving actionable insights that can inform model development. However, rigorous causal evaluations in this domain face significant methodological challenges, including complex confounding effects and prohibitive computational costs associated with extensive retraining. To tackle these challenges, we propose a causal representation learning framework wherein observed benchmark performance is modeled as a linear transformation of a few latent capability factors. Crucially, these latent factors are identified as causally interrelated after appropriately controlling for the base model as a common confounder. Applying this approach to a comprehensive dataset encompassing over 1500 models evaluated across six benchmarks from the Open LLM Leaderboard, we identify a concise three-node linear causal structure that reliably explains the observed performance variations. Further interpretation of this causal structure provides substantial scientific insights beyond simple numerical rankings: specifically, we reveal a clear causal direction starting from general problem-solving capabilities, advancing through instruction-following proficiency, and culminating in mathematical reasoning ability. Our results underscore the essential role of carefully controlling base model variations during evaluation, a step critical to accurately uncovering the underlying causal relationships among latent model capabilities.