Recent advancements in Large Language Models (LLMs) for code optimization have enabled industrial platforms to automate software performance engineering at unprecedented scale and speed. Yet, organizations in regulated industries face strict constraints on which LLMs they can use - many cannot utilize commercial models due to data privacy regulations and compliance requirements, creating a significant challenge for achieving high-quality code optimization while maintaining cost-effectiveness. We address this by implementing a Mixture-of-Agents (MoA) approach that directly synthesizes code from multiple specialized LLMs, comparing it against TurinTech AI's vanilla Genetic Algorithm (GA)-based ensemble system and individual LLM optimizers using real-world industrial codebases. Our key contributions include: (1) First MoA application to industrial code optimization using real-world codebases; (2) Empirical evidence that MoA excels with open-source models, achieving 14.3% to 22.2% cost savings and 28.6% to 32.2% faster optimization times for regulated environments; (3) Deployment guidelines demonstrating GA's advantage with commercial models while both ensembles outperform individual LLMs; and (4) Real-world validation across 50 code snippets and seven LLM combinations, generating over 8,700 variants, addresses gaps in industrial LLM ensemble evaluation. This provides actionable guidance for organizations balancing regulatory compliance with optimization performance in production environments.

Tasks that require character-level reasoning, such as counting or locating characters within words, remain challenging for contemporary language models. A common conjecture is that language models' reliance on subword units, rather than characters, contributes to their struggles with character-level tasks, yet recent studies offer conflicting conclusions about the role of tokenization, leaving its impact unclear. To address this gap, we introduce CharBench, a comprehensive benchmark of character-level tasks that is two orders of magnitude larger than existing alternatives. We evaluate a diverse range of leading open-weight and proprietary models on CharBench and find that it presents a significant challenge to modern LLMs, with an average accuracy of 43.6% and 32.3% on some tasks. We present an in-depth analysis of how intrinsic properties of words and their segmentations into tokens correspond to model performance. For counting tasks, we find that tokenization properties are weakly correlated with correctness, while the length of the queried word and the actual character count play a more significant part. In contrast, for tasks requiring intra-word positional understanding, performance is negatively correlated with the length of the token containing the queried character, suggesting that longer tokens obscure character position information for LLMs. We encourage future work to build on the benchmark and evaluation methodology introduced here as tools for improving model performance on such tasks.

Reinforcement learning-based retrieval-augmented generation (RAG) methods enhance the reasoning abilities of large language models (LLMs). However, most rely only on final-answer rewards, overlooking intermediate reasoning quality. This paper analyzes existing RAG reasoning models and identifies three main failure patterns: (1) information insufficiency, meaning the model fails to retrieve adequate support; (2) faulty reasoning, where logical or content-level flaws appear despite sufficient information; and (3) answer-reasoning inconsistency, where a valid reasoning chain leads to a mismatched final answer. We propose TIRESRAG-R1, a novel framework using a think-retrieve-reflect process and a multi-dimensional reward system to improve reasoning and stability. TIRESRAG-R1 introduces: (1) a sufficiency reward to encourage thorough retrieval; (2) a reasoning quality reward to assess the rationality and accuracy of the reasoning chain; and (3) a reflection reward to detect and revise errors. It also employs a difficulty-aware reweighting strategy and training sample filtering to boost performance on complex tasks. Experiments on four multi-hop QA datasets show that TIRESRAG-R1 outperforms prior RAG methods and generalizes well to single-hop tasks. The code and data are available at: https://github.com/probe2/TIRESRAG-R1.

Charts are a fundamental visualization format widely used in data analysis across research and industry. While enabling users to edit charts based on high-level intentions is of great practical value, existing methods primarily rely on natural language instructions, which are often too ambiguous to support fine-grained editing. In this work, we introduce a novel paradigm for multimodal chart editing, where user intent is expressed through a combination of natural language and visual indicators that explicitly highlight the elements to be modified. To support this paradigm, we present Chart$\text{M}^3$, a new benchmark for Multimodal chart editing with Multi-level complexity and Multi-perspective evaluation. Chart$\text{M}^3$ contains 1,000 samples spanning four levels of editing difficulty. Each sample includes triplets in the form of (chart, code, multimodal instructions). To comprehensively evaluate chart editing models, Chart$\text{M}^3$ provides metrics that assess both visual appearance and code correctness. Our benchmark reveals significant limitations in current multimodal large language models (MLLMs), including GPT-4o, particularly in their ability to interpret and act on visual indicators. To address this, we construct Chart$\text{M}^3$-Train, a large-scale training set with 24,000 multimodal chart editing samples. Fine-tuning MLLMs on this dataset leads to substantial improvements, demonstrating the importance of multimodal supervision in building practical chart editing systems. Our datasets, codes, and evaluation tools are available at https://github.com/MLrollIT/ChartM3. %https://github.com/MLrollIT/ChartM3Our datasets, codes, and evaluation tools are available at https://github.com/yaolinli/VCE.

The emergence of reasoning models and their integration into practical AI chat bots has led to breakthroughs in solving advanced math, deep search, and extractive question answering problems that requires a complex and multi-step thought process. Yet, a complete understanding of why these models hallucinate more than general purpose language models is missing. In this investigative study, we systematicallyexplore reasoning failures of contemporary language models on multi-hop question answering tasks. We introduce a novel, nuanced error categorization framework that examines failures across three critical dimensions: the diversity and uniqueness of source documents involved ("hops"), completeness in capturing relevant information ("coverage"), and cognitive inefficiency ("overthinking"). Through rigorous hu-man annotation, supported by complementary automated metrics, our exploration uncovers intricate error patterns often hidden by accuracy-centric evaluations. This investigative approach provides deeper insights into the cognitive limitations of current models and offers actionable guidance toward enhancing reasoning fidelity, transparency, and robustness in future language modeling efforts.

LLM-powered conversational assistants are often deployed in a one-size-fits-all manner, which fails to accommodate individual user preferences. Recently, LLM personalization -- tailoring models to align with specific user preferences -- has gained increasing attention as a way to bridge this gap. In this work, we specifically focus on a practical yet challenging setting where only a small set of preference annotations can be collected per user -- a problem we define as Personalized Preference Alignment with Limited Data (PPALLI). To support research in this area, we introduce two datasets -- DnD and ELIP -- and benchmark a variety of alignment techniques on them. We further propose FaST, a highly parameter-efficient approach that leverages high-level features automatically discovered from the data, achieving the best overall performance.

How can we build general-purpose robot systems for open-world semantic navigation, e.g., searching a novel environment for a target object specified in natural language? To tackle this challenge, we introduce OSG Navigator, a modular system composed of foundation models, for open-world Object-Goal Navigation (ObjectNav). Foundation models provide enormous semantic knowledge about the world, but struggle to organise and maintain spatial information effectively at scale. Key to OSG Navigator is the Open Scene Graph representation, which acts as spatial memory for OSG Navigator. It organises spatial information hierarchically using OSG schemas, which are templates, each describing the common structure of a class of environments. OSG schemas can be automatically generated from simple semantic labels of a given environment, e.g., "home" or "supermarket". They enable OSG Navigator to adapt zero-shot to new environment types. We conducted experiments using both Fetch and Spot robots in simulation and in the real world, showing that OSG Navigator achieves state-of-the-art performance on ObjectNav benchmarks and generalises zero-shot over diverse goals, environments, and robot embodiments.

Abstract--The continual learning capability of large language models (LLMs) is crucial for advancing artificial general intelligence. However, continual fine-tuning LLMs across various domains often suffers from catastrophic forgetting, characterized by: 1) significant forgetting of their general capabilities, and 2) sharp performance declines in previously learned tasks. T o simultaneously address both issues in a simple yet stable manner, we propose General Sample Replay (GeRe), a framework that use usual pretraining texts for efficient anti-forgetting. Beyond revisiting the most prevalent replay-based practices under GeRe, we further leverage neural states to introduce a enhanced activation states constrained optimization method using threshold-based margin (TM) loss, which maintains activation state consistency during replay learning. We are the first to validate that a small, fixed set of pre-collected general replay samples is sufficient to resolve both concerns--retaining general capabilities while promoting overall performance across sequential tasks. Indeed, the former can inherently facilitate the latter. Through controlled experiments, we systematically compare TM with different replay strategies under the GeRe framework, including vanilla label fitting, logit imitation via KL divergence and feature imitation via L1/L2 losses. Results demonstrate that TM consistently improves performance and exhibits better robustness. Our work paves the way for efficient replay of LLMs for the future. Our code and data are available at https://github.com/Qznan/GeRe. The finetuned model forgets its original world knowledge or basic instruction-following skills [1], [2]. Y unan Zhang, Shuoran Jiang, Y ang Fan, Mengchen Zhao, Xiangping Wu, Qingcai Chen are with the Department of Computer Science and T echnology, Harbin Institute of T echnology, Shenzhen, China. Qingcai Chen and Xiangping Wu are the corresponding authors.

Studies on the use o f AI tools in CS course s focus on prescriptive uses of AI tools that preemptively assign a use case to the tool. To our knowledge, this is the first user study that surveys CS students on how they use AI tools for their coding tasks by personal choice. We surveyed 26 CS students and practitioners with various programming experiences using AI tools for coding tasks (i.e., writ e, debug, etc.). When asked about the most common resources they have used to write a 300 - line program from scratch, blog entries ( e.g., Stack Exchange, etc.) were their top choice, followed by AI coding assistants (e.g., GitHub Copilot, etc.). When asked about resources to debug code, AI chatbots (e.g., ChatGPT, etc.) were the most common choice, followed by blog entries. AI coding assistants are used more for writing code, while AI chatbots are used for debugging tasks. Respondents with all programming experience prefer online resources for coding tasks - whether AI - powered or not - rather than direct human help from peers an d instructors.

Group Relative Policy Optimization (GRPO) has proven to be an effective tool for post-training language models (LMs). However, AI systems are increasingly expressed as modular programs that mix together multiple LM calls with distinct prompt templates and other tools, and it is not clear how best to leverage GRPO to improve these systems. We begin to address this challenge by defining mmGRPO, a simple multi-module generalization of GRPO that groups LM calls by module across rollouts and handles variable-length and interrupted trajectories. We find that mmGRPO, composed with automatic prompt optimization, improves accuracy by 11% on average across classification, many-hop search, and privacy-preserving delegation tasks against the post-trained LM, and by 5% against prompt optimization on its own. We open-source mmGRPO in DSPy as the dspy.GRPO optimizer.