End-to-end Large Speech Language Models~(\textbf{LSLMs}) demonstrate strong potential in response latency and speech comprehension capabilities, showcasing general intelligence across speech understanding tasks. However, the ability to follow speech instructions has not been fully realized due to the lack of datasets and heavily biased training tasks. Leveraging the rich ASR datasets, previous approaches have used Large Language Models~(\textbf{LLMs}) to continue the linguistic information of speech to construct speech instruction datasets. Yet, due to the gap between LLM-generated results and real human responses, the continuation methods further amplify these shortcomings. Given the high costs of collecting and annotating speech instruction datasets by humans, using speech synthesis to construct large-scale speech instruction datasets has become a balanced and robust alternative. Although modern Text-To-Speech~(\textbf{TTS}) models have achieved near-human-level synthesis quality, it is challenging to appropriately convert out-of-distribution text instruction to speech due to the limitations of the training data distribution in TTS models. To address this issue, we propose a query rewriting framework with multi-LLM knowledge fusion, employing multiple agents to annotate and validate the synthesized speech, making it possible to construct high-quality speech instruction datasets without relying on human annotation. Experiments show that this method can transform text instructions into distributions more suitable for TTS models for speech synthesis through zero-shot rewriting, increasing data usability from 72\% to 93\%. It also demonstrates unique advantages in rewriting tasks that require complex knowledge and context-related abilities.

Proto-personas are commonly used during early-stage Product Discovery, such as Lean Inception, to guide product definition and stakeholder alignment. However, the manual creation of proto-personas is often time-consuming, cognitively demanding, and prone to bias. In this paper, we propose and empirically investigate a prompt engineering-based approach to generate proto-personas with the support of Generative AI (GenAI). Our goal is to evaluate the approach in terms of efficiency, effectiveness, user acceptance, and the empathy elicited by the generated personas. We conducted a case study with 19 participants embedded in a real Lean Inception, employing a qualitative and quantitative methods design. The results reveal the approach's efficiency by reducing time and effort and improving the quality and reusability of personas in later discovery phases, such as Minimum Viable Product (MVP) scoping and feature refinement. While acceptance was generally high, especially regarding perceived usefulness and ease of use, participants noted limitations related to generalization and domain specificity. Furthermore, although cognitive empathy was strongly supported, affective and behavioral empathy varied significantly across participants. These results contribute novel empirical evidence on how GenAI can be effectively integrated into software Product Discovery practices, while also identifying key challenges to be addressed in future iterations of such hybrid design processes.

Applications range from technical analysis of a company's fundamental value, wider market sentiment, factor analysis and most tasks involving some form of natural language processing (NLP) [1, 2]. The implications to trading systems will likely be a dramatic increase in the rate and volume of market insights that can be generated to inform decisions. The overall capabilities of LLMs have dramatically increased over the last five years [3]. This has led to an increase in the number of LLMs available, both as proprietary models from frontier labs or as smaller models with open-weights which can be run locally. Given this, the influence of LLMs on trading decisions is expected to be varied and highly model specific. Early work is starting to compare and benchmark these models in tasks specific to financial applications, such as trading decisions, portfolio optimisation, and market analysis [4-10]. As the number of models increases, and their underlying strengths and weaknesses become more apparent, it is expected that different classes of pre-trained models will be more regularly deployed to achieve certain objectives [11, 12]. While these objectives are likely to be significantly linked to NLP-based tasks, such as text summarisation, analysis, and generation, recent LLM architectures give early evidence that more complex tasks can also be automated. These LLMs, such as the'o' series from OpenAI or'R1' from DeepSeek, generate'reasoning' tokens which result in the model performing more in-context analysis of the generated output and has lead to improved performance over a number of key evaluation measures [13, 14].

Millions of biological sample records collected in the last few centuries archived in natural history collections are un-georeferenced. Georeferencing complex locality descriptions associated with these collection samples is a highly labour-intensive task collection agencies struggle with. None of the existing automated methods exploit maps that are an essential tool for georeferencing complex relations. We present preliminary experiments and results of a novel method that exploits multi-modal capabilities of recent Large Multi-Modal Models (LMM). This method enables the model to visually contextualize spatial relations it reads in the locality description. We use a grid-based approach to adapt these auto-regressive models for this task in a zero-shot setting. Our experiments conducted on a small manually annotated dataset show impressive results for our approach ($\sim$1 km Average distance error) compared to uni-modal georeferencing with Large Language Models and existing georeferencing tools. The paper also discusses the findings of the experiments in light of an LMM's ability to comprehend fine-grained maps. Motivated by these results, a practical framework is proposed to integrate this method into a georeferencing workflow.

To ensure equitable access to the benefits of large language models (LLMs), it is essential to evaluate their capabilities across the world's languages. We introduce the AI Language Proficiency Monitor, a comprehensive multilingual benchmark that systematically assesses LLM performance across up to 200 languages, with a particular focus on low-resource languages. Our benchmark aggregates diverse tasks including translation, question answering, math, and reasoning, using datasets such as FLORES+, MMLU, GSM8K, TruthfulQA, and ARC. We provide an open-source, auto-updating leaderboard and dashboard that supports researchers, developers, and policymakers in identifying strengths and gaps in model performance. In addition to ranking models, the platform offers descriptive insights such as a global proficiency map and trends over time. By complementing and extending prior multilingual benchmarks, our work aims to foster transparency, inclusivity, and progress in multilingual AI. The system is available at https://huggingface.co/spaces/fair-forward/evals-for-every-language.

Generating expressive audio performances from music scores requires models to capture both instrument acoustics and human interpretation. Traditional music performance synthesis pipelines follow a two-stage approach, first generating expressive performance MIDI from a score, then synthesising the MIDI into audio. However, the synthesis models often struggle to generalise across diverse MIDI sources, musical styles, and recording environments. To address these challenges, we propose MIDI-VALLE, a neural codec language model adapted from the VALLE framework, which was originally designed for zero-shot personalised text-to-speech (TTS) synthesis. For performance MIDI-to-audio synthesis, we improve the architecture to condition on a reference audio performance and its corresponding MIDI. Unlike previous TTS-based systems that rely on piano rolls, MIDI-VALLE encodes both MIDI and audio as discrete tokens, facilitating a more consistent and robust modelling of piano performances. Furthermore, the model's generalisation ability is enhanced by training on an extensive and diverse piano performance dataset. Evaluation results show that MIDI-VALLE significantly outperforms a state-of-the-art baseline, achieving over 75% lower Frechet Audio Distance on the ATEPP and Maestro datasets. In the listening test, MIDI-VALLE received 202 votes compared to 58 for the baseline, demonstrating improved synthesis quality and generalisation across diverse performance MIDI inputs.

Structured reasoning over natural language inputs remains a core challenge in artificial intelligence, as it requires bridging the gap between unstructured linguistic expressions and formal logical representations. In this paper, we propose a novel \textbf{bi-level framework} that maps language to logic through a two-stage process: high-level task abstraction and low-level logic generation. At the upper level, a large language model (LLM) parses natural language queries into intermediate structured representations specifying the problem type, objectives, decision variables, and symbolic constraints. At the lower level, the LLM uses these representations to generate symbolic workflows or executable reasoning programs for accurate and interpretable decision making. The framework supports modular reasoning, enforces explicit constraints, and generalizes across domains such as mathematical problem solving, question answering, and logical inference. We further optimize the framework with an end-to-end {bi-level} optimization approach that jointly refines both the high-level abstraction and low-level logic generation stages. Experiments on multiple realistic reasoning benchmarks demonstrate that our approach significantly outperforms existing baselines in accuracy, with accuracy gains reaching as high as 40\%. Moreover, the bi-level design enhances transparency and error traceability, offering a promising step toward trustworthy and systematic reasoning with LLMs.

Latent Dirichlet Allocation (LDA) is a prominent generative probabilistic model used for uncovering abstract topics within document collections. In this paper, we explore the effectiveness of augmenting topic models with Large Language Models (LLMs) through integration into two key phases: Initialization and Post-Correction. Since the LDA is highly dependent on the quality of its initialization, we conduct extensive experiments on the LLM-guided topic clustering for initializing the Gibbs sampling algorithm. Interestingly, the experimental results reveal that while the proposed initialization strategy improves the early iterations of LDA, it has no effect on the convergence and yields the worst performance compared to the baselines. The LLM-enabled post-correction, on the other hand, achieved a promising improvement of 5.86% in the coherence evaluation. These results highlight the practical benefits of the LLM-in-the-loop approach and challenge the belief that LLMs are always the superior text mining alternative.

While large language models (LLMs) have advanced procedural planning for embodied AI systems through strong reasoning abilities, the integration of multimodal inputs and counterfactual reasoning remains underexplored. To tackle these challenges, we introduce LLaPa, a vision-language model framework designed for multimodal procedural planning. LLaPa generates executable action sequences from textual task descriptions and visual environmental images using vision-language models (VLMs). Furthermore, we enhance LLaPa with two auxiliary modules to improve procedural planning. The first module, the Task-Environment Reranker (TER), leverages task-oriented segmentation to create a task-sensitive feature space, aligning textual descriptions with visual environments and emphasizing critical regions for procedural execution. The second module, the Counterfactual Activities Retriever (CAR), identifies and emphasizes potential counterfactual conditions, enhancing the model's reasoning capability in counterfactual scenarios. Extensive experiments on ActPlan-1K and ALFRED benchmarks demonstrate that LLaPa generates higher-quality plans with superior LCS and correctness, outperforming advanced models. The code and models are available https://github.com/sunshibo1234/LLaPa.

There are currently two main paradigms for evaluating large language models (LLMs), reference-based evaluation and preference-based evaluation. The first, carried over from the evaluation of machine learning models in general, relies on pre-defined task instances, for which reference task executions are available. The second, best exemplified by the LM-arena, relies on (often self-selected) users bringing their own intents to a site that routes these to several models in parallel, among whose responses the user then selects their most preferred one. The former paradigm hence excels at control over what is tested, while the latter comes with higher ecological validity, testing actual use cases interactively. Recently, a third complementary paradigm has emerged that combines some of the strengths of these approaches, offering control over multi-turn, reference-free, repeatable interactions, while stressing goal-directedness: dialogue game based evaluation. While the utility of this approach has been shown by several projects, its adoption has been held back by the lack of a mature, easily re-usable implementation. In this paper, we present clembench, which has been in continuous development since 2023 and has in its latest release been optimized for ease of general use. We describe how it can be used to benchmark one's own models (using a provided set of benchmark game instances in English), as well as how easily the benchmark itself can be extended with new, tailor-made targeted tests.