Large language models (LLMs) demonstrate unprecedented capabilities and define the state of the art for almost all natural language processing (NLP) tasks and also for essentially all Language Technology (LT) applications. LLMs can only be trained for languages for which a sufficient amount of pre-training data is available, effectively excluding many languages that are typically characterised as under-resourced. However, there is both circumstantial and empirical evidence that multilingual LLMs, which have been trained using data sets that cover multiple languages (including under-resourced ones), do exhibit strong capabilities for some of these under-resourced languages. Eventually, this approach may have the potential to be a technological off-ramp for those under-resourced languages for which "native" LLMs, and LLM-based technologies, cannot be developed due to a lack of training data. This paper, which concentrates on European languages, examines this idea, analyses the current situation in terms of technology support and summarises related work. The article concludes by focusing on the key open questions that need to be answered for the approach to be put into practice in a systematic way.

Warning: This paper contains potentially offensive and harmful text. Large Reasoning Models (LRMs) have recently extended their powerful reasoning capabilities to safety checks--using chain-of-thought reasoning to decide whether a request should be answered. While this new approach offers a promising route for balancing model utility and safety, its robustness remains underexplored. To address this gap, we introduce Malicious-Educator, a benchmark that disguises extremely dangerous or malicious requests beneath seemingly legitimate educational prompts. Our experiments reveal severe security flaws in popular commercial-grade LRMs, including OpenAI o1/o3, DeepSeek-R1, and Gemini 2.0 Flash Thinking. For instance, although OpenAI's o1 model initially maintains a high refusal rate of about 98%, subsequent model updates significantly compromise its safety; and attackers can easily extract criminal strategies from DeepSeek-R1 and Gemini 2.0 Flash Thinking without any additional tricks. To further highlight these vulnerabilities, we propose Hijacking Chain-of-Thought (H-CoT), a universal and transferable attack method that leverages the model's own displayed intermediate reasoning to jailbreak its safety reasoning mechanism. Under H-CoT, refusal rates sharply decline--dropping from 98% to below 2%--and, in some instances, even transform initially cautious tones into ones that are willing to provide harmful content. We hope these findings underscore the urgent need for more robust safety mechanisms to preserve the benefits of advanced reasoning capabilities without compromising ethical standards.

The breakthrough of generative large language models (LLMs) that can solve different tasks through chat interaction has led to a significant increase in the use of general benchmarks to assess the quality or performance of these models beyond individual applications. There is also a need for better methods to evaluate and also to compare models due to the ever increasing number of new models published. However, most of the established benchmarks revolve around the English language. This paper analyses the benefits and limitations of current evaluation datasets, focusing on multilingual European benchmarks. We analyse seven multilingual benchmarks and identify four major challenges. Furthermore, we discuss potential solutions to enhance translation quality and mitigate cultural biases, including human-in-the-loop verification and iterative translation ranking. Our analysis highlights the need for culturally aware and rigorously validated benchmarks to assess the reasoning and question-answering capabilities of multilingual LLMs accurately.

In LLM evaluations, reasoning is often distinguished from recall/memorization by performing numerical variations to math-oriented questions. Here we introduce a general variation method for multiple-choice questions that completely dissociates the correct answer from previously seen tokens or concepts, requiring LLMs to understand and reason (rather than memorizing) in order to answer correctly. Using this method, we evaluate state-of-the-art proprietary and open-source LLMs on two datasets available in English and Spanish: the public MMLU benchmark and the private UNED-Access 2024 dataset. Results show that all models experience remarkable accuracy drops under our proposed variation, with an average loss of 57% on MMLU and 50% on UNED-Access 2024, ranging from 10% to 93% across models. Notably, the most accurate model in our experimentation (OpenAI-o3-mini) is not the most robust (DeepSeek-R1-70B), suggesting that the best models in standard evaluations may not be the ones with better reasoning capabilities. Also, we see larger accuracy drops in public (vs private) datasets and questions posed in their original language (vs a manual translation), which are signs of contamination and also point to a relevant role of recall/memorization in current LLMs' answers.

Existing end-to-end speech large language models (LLMs) usually rely on large-scale annotated data for training, while data-efficient training has not been discussed in depth. We focus on two fundamental problems between speech and text: the representation space gap and sequence length inconsistency. We propose Soundwave, which utilizes an efficient training strategy and a novel architecture to address these issues. Results show that Soundwave outperforms the advanced Qwen2-Audio in speech translation and AIR-Bench speech tasks, using only one-fiftieth of the training data. Further analysis shows that Soundwave still retains its intelligence during conversation. The project is available at https://github.com/FreedomIntelligence/Soundwave.

We introduce Fraud-R1, a benchmark designed to evaluate LLMs' ability to defend against internet fraud and phishing in dynamic, real-world scenarios. Fraud-R1 comprises 8,564 fraud cases sourced from phishing scams, fake job postings, social media, and news, categorized into 5 major fraud types. Unlike previous benchmarks, Fraud-R1 introduces a multi-round evaluation pipeline to assess LLMs' resistance to fraud at different stages, including credibility building, urgency creation, and emotional manipulation. Furthermore, we evaluate 15 LLMs under two settings: 1. Helpful-Assistant, where the LLM provides general decision-making assistance, and 2. Role-play, where the model assumes a specific persona, widely used in real-world agent-based interactions. Our evaluation reveals the significant challenges in defending against fraud and phishing inducement, especially in role-play settings and fake job postings. Additionally, we observe a substantial performance gap between Chinese and English, underscoring the need for improved multilingual fraud detection capabilities.

Generating SQLs from user queries is a long-standing challenge, where the accuracy of initial schema linking significantly impacts subsequent SQL generation performance. However, current schema linking models still struggle with missing relevant schema elements or an excess of redundant ones. A crucial reason for this is that commonly used metrics, recall and precision, fail to capture relevant element missing and thus cannot reflect actual schema linking performance. Motivated by this, we propose an enhanced schema linking metric by introducing a restricted missing indicator. Accordingly, we introduce Knapsack optimization-based Schema Linking Agent (KaSLA), a plug-in schema linking agent designed to prevent the missing of relevant schema elements while minimizing the inclusion of redundant ones. KaSLA employs a hierarchical linking strategy that first identifies the optimal table linking and subsequently links columns within the selected table to reduce linking candidate space. In each linking process, it utilize a knapsack optimization approach to link potentially relevant elements while accounting for a limited tolerance of potential redundant ones.With this optimization, KaSLA-1.6B achieves superior schema linking results compared to large-scale LLMs, including deepseek-v3 with state-of-the-art (SOTA) schema linking method. Extensive experiments on Spider and BIRD benchmarks verify that KaSLA can significantly improve the SQL generation performance of SOTA text-to-SQL models by substituting their schema linking processes.

Large Language Models (LLMs) fine-tuning technologies have achieved remarkable results. However, traditional LLM fine-tuning approaches face significant challenges: they require large Floating Point (FP) computation, raising privacy concerns when handling sensitive data, and are impractical for resource-constrained edge devices. While Parameter-Efficient Fine-Tuning (PEFT) techniques reduce trainable parameters, their reliance on floating-point arithmetic creates fundamental incompatibilities with edge hardware. In this work, we introduce a novel framework for on-device LLM fine-tuning that eliminates the need for floating-point operations in both inference and training, named GSQ-Tuning. At its core is the Group-Shared Exponents Integer format, which efficiently represents model parameters in integer format using shared exponents among parameter groups. When combined with LoRA-like adapters, this enables fully integer-based fine-tuning that is both memory and compute efficient. We demonstrate that our approach achieves accuracy comparable to FP16-based fine-tuning while significantly reducing memory usage (50%). Moreover, compared to FP8, our method can reduce 5x power consumption and 11x chip area with same performance, making large-scale model adaptation feasible on edge devices.

Query-driven recommendation with unknown items poses a challenge for users to understand why certain items are appropriate for their needs. Query-driven Contrastive Summarization (QCS) is a methodology designed to address this issue by leveraging language-based item descriptions to clarify contrasts between them. However, existing state-of-the-art contrastive summarization methods such as STRUM-LLM fall short of this goal. To overcome these limitations, we introduce Q-STRUM Debate, a novel extension of STRUM-LLM that employs debate-style prompting to generate focused and contrastive summarizations of item aspects relevant to a query. Leveraging modern large language models (LLMs) as powerful tools for generating debates, Q-STRUM Debate provides enhanced contrastive summaries. Experiments across three datasets demonstrate that Q-STRUM Debate yields significant performance improvements over existing methods on key contrastive summarization criteria, thus introducing a novel and performant debate prompting methodology for QCS.

This paper investigates whether Large Language Models (LLMs), fine-tuned on synthetic but domain-representative data, can perform the twofold task of (i) slot and intent detection and (ii) natural language response generation for a smart home assistant, while running solely on resource-limited, CPU-only edge hardware. We fine-tune LLMs to produce both JSON action calls and text responses. Our experiments show that 16-bit and 8-bit quantized variants preserve high accuracy on slot and intent detection and maintain strong semantic coherence in generated text, while the 4-bit model, while retaining generative fluency, suffers a noticeable drop in device-service classification accuracy. Further evaluations on noisy human (non-synthetic) prompts and out-of-domain intents confirm the models' generalization ability, obtaining around 80--86\% accuracy. While the average inference time is 5--6 seconds per query -- acceptable for one-shot commands but suboptimal for multi-turn dialogue -- our results affirm that an on-device LLM can effectively unify command interpretation and flexible response generation for home automation without relying on specialized hardware.