India is a diverse society with unique challenges in developing AI systems, including linguistic diversity, oral traditions, data accessibility, and scalability. Existing foundation models are primarily trained on English, limiting their effectiveness for India's population. Indic languages comprise only 1 percent of Common Crawl corpora despite India representing 18 percent of the global population, leading to linguistic biases. Thousands of regional languages, dialects, and code mixing create additional representation challenges due to sparse training data. We introduce Krutrim LLM, a 2 trillion token multilingual model designed for India's linguistic landscape. It incorporates the largest known Indic dataset, mitigating data scarcity and ensuring balanced performance across dialects. Krutrim outperforms or matches state-of-the-art models on Indic benchmarks while maintaining competitive English performance. Despite being significantly smaller in training flops, Krutrim LLM matches or exceeds models like LLAMA-2 on 10 out of 16 tasks, with an average score of 0.57 versus 0.55. This evidences Krutrim's flexible multilingual fluency across diverse linguistic contexts. Krutrim is integrated with real-time search to improve factual accuracy in conversational AI applications. This enhances accessibility for over 1 billion users worldwide. Through intentional design choices addressing data imbalances, Krutrim LLM signifies meaningful progress in building ethical, globally representative AI models.

This paper presents ViSP, a high-quality Vietnamese dataset for sentence paraphrasing, consisting of 1.2M original-paraphrase pairs collected from various domains. The dataset was constructed using a hybrid approach that combines automatic paraphrase generation with manual evaluation to ensure high quality. We conducted experiments using methods such as back-translation, EDA, and baseline models like BART and T5, as well as large language models (LLMs), including GPT-4o, Gemini-1.5, Aya, Qwen-2.5, and Meta-Llama-3.1 variants. To the best of our knowledge, this is the first large-scale study on Vietnamese paraphrasing. We hope that our dataset and findings will serve as a valuable foundation for future research and applications in Vietnamese paraphrase tasks.

Automatically summarizing large text collections is a valuable tool for document research, with applications in journalism, academic research, legal work, and many other fields. In this work, we contrast two classes of systems for large-scale multi-document summarization (MDS): compression and full-text. Compression-based methods use a multi-stage pipeline and often lead to lossy summaries. Full-text methods promise a lossless summary by relying on recent advances in long-context reasoning. To understand their utility on large-scale MDS, we evaluated them on three datasets, each containing approximately one hundred documents per summary. Our experiments cover a diverse set of long-context transformers (Llama-3.1, Command-R, Jamba-1.5-Mini) and compression methods (retrieval-augmented, hierarchical, incremental). Overall, we find that full-text and retrieval methods perform the best in most settings. With further analysis into the salient information retention patterns, we show that compression-based methods show strong promise at intermediate stages, even outperforming full-context. However, they suffer information loss due to their multi-stage pipeline and lack of global context. Our results highlight the need to develop hybrid approaches that combine compression and full-text approaches for optimal performance on large-scale multi-document summarization.

Theory of Mind (ToM), the ability to attribute mental states to others and predict their behaviour, is fundamental to social intelligence. In this paper, we survey studies evaluating behavioural and representational ToM in Large Language Models (LLMs), identify important safety risks from advanced LLM ToM capabilities, and suggest several research directions for effective evaluation and mitigation of these risks.

The utilisation of foundation models as smartphone assistants, termed app agents, is a critical research challenge. These agents aim to execute human instructions on smartphones by interpreting textual instructions and performing actions via the device's interface. While promising, current approaches face significant limitations. Methods that use large proprietary models, such as GPT-4o, are computationally expensive, while those that use smaller fine-tuned models often lack adaptability to out-of-distribution tasks. In this work, we introduce AppVLM, a lightweight Vision-Language Model (VLM). First, we fine-tune it offline on the AndroidControl dataset. Then, we refine its policy by collecting data from the AndroidWorld environment and performing further training iterations. Our results indicate that AppVLM achieves the highest action prediction accuracy in offline evaluation on the AndroidControl dataset, compared to all evaluated baselines, and matches GPT-4o in online task completion success rate in the AndroidWorld environment, while being up to ten times faster. This makes AppVLM a practical and efficient solution for real-world deployment.

Tokenization constitutes an essential preprocessing step within Natural Language Processing (NLP), exerting a direct impact on large language models' (LLMs) capacity to capture syntactic, morphosyntactic, and semantic details. This paper introduces a novel framework for the systematic evaluation of tokenization strategies, with a particular focus on mitigating the challenges associated with morphologically-rich and low-resource languages. Using a Turkish dataset of 6,200 multiple-choice questions derived from the Massive Multitask Language Understanding (MMLU) benchmark, the framework evaluates tokenizers across five key metrics: vocabulary size, token count, processing time, language-specific token percentages (%TR), and token purity. These metrics, proposed in this study, offer a structured approach to assessing how effectively tokenizers preserve linguistic structures. While %TR measures the proportion of valid words generated in the target language, %Pure evaluates the alignment of tokens with meaningful linguistic units, such as roots and valid morphemes, ensuring minimal semantic fragmentation. The findings reveal that language-specific token percentages, introduced as a critical evaluation metric, exhibit a stronger correlation with downstream performance (e.g., MMLU scores) compared to token purity, emphasizing their importance in enhancing model accuracy and robustness. Furthermore, the analysis demonstrates that larger model parameters do not necessarily yield superior tokenization quality or improved results, highlighting the need for tailored tokenization strategies that prioritize linguistic alignment over mere computational scaling. By addressing both computational efficiency and linguistic fidelity, this framework establishes a new standard for developing robust tokenization methods optimized for morphologically complex and low-resource languages. Future work will focus on advancing morphological analysis techniques, exploring domain-specific customizations, and conducting cross-linguistic evaluations to further refine tokenization practices across diverse linguistic contexts.

Large Language Models (LLMs) benefit from training on ever larger amounts of textual data, but as a result, they increasingly incur the risk of leaking private information. The ability to selectively remove knowledge from LLMs is, therefore, a highly desirable capability. In this paper, we propose LUNAR, a novel unlearning methodology grounded in the Linear Representation Hypothesis. LUNAR operates by redirecting the representations of unlearned data to regions that trigger the model's inherent ability to express its inability to answer. LUNAR achieves state-of-the-art unlearning performance while significantly enhancing the controllability of the unlearned model during inference. Specifically, LUNAR achieves between 2.9x to 11.7x improvements on combined "unlearning efficacy" and "model utility" score ("Deviation Score") on the PISTOL dataset across various base models. We also demonstrate, through quantitative analysis and qualitative examples, LUNAR's superior controllability in generating coherent and contextually aware responses, mitigating undesired side effects of existing methods. Moreover, we demonstrate that LUNAR is robust against white-box adversarial attacks and versatile in handling real-world scenarios, such as processing sequential unlearning requests.

The imitation of voice, targeted on specific speech attributes such as timbre and speaking style, is crucial in speech generation. However, existing methods rely heavily on annotated data, and struggle with effectively disentangling timbre and style, leading to challenges in achieving controllable generation, especially in zero-shot scenarios. To address these issues, we propose Vevo, a versatile zeroshot voice imitation framework with controllable timbre and style. Vevo operates in two core stages: (1) Content-Style Modeling: Given either text or speech's content tokens as input, we utilize an autoregressive transformer to generate the content-style tokens, which is prompted by a style reference; (2) Acoustic Modeling: Given the content-style tokens as input, we employ a flow-matching transformer to produce acoustic representations, which is prompted by a timbre reference. To obtain the content and content-style tokens of speech, we design a fully self-supervised approach that progressively decouples the timbre, style, and linguistic content of speech. Specifically, we adopt VQ-VAE [1] as the tokenizer for the continuous hidden features of HuBERT [2]. We treat the vocabulary size of the VQ-VAE codebook as the information bottleneck, and adjust it carefully to obtain the disentangled speech representations. Solely self-supervised trained on 60K hours of audiobook speech data, without any fine-tuning on style-specific corpora, Vevo matches or surpasses existing methods in accent and emotion conversion tasks. Additionally, Vevo's effectiveness in zero-shot voice conversion and text-to-speech tasks further demonstrates its strong generalization and versatility. The imitation of voice has long been an important issue in the field of speech generation. This includes the imitation of speaker identity [3, 4], the imitation of speaking style such as accent [5, 6] or emotion [7], and a broader concept of voice cloning such as in zero-shot text-to-speech (TTS) task [8]. These techniques have a wide range of applications, including spoken language learning [5, 6, 9], voice anonymization [10], voice assistants [11, 12], and video dubbing [11, 12, 13]. To achieve targeted and controllable imitation over various speech attributes, many studies focuses on factorizing speech into multiple sub-spaces [14, 15, 16, 17]. In this work, we follow this idea and decompose speech into three key attributes: linguistic content (what to speak), style (how to speak), and timbre (who speaks).

Mobile emailing demands efficiency in diverse situations, which motivates the use of AI. However, generated text does not always reflect how people want to respond. This challenges users with AI involvement tradeoffs not yet considered in email UIs. We address this with a new UI concept called Content-Driven Local Response (CDLR), inspired by microtasking. This allows users to insert responses into the email by selecting sentences, which additionally serves to guide AI suggestions. The concept supports combining AI for local suggestions and message-level improvements. Our user study (N=126) compared CDLR with manual typing and full reply generation. We found that CDLR supports flexible workflows with varying degrees of AI involvement, while retaining the benefits of reduced typing and errors. This work contributes a new approach to integrating AI capabilities: By redesigning the UI for workflows with and without AI, we can empower users to dynamically adjust AI involvement.

In this paper, we introduce a learning analytics framework to analyze the in-context learning (ICL) behavior of large language models (LLMs) through the lens of the Zone of Proximal Development (ZPD), an established theory in educational psychology. ZPD delineates the space between what a learner is capable of doing unsupported and what the learner cannot do even with support. We adapt this concept to ICL, measuring the ZPD of LLMs based on model performance on individual examples with and without ICL. Furthermore, we propose an item response theory (IRT) model to predict the distribution of zones for LLMs. Our findings reveal a series of intricate and multifaceted behaviors of ICL, providing new insights into understanding and leveraging this technique. Finally, we demonstrate how our framework can enhance LLM in both inference and fine-tuning scenarios: (1) By predicting a model's zone of proximal development, we selectively apply ICL to queries that are most likely to benefit from demonstrations, achieving a better balance between inference cost and performance; (2) We propose a human-like curriculum for fine-tuning, which prioritizes examples within the model's ZPD. The curriculum results in improved performance, and we explain its effectiveness through an analysis of the training dynamics of LLMs.