Large language models (LLMs) encode problem difficulty as an internal signal that can be linearly decoded from their residuals. Given their multilingual capabilities, we investigate whether this meta-cognitive signal is language-agnostic and how it is organized across the model’s layers by training linear probes on the AMC subset of the Easy2Hard benchmark, translated into 21 languages. We found that difficulty-related signals emerge at two distinct stages of the model internals, corresponding to shallow (early-layers) and deep (later-layers) internal representations, that exhibit functionally different behaviors. Probes trained on deep representations achieve high accuracy when evaluated on the same language but exhibit weaker cross-lingual transfer. In contrast, probes trained on shallow representations generalize better across languages, despite achieving lower within-language performance. This closely aligns with existing findings in LLM interpretability, showing that models tend to operate in an abstract conceptual space before producing language-specific outputs. Our results suggest that this two-stage organizational principle extends beyond simple semantic processing to meta-cognitive properties such as problem difficulty, highlighting an internal control signal that is not tied to surface meaning.

We present a novel approach for enhancing diversity and control in data annotation tasks by personalizing large language models (LLMs). We investigate the impact of injecting diverse persona descriptions into LLM prompts across two studies, exploring whether personas increase annotation diversity and whether the impacts of individual personas on the resulting annotations are consistent and controllable. Our results indicate that persona-prompted LLMs generate more diverse annotations than LLMs prompted without personas, and that the effects of personas on LLM annotations align with subjective differences in human annotations. These effects are both controllable and repeatable, making our approach a valuable tool for enhancing data annotation in subjective NLP tasks such as toxicity detection.

This study introduces KriRAG, a novel Retrieval-Augmented Generation (RAG) architecture designed to assist criminal investigators in analyzing information and overcoming the challenge of information overload. KriRAG structures and summarizes extensive document collections based on existing investigative queries, providing relevant document references and detailed answers for each query. Working with unstructured data from two homicide case files comprising approximately 3,700 documents and 13,000 pages, a comprehensive evaluation methodology is established, incorporating semantic retrieval, scoring, reasoning, and query response accuracy. The system’s outputs are evaluated against queries and answers provided by criminal investigators, demonstrating promising performance with 97.5% accuracy in relevance assessment and 77.5% accuracy for query responses. These findings provide a rigorous foundation for other query-oriented and open-ended retrieval applications. KriRAG is designed to run offline on limited hardware, ensuring sensitive data handling and on-device availability.

Monitoring vaccine behaviour through social media can guide health policy. We present a new dataset of 9471 tweets posted in Australia from 2020 to 2022, annotated with sentiment toward vaccines and also 5C, the five types of behaviour toward vaccines, a scheme commonly used in health psychology literature. We benchmark our dataset using BERT and Gradient Boosting Machine and show that jointly training both sentiment and 5C tasks (F1=48) outperforms individual training (F1=39) in this highly imbalanced data. Our sentiment analysis indicates close correlation between the sentiments and prominent events during the pandemic. We hope that our dataset and benchmark models will inform further work in online monitoring of vaccine behaviour. The dataset and benchmark methods are accessible online.