Arghodeep Nandi

Author directory

2026

Prompting has transformed large language models into general-purpose decision systems, where token-level probabilities are often treated as indicators of model preference and correctness. However, directly interpreting these probabilities is challenging due to calibration disparities, tokenizer variability, and inconsistent output distributions across models, limiting their statistical comparability. This raises a fundamental question: *when can a model be said to be confident in its decision or preference?* Addressing this question remains an open and important area of research. We introduce **PLAUSIEVE**, a novel distribution-based statistical framework that estimates model preference through a forward pass without token generation. **PLAUSIEVE** provides a calibration-robust and model-independent mechanism for quantifying preference without relying on fixed thresholds or individual tokens. We apply **PLAUSIEVE** in estimating model belief, across two prompt settings (direct and epistemic), two confidence levels, and five models from four families to study prompt–architecture interactions. Our analysis shows significant improvements under epistemic prompting, with nearly 100% gains in certain model combinations. Notably, **Llama-3.2-3B** outperforms the larger **Gemma-7B** under epistemic prompting in terms of both decisiveness and veracity, demonstrating that model architecture plays a more significant role than scale in shaping belief formation.

2025

Misinformation remains one of the most significant issues in the digital age. While automated fact-checking has emerged as a viable solution, most current systems are limited to evaluating factual accuracy. However, the detrimental effect of misinformation transcends simple falsehoods; it takes advantage of how individuals perceive, interpret, and emotionally react to information. This underscores the need to move beyond factuality and adopt more human-centered detection frameworks. In this survey, we explore the evolving interplay between traditional fact-checking approaches and psychological concepts such as cognitive biases, social dynamics, and emotional responses. By analyzing state-of-the-art misinformation detection systems through the lens of human psychology and behavior, we reveal critical limitations of current methods and identify opportunities for improvement. Additionally, we outline future research directions aimed at creating more robust and adaptive frameworks, such as neuro-behavioural models that integrate technological factors with the complexities of human cognition and social influence. These approaches offer promising pathways to more effectively detect and mitigate the societal harms of misinformation.