Arjun Mukherjee

From Words to Wit: Humor-Aware Information Retrieval in English and Hinglish

IReL_IIT(BHU) at HAHA 2026: A Fine-Tuning Pipeline for Satire and LLM-Generated Humor Detection in Spanish News Headlines

Correctness Is a Direction: Geometric Answer Selection in Language Models

Answer correctness is encoded as a recoverable geometric direction in the hidden states of language models. We show that the mean displacement from incorrect to correct answer representations, computed at approximately 70\% of model depth from fifty labeled examples with no parameter updates, yields a scoring direction...

Decodable But Not Detachable: Training Data Granularity Determines Parametric Modularity in Large Language Models

This work applies a uniform causal methodology across two domain granularities, three model families, and eight domains to identify domain-specific parametric shells: concentrated, causally necessary neuron populations whose removal selectively degrades a target domain while sparing others.