Despite efforts to expand the knowledge of large language models (LLMs), knowledge gaps—missing or outdated information in LLMs—might always persist given the evolving nature of knowledge. In this work, we study approaches to identify LLM knowledge gaps and abstain from answering questions when knowledge gaps are present. We first adapt existing approaches to model calibration or adaptation through fine-tuning/prompting and analyze their ability to abstain from generating low-confidence outputs. Motivated by their failures in self-reflection and over-reliance on held-out sets, we propose two novel approaches that are based on model collaboration, i.e., LLMs probing other LLMs for knowledge gaps, either cooperatively or competitively. Extensive experiments with three LLMs on four QA tasks featuring diverse knowledge domains demonstrate that both cooperative and competitive approaches to unveiling LLM knowledge gaps achieve up to 19.3% improvements on abstain accuracy against the strongest baseline. Further analysis reveals that our abstention methods pinpoint failure cases in retrieval augmentation and knowledge gaps in multi-hop reasoning.

This paper describes our system used in the SemEval-2024 Task 4 Multilingual Detection of Persuasion Techniques in Memes. Our team proposes a detection system that employs a Teacher Student Fusion framework. Initially, a Large Language Model serves as the teacher, engaging in abductive reasoning on multimodal inputs to generate background knowledge on persuasion techniques, assisting in the training of a smaller downstream model. The student model adopts CLIP as an encoder for text and image features, and we incorporate an attention mechanism for modality alignment. Ultimately, our proposed system achieves a Macro-F1 score of 0.8103, ranking 1st out of 20 on the leaderboard of Subtask 2b in English. In Bulgarian, Macedonian and Arabic, our detection capabilities are ranked 1/15, 3/15 and 14/15.