Large Language Models (LLMs) have taken the research field of Natural Language Processing by storm. Researchers are not only investigating their capabilities and possible applications, but also their weaknesses and how they may be exploited.This has resulted in various attacks and “jailbreaking” approaches that have gained large interest within the community.The vulnerability of LLMs to certain types of input may pose major risks regarding the real-world usage of LLMs in productive operations.We therefore investigate the relationship between a LLM’s uncertainty and its vulnerability to jailbreaking attacks.To this end, we focus on a probabilistic point of view of uncertainty and employ a state-of-the art open-source LLM.We investigate an attack that is based on linguistic obfuscation.Our results indicate that the model is subject to a higher level of uncertainty when confronted with manipulated prompts that aim to evade security mechanisms.This study lays the foundation for future research into the link between model uncertainty and its vulnerability to jailbreaks.

High-quality training data for Task-Oriented Dialog (TOD) systems is costly to come by if no corpora are available. One method to extend available data is data augmentation. Yet, the research into and adaptation of data augmentation techniques for TOD systems is limited in comparison with other data modalities. We propose a novel, causally-flavored data augmentation technique called Counterfactual Dialog Mixing (CDM) that generates realistic synthetic dialogs via counterfactuals to increase the amount of training data. We demonstrate the method on a benchmark dataset and show that a model trained to classify the counterfactuals from the original data fails to do so, which strengthens the claim of creating realistic synthetic dialogs. To evaluate the effectiveness of CDM, we train a current architecture on a benchmark dataset and compare the performance with and without CDM. By doing so, we achieve state-of-the-art on some metrics. We further investigate the external generalizability and a lower resource setting. To evaluate the models, we adopted an interactive evaluation scheme.

Modern dialog systems rely on Deep Learning to train transformer-based model architectures. These notoriously rely on large amounts of training data. However, the collection of conversational data is often a tedious and costly process. This is especially true for Task-Oriented Dialogs, where the system ought to help the user achieve specific tasks, such as making reservations. We investigate a controlled strategy for dialog synthesis. Our method generates utterances based on dialog annotations in a sequence-to-sequence manner. Besides exploring the viability of the approach itself, we also explore the effect of constrained beam search on the generation capabilities. Moreover, we analyze the effectiveness of the proposed method as a data augmentation by studying the impact the synthetic dialogs have on training dialog systems. We perform the experiments in multiple settings, simulating various amounts of ground-truth data. Our work shows that a controlled generation approach is a viable method to synthesize Task-Oriented Dialogs, that can in turn be used to train dialog systems. We were able to improve this process by utilizing constrained beam search.

This study explores the capacity of large language models (LLMs) to efficiently generate summaries of informal educational content tailored for platforms like TikTok. It also investigates how both humans and LLMs assess the quality of these summaries, based on a series of experiments, exploring the potential replacement of human evaluation with LLMs. Furthermore, the study delves into how experienced content creators perceive the utility of automatic summaries for TikTok videos. We employ strategic prompt selection techniques to guide LLMs in producing engaging summaries based on the characteristics of viral TikTok content, including hashtags, captivating hooks, storytelling, and user engagement. The study leverages OpenAI’s GPT-4 model to generate TikTok content summaries, aiming to align them with the essential features identified. By employing this model and incorporating human evaluation and expert assessment, this research endeavors to shed light on the intricate dynamics of modern content creation, where AI and human ingenuity converge. Ultimately, it seeks to enhance strategies for disseminating and evaluating educational information effectively in the realm of social media.