Daniel Fitzpatrick

The Sarcastic Side of AI: Metacognitive Prompting

In this session, we will introduce Pragmatic Metacognitive Prompting (PMP) as a novel approach to improve sarcasm detection in Large Language Models (LLMs). PMP leverages pragmatic reasoning and metacognitive strategies to enhance the interpretation of implied meanings and contextual cues, achieving improved performance on sarcasm detection benchmarks. This entertaining, interactive talk explains the integration of pragmatic theories into LLM prompting, offering a new direction for sentiment analysis research. It demonstrates how linguistic principles – such as implicature – influence human reasoning in complex contexts, enhancing sarcasm detection capabilities. The target audience includes AI researchers, NLP practitioners, and data scientists interested in sentiment analysis and language model enhancement. Key takeaways include understanding the application of pragmatic theories in Natural Language Processing, insights into advanced sarcasm detection methods, and practical knowledge of implementing metacognitive prompting.

Moving from lithography into the digital world, Dan started designing and programming user interfaces back in 1999. Today, along with Dr. Tillmann Pross, he leads Frank Reply GmbH – the Reply expert team for Conversational AI. This still doesn't stop him from programming voice interfaces and researching related technologies! Furthermore, as Practice Lead, he heads a worldwide think tank for Voice Machine Interfaces at Reply AG.