GenAI in Action: Structured, Verified and Explainable Text Generation

Dear all,

We are back! And as usual we will be looking at a hot topic in data science (and beyond) ... generative AI! And why not ask one of the experts to talk us through the opportunities, problems and solutions of GenAI? Well, that's exactly our thinking. We are very happy to welcome Dietrich Trautmann to be our guest this time.

Dietrich will join us remotely but we have also booked a room at uni (H25 in the Vielberth building), and the plan is to meet up there with a cold drink (or two) and have a proper Meetup. Anyone who wants to join remotely is welcome to do so (we will send out a Zoom link closer to the date as always). Let's see how that works out and whether this might be a model we could try out again in future.

We will also team up with the Text Analytics London Meetup. We expect many familiar faces joining us from there ...

Are you ready? Then we hope to see you next week,
Udo, Dyaa and Tony

Speaker:
Dietrich Trautmann

Title:
Structured, Verified and Explainable Text Generation

Abstract:
Large language models (LLMs) have revolutionized text generation, but their outputs often lack structure, verification, and explainability. This talk explores practical approaches to address these challenges in applied settings. We'll examine techniques for guiding LLMs to produce structured text that adheres to specific formats and constraints. We'll also discuss methods for verifying the accuracy and reliability of generated content, crucial for applications where precision is essential. Finally, we'll investigate methods that enhance the explainability of LLM outputs, allowing users to understand and trust the model's decision-making process. By integrating these three components - structure, verification, and explainability - we aim to showcase how LLM-driven text generation can be effectively and responsibly deployed in diverse real-world scenarios.

Throughout the talk, we'll focus on practical implementations and case studies, highlighting the potential and limitations of these approaches across various industries.

Short Bio:
Dietrich Trautmann is a seasoned expert in the field of natural language processing (NLP). He held several positions as Data Scientist and Applied Research Scientist, working mostly on unstructured textual data. With an academic background in computer science from the Technical University of Munich (TUM) and the Ludwig-Maximilians University of Munich (LMU), he also contributed with publications to several conferences. Currently, Dietrich is dedicated to advancing the application of large language models (LLMs), transforming them from proof-of-concept stages to dependable, high-impact solutions across a variety of domains and use cases.