Motivation

As a public-service media organization, ZDF is responsible for informing, educating, advising, and entertaining the public in Germany. To remain competitive amid rapid technological advancements, we need advanced language models capable of tasks like text summarization, transcription, content recommendations, fact-checking, and bias detection. Our goal is to maintain control and transparency over these models, ensuring independence from external vendors while optimizing resource efficiency and accuracy. In collaboration with TUM-DI-LAB, we are seeking talented students to develop a small, efficient language model (SLM) prototype for news article summarization and compare its performance to that of customized large language models (LLMs).
 

Goals

This project aims to develop a prototype small language model for text summarization that can be scaled for broader applications. The main objective is to evaluate its effectiveness compared to customized, fine-tuned models in summarizing ZDF’s news articles. We are looking for motivated students to drive these innovations in public-service media.

As part of the project team, you will collaborate with our development and data science experts, working with real data. Key tasks include:

1. Data Familiarization: Gain an understanding of our dataset, infrastructure, and news articles.
2. Customizing LLMs: Explore pre-trained models to identify patterns and insights for effective content summarization.
3. Building the SLM Prototype: Design and implement a small language model, assessing its accuracy and reliability against customized LLMs.
4. Recommendation Report: Evaluate and compare the performance of the LLMs and SLM prototype across different content types (e.g., short news, long-form articles) based on readability, factual accuracy, compression ratio, and alignment with editorial standards.

The vision is to create a tool for generating high-quality text summaries that can also serve as a prototype for future text generation applications.

Requirements 

To succeed in this project, we recommend students bring the following skills:

• Problem-Solving & Critical Thinking: Ability to deconstruct complex problems and develop creative solutions.
• AI & Machine Learning: Familiarity with frameworks like TensorFlow, PyTorch, or scikit-learn, with strong AI/ML knowledge and Python proficiency for model development.
• Curiosity & Enthusiasm: A passion for learning, innovation, and contributing to public-service media.

Join us to shape the future of public-service media and contribute to a more informed world. Apply now!

Apply to this project here