Exploring LLMs: Building, Fine-Tuning and Detection

Exploring LLMs: Building, Fine-Tuning and Detection
Our 43th DataTalks meetup will be hosted by DoubleVerify, and we'll discuss about LLMs fundamentals, including their architecture, fine-tuning techniques for specialized tasks, and methods for detecting AI-generated content. The session will cover both theoretical understanding and practical applications of LLM technology.

When: Tuesday, January 28, 2025, from 17:30 to 19:30 IST
Where: Alon Tower #2, 94 Yigal Alon St., Floor 27, Tel Aviv

Agenda

• 17:30: Welcome: Networking, Food and Drinks
• 18:00: Fine-Tuning Large Language Models: Techniques and Approaches for Task-Specific Adaptation
  Mike Erlihson, Head of AI, Cyber Stealth
  In the talk, Mike will explain why and how fine-tuning is essential for Large Language Models (LLMs). He’ll cover key techniques like Supervised Fine-Tuning (SFT) and Reinforcement Learning with Human Feedback (RLHF), as well as efficient methods and alternatives like in-context learning (ICL) and retrieval-augmented generation (RAG). By the end, you'll understand how to adapt LLMs for specific tasks and improve their performance in specialized applications.
• 18:45: Detecting AI-Generated Texts
  James Eiger, Sr. Data Scientist, DoubleVerify
  While LLMs vary in training methods, data, pre-prompts, and limitations, they mostly operate under a similar framework. As such, by leveraging how they operate, we are able to detect texts generated by a wide variety of LLMs with a high degree of accuracy. James will explain how texts are generated and how we can build features around this to detect when a text has been generated by AI.

See you there!