Kara Büyünün Ardında | Burak Sina Akbudak | TEDxIzmir Fen Lisesi Youth

Quick Overview

The presentation by Burak Sina Akbudak at TEDxIzmir Fen Lisesi Youth focuses on demystifying Artificial Intelligence, particularly Large Language Models (LLMs), by outlining the key steps in their creation—Data Sets, Model Architecture, Training, Validation, Extra Precautions, and Inference—while emphasizing the accelerating pace of development and the need for humans to adapt to AI as a powerful new tool, rather than something entirely foreign.

Key Points: The speaker, Burak Sina Akbudak (aka cos beta), is currently studying Computer Engineering and has experience in NLP projects and competitions like ZOOAA Turkey. AI development is structured in clear phases: Pre-training (using massive data sets like Common Crawl to teach language basics), Fine-Tuning (specializing the model for specific domains via labeling), and RLHF (Reinforcement Learning from Human Feedback) to align outputs with human preferences. The scale of data processing is immense; Common Crawl adds 250 TB of data monthly (2 billion web pages) since 2015, accumulating petabytes of data. Model complexity is demonstrated by the rapid growth in parameters, moving from the 13-year-old Human model (<100 Million tokens) to DeepSeek-V3 (2024) processing 14.8 Trillion tokens. Practical AI applications span Natural Language Processing (code generation, Q&A), Computer Vision (object tracking, pose estimation, image generation via Diffusion/GANs), and Audio/Music processing (synthesis, source separation, classification). The core message for the future is that 'Tools change, humanity adapts' (Araçlar değişir, insanlık adapte olur), comparing AI access to the progression from libraries to the internet to AI itself. The training process, especially for LLMs, requires massive computational resources, exemplified by the Grok-3 model needing 200,000 GPUs, costing $5 billion.

Context: Burak Sina Akbudak delivered this talk at TEDxIzmir Fen Lisesi Youth, focusing on explaining the underlying mechanics and practical applications of modern Artificial Intelligence, especially Large Language Models (LLMs). The presentation systematically walks the audience through the lifecycle of building an AI model, from initial data collection and architecture design to advanced fine-tuning and validation techniques, using analogies like architecture (Mimari) versus the final product (Model) to clarify complex concepts.