In Episode 7 of Mr. Fred’s Tech Talks, I dive deeper into Large Language Models (LLMs) and explore how they’re trained. Using the fun analogy of a parrot that never stops practicing, Mr. Fred explains the 9-step training pipeline: from collecting massive datasets and tokenizing text, to neurons, weights, backpropagation, GPUs, fine-tuning, and safety alignment...but in a LOW TECH JARGON way.
I’ll also talk about probability math, why LLMs don’t really “understand” but instead predict the most likely next word, like rolling loaded dice. Along the way, enjoy some nostalgic sound bites from movies and TV that connect the dots between memory, patterns, and AI.
🎧 Highlights:
Whether you’re a parent, teacher, student, or just curious about AI, this episode will give you a fun and clear view of how language models actually learn.
","author_name":"Fred Aebli"}