Inside GPT – Large Language Models Demystified
summary
Downloads include Jupyter notebooks and a zip that run GPT2 from scratch (about 126M parameters, 12 layers, 12 heads, 768-d embeddings) so you can step through tokenization, positional embeddings, attention and logits on a CPU. Live demos show a weak GPT2 chatbot, a prototype that queries a local Nobel database (and accidentally inserts a 2025 laureate), and how sampling choices like temperature and top-p change deterministic logits into varied outputs.
AI-written summary · about
chapters (12)
- 0:00Welcome
- 1:49Language model hallucinations
- 5:27Practical applications
- 10:26Chatbot and LLM architecture
- 15:01Prompt engineering
- 17:31GPT2 demo and limits
- 22:11Tokenization challenges
- 33:04Embeddings and vector similarity
- 41:06Transformer and attention
- 51:19Model layers and generation
- 54:30Sampling parameters impact
- 1:00:51Advanced sampling and closing
related talks
01h 00mInside GPT – Large Language Models Demystified
47mInside GPT – Large Language Models Demystified
41mInside GPT – The Maths Behind the Magic
28mGPT or how I learned to stop worrying and love the generative AI
45mHow to use GPT in your projects: A Case Study
51mLarge Language Models for Real-World Applications - A Gentle Intro
43mBeyond the Hype: A Realistic Look at Large Language Models
56mLarge Language Models for Python Specialists