Large Language Models and Prompt Engineering for Developers
summary
Overview of practical constraints and fixes for using large language models as part of real apps: models are trained on web-scale data with a point-in-time cutoff (roughly 2021–2022) and grow to huge size (GPT-4 estimated ~1.5 trillion parameters), yet they have no memory, no action — they hallucinate (created a fake “Risto Roop” founder) and miscompute (cube root returned 21.6). Key specifics: token mechanics, context window (4,100–33,000 tokens), temperature effects, zero/few-shot vs fine-tuning, and engineering patterns like RAG, embeddings + vector databases, Azure OpenAI, GitHub Copilot, TypeChat, Semantic Kernel, and LangChain.
AI-written summary · about
chapters (13)
- 0:00LLMs and prompt engineering
- 2:30AI and generative models history
- 6:00Foundation models capabilities
- 8:51How LLMs work
- 12:08LLM memory limits
- 13:34Model calculation limits
- 14:59Tokens and context windows
- 18:10Sampling and temperature
- 20:39Creativity vs fact retrieval
- 22:48Prompt engineering basics
- 24:15Zero-shot, few-shot, fine-tuning
- 27:24Developer workflow tools
- 28:46Capabilities and responsibilities
related talks
43mBeyond the Hype: A Realistic Look at Large Language Models
27mLarge Language Models and the Future of AI
01h 00mInside GPT – Large Language Models Demystified
51mTowards a Pragmatic Understanding of Large Language Models
15mFuture with Large Language Models
51mLarge Language Models for Real-World Applications - A Gentle Intro
01h 00mUnleashGPT: Mastering the Implementation of Large Language Models
44mBuilding projects with language models