Avoid common LLM pitfalls
summary
Shows a working lineup of fixes: uses Gemini 1.5 Pro 1M+ context (demo claims 32k, 128, 1M and now 2M windows), builds a PDF-to-RAG pipeline with LangChain RAG pipeline and Firestore vector store that returns exact facts instead of hallucinations, implements function calling for live weather via geo+weather APIs, adds context caching (default 60 min) and automated checks with DeepEval/PromFu to score and red‑team failures; code and demos are in the repo.
AI-written summary · about
chapters (11)
- 0:00Introduction and overview
- 7:29Common LLM pitfalls overview
- 12:07Using LLM frameworks for processing
- 13:13Mitigating hallucinations with grounding
- 16:22Retrieval Augmented Generation (RAG) explained
- 23:02Function calling for real-time data
- 28:18Controlling LLM output formats
- 30:47Context caching to reduce costs
- 32:57Evaluating LLM outputs
- 37:16Testing for safety and security
- 44:03Summary and best practices
related talks
40mTechniques to avoid common LLM pitfalls
51mAvoid common LLM pitfalls
36mTechniques to avoid common LLM pitfalls
27mBuilding Useful Apps With Small, Local LLMs
48mCrafting your own RAG system: Leveraging 30+ LLMs for enhanced performance
01h 49mAI engineering with open access LLMs that lie, curse, and steal
49mRAG: from dumb implementation to serious results
48mGoing Beyond Prompt Engineering