Techniques to avoid common LLM pitfalls
summary
Overview of actionable fixes: use Retrieval-Augmented Generation (RAG) to ingest private PDFs (example uses 500-char chunks with 100-char overlap, embeddings into a vector DB) so a query returns the manual’s cargo capacity 13.5 cubic feet instead of a hallucination; add Function Calling to fetch live data (weather API dance) and use grounding via Google Search or Vertex AI Search; orchestrate pipelines with LangChain, enable context caching and batch generation (about 50% cost saving), and run evaluations with DeepEval/Promu and LLM Guard for safety.
AI-written summary · about
chapters (11)
- 0:00Introduction and speaker welcome
- 2:16LLM fundamentals overview
- 7:17Common LLM problems explained
- 10:29Solutions for LLM pitfalls
- 14:20RAG for private data grounding
- 20:46Function calling for real-time data
- 23:34Controlling LLM output format
- 25:20Cost optimization techniques
- 28:13Evaluating LLM performance
- 30:18Safety and security checks
- 35:36Summary and closing remarks
related talks
51mAvoid common LLM pitfalls
40mTechniques to avoid common LLM pitfalls
57mLLMs gone wild
53mFrom RAG to Riches: Secrets of Data-Augmented LLMs
18mLLMs and GenAI: A Guide to Building Your Own RAG Model for Document Interaction
33mDeep research at scale
33mDesigning LLM Applications for Massive Datasets
47mBuilding and Testing AI-Agent Powered LLM Applications: A Live Demonstration