RAG, Tooling, MCP, Agents ... and What is Missing?
summary
Explains concrete wins and failures: retrieval-augmented generation avoids costly retraining by embedding recent docs but requires careful chunking (15 strategies) and often pairs with elasticsearch rather than huge vector DBs; tooling via the MCP multicontext protocol enables function calling and orchestration; agentic setups use recursive self‑learning and code‑execution feedback but still fail in embodied tasks (the robot butter trial runs at well under human performance). Notes OpenAI’s double‑digit quarterly loss, local 4‑bit quantized models on a 16GB machine, and real legal/fraud hallucination cases.
AI-written summary · about
chapters (22)
- 0:00Introduction
- 1:10Retrieval augmented generation matters
- 3:18Criticism of LLMs
- 9:00Hinton’s AI perspective
- 12:32Hype, solutions, future steps
- 13:36How RAG works
- 16:02Knowledge graphs and context
- 17:10Tooling and MCP
- 19:54Agentic systems
- 20:18Self-learning feedback
- 21:19File search and hybrid retrieval
- 22:06Local models trends
- 23:28Privacy and on-premise
- 24:33Model efficiency trends
- 25:38Limitations of LLMs and power efficiency
- 26:44LLMs for AGI?
- 28:07LLMs as AGI building blocks
- 30:18Q&A: RAG techniques
- 31:45Q&A: AI investing
- 32:31Q&A: AI architectures
- 33:38Q&A: LLMs and empathy
- 35:47Closing thoughts and event wrap-up
related talks
01h 01mAI Agents In-Depth – Function Calling, MCP and Tool Use Under the Hood
22mThe Power of RAG at Scale: A Year of Explosive Growth and Observab...
45mAgents, Tools, and MCP, oh my! Next-level AI concepts for developers
50mRetrieval Augmented Generation Beyond Toy Demos
27mOptimizing RAG Systems in Practice: Challenges and Proven Solutions
42mTo RAG or not to RAG
22mAgentic AI for Engineering Leaders | Strategy & Execution Guide
31mAdvanced Enterprise RAG Systems