Exploring LLMs across clouds – A Year in the Life
summary
Side-by-side comparison of cloud LLM stacks: Google’s Gemini 1.5 offers multimodal strength and charges about $1.75 input / $5 output per million tokens, Microsoft’s GPT 40 delivers the best multimodal reasoning but costs roughly twice as much, and Amazon’s Titan text Express aims at cheap text-only embedding workloads while leaning on open‑source models. Core engineering details: embeddings → vector DB (Pinecone, BigQuery, Cosmos/PG), retrievers (Kendra/OpenSearch/Vertex AI Search), RAG orchestration and toolkits (Bedrock, Vertex AI, Azure OpenAI, Lex/Dialogflow/Co‑pilot Studio). Practical claims: Exclusive infra deals, token pricing, and a 2x+ productivity boost for developers.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 1:53What GPT and LLMs mean
- 5:40Milestones in GPT models
- 7:46Cloud providers' LLM approaches
- 10:33Foundation models: Amazon, Google, Microsoft
- 13:44Benchmarking LLMs and licensing
- 15:50LLM pricing models
- 18:40Knowledge engineering and search
- 20:46Google and Microsoft AI assistants
- 21:48Developer and chat assistants
- 23:14Retrieval augmented generation
- 25:44RAG ecosystem overview
- 28:54Real-world LLM applications
- 29:58Generative AI productivity tools
- 30:43AI tool adoption management
- 31:03Summary and future of generative AI
related talks
49mBen Hall & Barbara Fusinska
27mThe Cutting Edge of AI: From LLMs to Multimodal Intelligence
40mTechniques to avoid common LLM pitfalls
39mA New Stack for the Gen AI Era
37mExpanding the Capabilities of LLMs Through Community Collaboration
32mGenerative AI Text and Multimodal Embedding Models for Real World Use Cases
24mHarnessing the Power of LLMs in Advanced Analytics
26mHow Does LLM and Generative AI Really Work?