❓ Successful Building a GenAI use cases just requires the latest and greatest frontier model, right?!

❓ Successful Building a GenAI use cases just requires the latest and greatest frontier model, right?!

🔍 In my conversations with customers I often realize that the choice of the best & shiniest model is highly occupying their resources, while thinking and focusing on architecting and building the use cases which ultimately should fulfil users needs gets very little attention. This naturally fuelled by the fast-pace and loud announcements of new frontier models, each announced with claims of superior benchmark results and new capabilities—and benchmark wins don’t automatically translate into better results for your specific use case. What we often forget is that the majority of capabilities become commodity among different models very fast — at least that is the pattern I keep seeing across release cycles. Hence—whenever several models can deliver the task quality you need—it often makes much more sense to focus on the use cases and building the generative application for the use case in a way it really addresses the user’s need and is able to evolve to utilize new release models where this makes sense. I wrote about this in my blog post (https://medium.com/@stefanchristoph/how-do-you-chose-the-foundation-model-for-your-generative-ai-app-like-your-car-18f8a299678c.)

📖 𝗕𝘂𝘁 𝗵𝗼𝘄 𝘁𝗼 𝗯𝘂𝗶𝗹𝗱 𝘁𝗵𝗼𝘀𝗲 𝗮𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲𝘀? Very happily I came across Chip Huyen’s blog post “Building A Generative AI Platform” - https://huyenchip.com/2024/07/25/genai-platform.html in which she outlines an Architecture Blueprint for Generative AI Applications and runs the reader through each of the possible elements. I found that a very insightful read and highly recommend it if you aim to build your own applications or want to review your existing architecture.

👀 Chip Huyen is currently working on her book “AI Engineering” - looking forward to read it.

🏗 As I’m working at AWS, I naturally looked at how to build this Blueprint Architecture on AWS. Not going into the details here, but e.g. Amazon Bedrock Guardrails’ Independent API—the ApplyGuardrail API (https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails-use-independent-api.html)—fits the blueprint’s input/output-guardrails component: it evaluates content as a standalone call, without invoking a foundation model, so you can place the check anywhere in your application flow. If you want to read more about how to build a Generative AI Architecture on AWS let me know!

• AI
• Architecture
• AWS