Presentation slides: 🔎 Search Smarter in Generative AI Apps
Takeaway links
- Learn more about Azure AI Search: https://aka.ms/AzureAISearch
- Explore Azure AI Studio for a complete RAG development experience: https://aka.ms/AzureAIStudio
- Dig more into quality evaluation details and why Azure AI Search will make your application generate better results: https://aka.ms/ragrelevance
Azure OpenAI Service on your data: 📺 https://www.youtube.com/watch?v=9IBEVMQh5FQ
- In Copilot Studio, creating a GPT doesn't train a Generative Pretrained Transformer model. In this case a "GPT" is a LLM combined wtih a System Message and a data source for RAG (think of it as "RAG in a box"). Like the ChatGPT GPTs. As Chimes said "Don't train a GPT" - i.e. don't train a Generative Pretrained Transformer, but feel free to build a "RAG in a box".
- You could use Florence Computer Vision model to analyze the content and then load that into a Vector index (as chunks). Or Azure Video Indexer. For specialised use cases (eg videos), it'll make sense to do pre-processing on the data using specialised tools (eg computer vision, Video Indexer, python scripts) before pointing Copilot at it. It'll be better and/or cheaper - horses for courses still applies.
- Coutesy of Copilot in Teams meetings (my prompt was "Gather all links shared in the chat into a list, including the link and a description of what it is")
- Here are the links shared in the chat with their descriptions:
- Dan Walker's LinkedIn post about his six-month activity 1
- Azure OpenAI Service models (The three (gen2 and gen3) embedding models available in AOAI (from OAI)) 2
- microsoft/semantic-kernel (Integrate cutting-edge LLM technology quickly and easily into your apps) 3
- Microsoft Copilot Studio sign-up page 4
- Microsoft Copilot for Azure (AI Companion and Assistant in Azure) 5
- Microsoft Build Vision Applied Credential (Vision Studio is part of it)
- Here are the links shared in the chat with their descriptions:
