Building and using AI models responsibly Q&A
summary
Preview release shows Azure Machine Learning now ingests images and open text, captures rich metadata, and adds interpretability overlays and monitoring for production models; teams use Responsible AI dashboard with custom metadata slices and feature attribution to find cohorts, and use Prompt Flow with GPT‑4 as an evaluator to measure groundedness and relevance when ground truth is absent. Azure Content Safety, custom blocklists, 32K context windows, and retrieval‑augmented generation are supported; endpoints do not auto‑upgrade model versions.
AI-written summary · about
chapters (21)
- 0:00Q&A
- 1:57Azure ML new features for Responsible AI
- 3:49Industry use cases of Responsible AI dashboard
- 5:42Evaluating LLMs with Prompt Flow
- 7:29Microsoft Copilot Responsible AI practices
- 9:40Customizing Copilot for enterprises
- 11:22Industry-specific Responsible AI apps
- 13:17Responsible AI dashboard setup
- 14:00Explainable AI in Responsible AI tools
- 16:36Security limits for AI bots
- 19:49Mitigating data poisoning
- 22:43Responsible AI dashboard + Prompt Flow
- 24:08Prompt Flow vs Responsible AI metrics
- 25:35Multilingual AI applications
- 27:21AI safety research collaboration
- 28:46Addressing public AI fears
- 31:42On-premises Responsible AI tools
- 33:34Microsoft’s open-source Responsible AI
- 36:09Monitoring AI compliance
- 39:01Azure ML + Azure OpenAI integration
- 40:32Model version updates in production
related talks
48mDeploy large language models responsibly with Azure AI
31mAzure AI Innovations at Microsoft
38mSafe Evaluation and Rollout of AI Models
50mAzure AI, Power new possibilities for every organization
51mThe Next Frontier: Azure OpenAI Models and the Arrival of GPT-4
01h 01mAI and ML imbedded in Azure Data Platform
01h 00m5 Levels of AI in Microsoft Azure
57mEasier AI allows everyone to build smart apps