Merge pull request #6966 from MicrosoftDocs/ai-scenario-edits-pr

[BULK] AIEdit-discovered typos

Author: AnnaMHuff

docs/scenarios/ai/govern.md

@@ -51,7 +51,7 @@ AI governance policies provide a structured framework for responsible AI use wit
 
 ## Enforce AI governance policies
 
-The enforcement of your AI governance policies maintains consistent and ethical AI practices across your organization. You should use automated tools and manual interventions ensure policy adherence across all AI deployments. Here's how:
+The enforcement of your AI governance policies maintains consistent and ethical AI practices across your organization. You should use automated tools and manual interventions to ensure policy adherence across all AI deployments. Here's how:
 
 1. **Automate policy enforcement where possible.** Automated enforcement reduces human error and ensures consistent policy application across all AI deployments. Automation provides real-time monitoring and immediate response to policy violations, which manual processes cannot match effectively. Use platforms like Azure Policy and Microsoft Purview to enforce policies automatically across AI deployments, and regularly assess areas where automation can improve policy adherence.
 
@@ -93,4 +93,4 @@ The following table lists some common AI risks and provides a mitigation strateg
 | R007| Adversarial attack | Use PyRIT to test AI workloads for vulnerabilities and strengthen defenses.|The Security Development Lifecycle and AI red team testing must be used to secure AI workloads against adversarial attacks. |
 | R008| Insider threats| Use Microsoft Entra ID to enforce strict access controls that are based on roles and group memberships to limit insider access to sensitive data.| Strict identity and access management and continuous monitoring must be used to mitigate insider threats. |
 | R009| Unexpected costs| Use Microsoft Cost Management to track CPU, GPU, memory, and storage usage to ensure efficient resource utilization and prevent cost spikes. |Monitoring and optimization of resource usage and automated detection of cost overruns must be used to manage unexpected costs.|
-| R010| Underutilization of AI resources| Monitor AI service metrics, like request rates and response times, to optimize usage.| Performance metrics and automated scalability must be used to optimize AI resource utilization. |
+| R010| Underutilization of AI resources| Monitor AI service metrics, like request rates and response times, to optimize usage.| Performance metrics and automated scalability must be used to optimize AI resource utilization. |

docs/scenarios/ai/infrastructure/compute.md

@@ -22,9 +22,9 @@ AI workloads require specialized virtual machines (VMs) to handle high computati
 
 Choose a suitable virtual machine image, such as the Data Science Virtual Machines, to access preconfigured tools for AI workloads quickly. This choice saves time and resources while providing the software necessary for efficient AI processing
 
-- *Start with the Data Science Virtual Machines images.* The [Data Science Virtual Machine](/azure/machine-learning/data-science-virtual-machine/overview) image offers preconfigured access to data science tools. These tools include PyTorch, TensorFlow, scikit-learn, Jupyter, Visual Studio Code, Azure CLI, and PySpark. When used with GPUs, the image also includes Nvidia drivers, CUDA Toolkit, and cuDNN. These images serve as your baseline image. If you need more software, add it via a script at boot time or embed into a custom image. They maintain compatibility with your orchestration solutions.
+- *Start with the Data Science Virtual Machines images.* The [Data Science Virtual Machine](/azure/machine-learning/data-science-virtual-machine/overview) image offers preconfigured access to data science tools. These tools include PyTorch, TensorFlow, scikit-learn, Jupyter, Visual Studio Code, Azure CLI, and PySpark. When used with GPUs, the image also includes Nvidia drivers, CUDA Toolkit, and cuDNN. These images serve as your baseline image. If you need more software, add it via a script at boot time or embed it into a custom image. They maintain compatibility with your orchestration solutions.
 
-- *Find alternative images as needed.* If the Data Science Virtual Machine image doesn't meet your needs, use the [Azure Marketplace](https://azuremarketplace.microsoft.com/marketplace/apps) or other search [methods](/azure/virtual-machines/overview#distributions) to find alternate images. For example, with GPUs, you might need [Linux images](/azure/virtual-machines/configure) that include InfiniBand drivers, NVIDIA drivers, communication libraries, MPI libraries, and monitoring tools.
+- *Find alternative images as needed.* If the Data Science Virtual Machine image doesn't meet your needs, use [Azure Marketplace](https://azuremarketplace.microsoft.com/marketplace/apps) or other search [methods](/azure/virtual-machines/overview#distributions) to find alternate images. For example, with GPUs, you might need [Linux images](/azure/virtual-machines/configure) that include InfiniBand drivers, NVIDIA drivers, communication libraries, MPI libraries, and monitoring tools.
 
 ## Pick a virtual machine size
 
@@ -64,4 +64,4 @@ Containers provide a consistent, reproducible environment that scales efficientl
 ## Next step
 
 > [!div class="nextstepaction"]
-> [Storage IaaS AI](./storage.md)
+> [Storage IaaS AI](./storage.md)

docs/scenarios/ai/infrastructure/governance.md

@@ -39,7 +39,7 @@ Cost management monitors and controls expenses related to AI workloads on Azure.
 
 - *View spending patterns.* Use the Azure [Cost analysis](/azure/cost-management-billing/costs/quick-acm-cost-analysis) tool to regularly review spending patterns. This process identifies trends and reveals areas for potential savings, especially in VM usage.
 
-- *Allow specific virtual machine SKUs.* Use Azure policy to allow only the virtual machines SKUs that align with your AI budget. The built-in policy definition [*Allowed virtual machine SKUs*](https://ms.portal.azure.com/#view/Microsoft_Azure_Policy/PolicyDetailBlade/definitionId/%2Fproviders%2FMicrosoft.Authorization%2FpolicyDefinitions%2Fcccc23c7-8427-4f53-ad12-b6a63eb452b3) can enforce this control.
+- *Allow specific virtual machine SKUs.* Use Azure policy to allow only the virtual machine SKUs that align with your AI budget. The built-in policy definition [*Allowed virtual machine SKUs*](https://ms.portal.azure.com/#view/Microsoft_Azure_Policy/PolicyDetailBlade/definitionId/%2Fproviders%2FMicrosoft.Authorization%2FpolicyDefinitions%2Fcccc23c7-8427-4f53-ad12-b6a63eb452b3) can enforce this control.
 
 - *Consider autoscaling.* Use a [virtual machine scale set](/azure/virtual-machine-scale-sets/overview) to dynamically adjust VM counts based on demand, optimizing costs.
 

docs/scenarios/ai/infrastructure/well-architected.md

@@ -90,7 +90,7 @@ Performance efficiency refers to maximizing the processing power of Azure infras
 
 Increase the clock rate of a graphics processing unit (GPU) to improve performance, especially for tasks requiring high graphical processing or complex computations. Higher clock speeds allow the GPU to execute more operations in a given time period, enhancing overall efficiency. Use this [GPU-optimization script](https://github.com/Azure/azurehpc/tree/master/experimental/gpu_optimizations#gpu-optimization) to set the GPU clock frequencies to their maximum values.
 
-- *Enable Accelerated Networking.* Accelerated Networking is a hardware acceleration technology that allows virtual machines to use single root I/O virtualization (SR-IOV) on supported virtual machines types. It provides lower latency, reduced jitter, and decreased CPU utilization. Enable accelerated Networking offers substantial enhancements in front-end network performance.
+- *Enable Accelerated Networking.* Accelerated Networking is a hardware acceleration technology that allows virtual machines to use single root I/O virtualization (SR-IOV) on supported virtual machine types. It provides lower latency, reduced jitter, and decreased CPU utilization. Enabling accelerated Networking offers substantial enhancements in front-end network performance.
 
 ### I/O tuning
 
@@ -118,7 +118,7 @@ NVIDIA Megatron-LM is an open-source framework for training large language model
 
 Megatron-LM deploys on Azure HPC infrastructure, and it uses Azure’s scalability for large language models without requiring on-premises hardware.
 
-#### Megatron-LM test set up
+#### Megatron-LM test setup
 
 Deploying Megatron-LM requires specific software and hardware.
 
@@ -144,7 +144,7 @@ This setup ensures efficient deployment and training of large language models on
 
 ### NCCL bandwidth test
 
-To verify and optimize GPU communication across nodes, run the NCCL bandwidth test. The NCCL bandwidth test is specialized tool within NCCL, a library that facilitates high-speed communication between GPUs. NCCL supports collective operations, including all-reduce, all-gather, reduce, broadcast, and reduce-scatter, across single or multi-GPU nodes, and achieves optimal performance on platforms with PCIe, NVLink, NVswitch, or networking setups like InfiniBand or TCP/IP. For more information, see [NVIDIA/NCCL tests](https://github.com/nvidia/nccl).
+To verify and optimize GPU communication across nodes, run the NCCL bandwidth test. The NCCL bandwidth test is a specialized tool within NCCL, a library that facilitates high-speed communication between GPUs. NCCL supports collective operations, including all-reduce, all-gather, reduce, broadcast, and reduce-scatter, across single or multi-GPU nodes, and achieves optimal performance on platforms with PCIe, NVLink, NVswitch, or networking setups like InfiniBand or TCP/IP. For more information, see [NVIDIA/NCCL tests](https://github.com/nvidia/nccl).
 
 #### NCCL performance metrics
 

docs/scenarios/ai/secure.md

@@ -17,7 +17,7 @@ This article helps you establish a security process for AI workloads in Azure. A
 
 AI workloads create new attack surfaces that traditional security measures can't address. You must systematically evaluate AI-specific vulnerabilities to build effective defenses. Here's how:
 
-1. **Identify AI system risks across your environment.** AI adoption introduces classes of risk that may not be explicitly addressed in traditional threat modeling. While your existing enterprise framework may already be capable of modeling these risks, AI governance requires deliberate validation that it does so. Begin with your established threat modeling framework, such a [STRIDE](/azure/security/develop/threat-modeling-tool-threats). Then reference AI‑specific risk inventories, like [MITRE ATLAS](https://atlas.mitre.org/) and [OWASP Generative AI risk](https://genai.owasp.org/) to confirm that AI‑specific attack techniques, misuse scenarios, and systemic risks are adequately represented. Use these sources to supplement, not replace, your existing framework and to inform consistent AI risk identification across the organization.
+1. **Identify AI system risks across your environment.** AI adoption introduces classes of risk that may not be explicitly addressed in traditional threat modeling. While your existing enterprise framework may already be capable of modeling these risks, AI governance requires deliberate validation that it does so. Begin with your established threat modeling framework, such as [STRIDE](/azure/security/develop/threat-modeling-tool-threats). Then reference AI‑specific risk inventories, like [MITRE ATLAS](https://atlas.mitre.org/) and [OWASP Generative AI risk](https://genai.owasp.org/) to confirm that AI‑specific attack techniques, misuse scenarios, and systemic risks are adequately represented. Use these sources to supplement, not replace, your existing framework and to inform consistent AI risk identification across the organization.
 
 2. **Assess AI data risks throughout your workflows.** Sensitive data in AI workflows increases the risk of insider threats and data leaks that can compromise business operations. Data risk assessment helps you prioritize security investments based on actual exposure levels. Use tools like [Microsoft Purview Insider Risk Management](/purview/insider-risk-management) to assess enterprise-wide data risks and prioritize them based on data sensitivity levels.
 
@@ -82,4 +82,4 @@ AI systems face evolving threats that require continuous monitoring to prevent s
 Govern AI, Manage AI, and Secure AI are continuous processes you must iterate through regularly. Revisit each AI Strategy, AI Plan, and AI Ready as needed. Use the AI adoption checklists to determine what your next step should be.
 
 > [!div class="nextstepaction"]
-> [AI checklists](index.md#ai-checklists)
+> [AI checklists](index.md#ai-checklists)