Differences

This shows you the differences between two versions of the page.

--- 711:generative_ai [2024/10/08 22:33] – [A few tips we learned so far] Meier, Erich
+++ 711:generative_ai [2025/02/21 09:25] (current) – [Create process content using AI] Linz, Andreas
@@ Line 1: / Line 1: @@
 ====== Create process content using AI ======
-Using Generative AI, Stages can help to draft activity, role, and work product descriptions or create a first set of activities for a workflow. As a modeler, you can pick a template for the description and select the best matching option that the AI is creating for you.
+Using Generative AI, Stages can help to draft activity, role, and work product descriptions or create a first set of activities for a workflow. As a modeler, you can pick from a predefined set of templates and select the best matching option for the content Stages AI should generate for you.
+**
+Please note: currently all AI related features are only available on UL hosted Stages servers (managed Services).**
 ===== Generate element descriptions =====
@@ Line 49: / Line 51: @@
 {{ :711:ai-6-generatedflow.png?direct }}
-===== A few tips we learned so far =====
+===== Translate process elements =====
-  * The AI generation works best for well-known element names, e.g. "Project Management", "Estimate resources", or "Project Manager"
+Click "Translate" on a process element to open the Translation dialog. Use the "Translate To..." button on the right to add a translation. The Name, Description, and Custom Attributes on the left will be translated into the selected language on the right tabs.
-  * Try to avoid internal abbreviations or organizational-specific terms.
+{{:711:stages_ai_translate.png?direct|}}
+===== Human in the loop =====
+Stages AI follows the principle of always having a human in the loop, so that AI-generated content cannot be automatically make it to end users without prior review or approval by a human user, either by using a process release automation or by manually creating a process version and marking it as valid.
+To support this principle, we advise all content authors, reviewers and approvers to always really review AI-generated content, in case the AI system produces non-sensical results.
+===== A few tips to generate good results =====
+  * The AI generation works best for well-known element names, e.g. "Project Management", "Estimate resources", or "Project Manager" where the AI system has a lot of data about
+  * Try to avoid internal abbreviations or organizational-specific terms.
 ===== How the AI output can be adjusted =====
@@ Line 59: / Line 71: @@
 The AI output is mainly influenced by the AI template you select when generating the content, also know as the "AI prompt".
-Adjusting the AI prompts can currently only be performed by UL Solutions experts.
+The AI output can be further optimized by providing additional documents. The AI system will then consult those documents and create responses in accordance with their content. Documents that work well are process-oriented standards like Automotive SPICE or ISO 15288 (Systems Engineering). Documents that do not mention specific processes like more abstract standards (e.g. ISO 9001) will less likely lead to an improved output.
-Please contact us if you would like to have a personal session with our experts to adjust your AI templates, gather experience with different types of prompts, and optimize the prompts for your organization or use cases.
+To minimize the risk of malfunction or misuse, adjusting the AI prompts and uploading additional documents can currently only be performed by UL Solutions experts.
+[[stages-support@ul.com|Please contact us]] if you would like to have a personal session with our experts to adjust your AI templates, gather experience with different types of prompts, upload additional documents, and generally optimize Stages AI for your organization or use cases.
 ===== Under the hood =====
-To generate the process content, Stages is connecting to a dedicated Microsoft Cognitive AI service instance, which is currently based on ''GPT-4.0''. Microsoft assures that the AI service does not store transmitted data and does not add it to its knowledge base, so it will not be shared with other users of the service.
+To generate the process content, Stages is connecting to dedicated Microsoft Azure AI service instances, which are currently based on ''GPT-4o'' and the Azure AI Translator. Microsoft assures that the AI services do not store transmitted data and do not add it to its knowledge base, so it will not be shared with other users of the service.
 To assure safety and security of the automated content generation via AI, the UL Solutions AI lab is monitoring the requests and responses exchanged between Stages and the AI service to detect anomalies and making sure that the AI service works as expected. The AI lab adheres to all relevant privacy and data security regulations and does not store or use the data beyond the purpose of monitoring AI safety.
@@ Line 72: / Line 87: @@
   * Name of the process element
-  * Name of the parent element (potentially)
+  * Names and descriptions of the parent element
-  * Name of associated elements (potentially)
+  * Names and descriptions of associated elements
-  * Name of the current workspace (potentially)
+  * Name of the current workspace
 The exact amount of data sent is defined in the AI template you select when generating the content.
@@ Line 84: / Line 99: @@
   * Your password or any other credentials
   * Your organization's name or any other identifier for your organization
-  * The IP address you are connecting from
+  * The IP address or location you are connecting from
+Let us know if you have any questions or concerns about your data and how it is being used.
-Let us know if you have any questions or concerns about your data and how it is being used. Your safety, security, and privacy is our top priority.
+Your safety, security, and privacy is our top priority.