Overview

​Access, Features and Limitations

The AI chatbots can be accessed on the Clarity platform. You will need to log in with your Yale email. 

The rate message functionality allows you to indicate whether you find a response helpful or not and to submit comments about the response. The information is not shared with anyone at Yale. The ratings do not impact the responses from the chatbot. In the future we may extend this functionality of these buttons to aggregate feedback in order to improve the performance of the chatbot, after confirming with the user that they agree to share the feedback.

A token count is displayed for your prompt and for the generated response. Instead of words or characters, tokens are a measurement tool used to assess the length of chat inputs and outputs, defining the amount of work the agent must perform. Usually, it is not something you as a user should be worried about.

None of the AI models on the Clarity platform are capable of browsing the internet to provide up-to-date information. The agents generate responses based on the training data of their underlying AI models. For information on the models’ knowledge cutoff dates, please refer to the Clarity Agents Page.

All of the Clarity featured agents support file uploads. 

​For searching within a file, the following file types are supported: 

  • Text Files: .doc, .docx, .pdf, .pptx, .txt, .md, .tex
  • Data Files: .json
  • Code Files: .c, .cpp, .cs, .css, .html, .java, .js, .php, .py, .rb, .sh, .ts

​For interpreting a file using code, the following file types are supported: 

  • Text Files: .doc, .docx, .pdf, .pptx, .txt, .md, .tex
  • Data Files: .csv, .xlsx, .json, .xml
  • Code Files: .c, .cpp, .cs, .css, .html, .java, .js, .php, .py, .rb, .sh, .ts
  • Image Files: .gif, .jpeg, .jpg, .png
  • Compressed/Archive Files: .tar, .zip

For more information on working with files and using Clarity to generate downloadable files, please refer to Working with Files in Clarity.

​Over 100 languages are supported. However, please be aware that while the AI chatbot can understand and generate text in these languages, the quality of the responses may vary based on the language’s complexity and the context of the request.

Currently, you can’t export your chat conversations, but you can print your chat conversations and save the chat conversation as a PDF. Please refer to this how-to guide outlining the steps to do so.  

Files may be rejected if they do not meet our security or formatting requirements. Common reasons include:

  • Format Mismatch: The internal file structure does not match the file extension (e.g., a renamed .exe file).
  • Unsupported File Type: The file format is not permitted by the system.
  • Security Filters: The file failed content safety checks or triggered our Microsoft Prompt Shield filters.

To protect our users, we use enhanced security filters to block potential risks, including prompt injection attacks. If an injection is detected, only the affected “chunks” of the file are blocked; however, for smaller files, this may result in the entire document being unavailable.

Note: We are aware that these security filters can occasionally trigger “false positives.” We are actively working to improve their accuracy.

How to fix this: If you believe your file is safe, try converting it to a different format (e.g., from .pdf to .docx) or copying the specific text you need directly into the chat prompt.

Custom Agents

Access to create Custom Agents will be granted upon approval of submitted use cases. Detailed information can be found on the Custom Agents Overview page.

Instructions for creating or managing Custom Agents upon use case approval are provided on the Create/Manage Your Agent page.

Pending use case approval, users can create up to 5 Custom Agents for a use case in Clarity. If you are looking to create agents for a different use case, an additional use case request needs to be submitted.

No, you cannot switch models for your Custom Agents. Create a new agent to use a different model. 

You can select from several different models when creating your Custom Agent. To learn more about how the models differ, please refer to the Agents page.

There is no limit to the number of users you can grant access to your agent, however they must be added individually. If your use case requires a large number of users to have access, you can reach out to ai.platforms@yale.edu for assistance.

Refreshing the Manage Agents page should cause your saved changes to be reflected. 

Review your custom agent’s system prompt as you may need to adjust it to better guide your agent’s behavior. Also, check any documents that you might have uploaded as the agent’s knowledge source that might affect the responses. You can also ask questions our Teams based Community of Practice  in the or visit our bi-weekly Office Hour Wednesdays between 10 and 11 (even weeks). 

While some models allow for adjusting temperature, you cannot modify the temperature for your agent in self-service. 

ePHI is not currently approved for Custom Agents or API access in Clarity. There are Health agents available for YSM community use. For more information, refer to the Health Sciences AI Toolkit and the Clarity Platform Security guidelines for approved data types.

Clarity API

API keys for Clarity are granted in relation to Custom Agents and must go through a review process before they are distributed. Review the Clarity API and Portkey webpage for more information. 

API keys expire after 90 days. If your Clarity API key expires, the AI Platform Services team will issue a new one; for Portkey, you can generate a new API key yourself.

API costs are charged monthly based on consumption. Please refer to the API Pricing page for more information.

Users can create up to 5 Custom Agents in Clarity.

After you have created your Custom Agent in Clarity you reach out to the ai.platforms@yale.edu to get an agent access token. The key will be shared with you through Azure Key Vault. 

Security Questions

Training and General Instructional Questions

Training material, including key terms and training videos on AI, can be found on the Learn About AI page.   

​Training material specific to the Clarity platform can be found on the Clarity Platform page.

​To upload a file during your conversation, you can use the “Attach File” button located at the bottom of the chat interface and next to the message box.  

​Here’s a step-by-step guide: 

  1. ​Select the paperclip button: Look for the paperclip icon labeled “Attach File” in the chat interface. 
  2. ​Select your file: A file explorer window will open, allowing you to select one or multiple files from your computer. Navigate to the file you want to upload and select it. 
  3. ​Upload the file: After selecting the file, confirm the upload by selecting the upload button. A progress indicator will display, indicating that the file is being uploaded and a success message will display when the file has been successfully uploaded. 

Once the file is uploaded, you can interact with the AI chatbot to get analyses, summaries or to process information from the document. 

There is no limit on number of file uploads and you can upload up to 10 files simultaneously. 

​The file size limit is 512MB for any type of file. An error message will show up on your chat session if the file limit has been exceeded. In addition, the size of the extracted text (not the size of the original file) cannot exceed 10,000,000 characters.

​There is no limit to how many chat sessions a user can have. If you don’t want to keep a chat, you have the option to delete it from the chat history. 

​To have multiple chat sessions at once, click on the “+” in the upper left corner of your chat window to start additional chat sessions. 

The AI chatbot on the Clarity platform is a general-purpose AI chatbot and it has not been trained on Yale specific data. AI chatbots, including the AI chatbot on the Clarity platform, may make mistakes in its responses. AI chatbots are very good at generating content, which is one of the reasons they are very popular, but that content can be inaccurate, misleading, or entirely fabricated, sometimes called “hallucinations”. This is one of the limitations of AI chatbots and it highlights why it is important to always fact-check the responses you receive. Remember that you are responsible for any content you produce or publish, including AI-generated material.  

Also, it’s important to understand the limitations of AI chatbots. They might not always have the information you need, especially if it’s very domain-specific or beyond their knowledge cut-off date.  AI chatbots are powered by large language models that may reflect biases in their training data. Consequently, the responses they produce reflect their inherent bias and the quality of their programming and training data.   

​The Clarity AI chatbot has several agents available. Please refer to the Clarity Agents Page which outlines each of the agents and their differences.

The featured agent chatbots on the Clarity platform know the current date. 

Image Generation

Clarity’s ability to generate images, which was previously available using DALL-E 3 in the GPT-4o and GPT-4o mini agents, is no longer available. We will be implementing an updated solution for image generation in the future, but in the meantime, image generation functionality is available in Microsoft Copilot Chat and Google Gemini for Education, both of which are part of Yale’s AI Toolkit.

OneDrive

Please refer to the Knowledge Base article on how to connect to your Yale OneDrive account in Clarity.  

Your profile picture appears in Clarity because it is linked to your Yale-affiliated Microsoft account. When you connect your OneDrive or use the same account details to log into Clarity, the associated profile information, including your profile picture, gets integrated. 

There are two ways that you can disconnect from your OneDrive in Clarity.  

You can make a soft disconnect by: 

  1. Select the paper clip icon next to the text message area. 
  2. Select the option “Disconnect OneDrive” and you will now not be able to select files from your OneDrive until you decide to reconnect to your OneDrive account.  

The soft disconnection from your OneDrive account does not revoke your access grant via Microsoft. Please follow the following steps, if you want to revoke the access grant via Microsoft too: 

  1. Navigate to Microsoft’s website. 
  2. Find the app named FoundationaLLM-Core-Portal and select the three dots next to the app and choose “Manage your application”. Do NOT select the option to “Remove” the app. 
  3. On the FoundationaLLM-Core-Portal page you have the option to the right to select “Revoke Consent”. 
  4. Now when you go back into Clarity and select a file from OneDrive, a pop-up window will ask you to grant Clarity the necessary permissions to access your OneDrive files. You will only have to accept the permissions to connect to your OneDrive. 

There is a hard limit of 512MB per file and you can upload up to 10 files simultaneously. All common file types are supported, including text files, spreadsheets, presentations, and documents. Please refer to the FAQ that outlines the supported file types.