"Direct Quotes" feature for Creator Studio

Definitely understand the need – as language models are getting stronger, their hallucination risk goes down. If you can share improper summarizations of those longer text responses, we can work with your team to address the gap & steer the AI agent to preserve the information that’s absolutely critical for your financial & contractual considerations. We should be able to preserve things like disclaimers in our current dialog generation capabilities.

However, if absolutely necessary, you can control exact copy of disclaimers, etc. via our notification action. This is by no means a best practice, but it is a workaround that you can take advantage of today. https://help.moveworks.com/docs/built-in-actions#send_plaintext_chat_notification

Thanks ​@Ajay Merchia. We’re seeing that any text longer than 200 characters is being summarized and the end user experience is compromised. Here’s the creator studio configuration (event) vs summarization we’re getting. 

Question configured in Creator Studio Event: A Statement of Work (SOW) documents the work products and services, including, but not limited to the work activities and deliverables to be supplied under a contract or as part of a project timeline. In contrast to a typical Staff Augmentation arrangement which is billed based on time worked, SOW agreements are usually billed based on fixed price deliverable or hitting specific milestones (Service Providers, Managed Services, Freelancer/Independent Contractor contracted via SOW) Is this what you are looking for?

Question Displayed as response:
Is this what you are looking for? / Is this the information you were seeking ?

This sounds more like a knowledge article - I’m a bit confused why you’re configuring it as a question instead of a text response, or adding it to your knowledge base

​@Ajay Merchia - We have a business requirement where employees can trigger Creator Studio with certain keywords to ask clarifying questions and guide them to the right process and platform. However, we found that Creator Studio was summarizing instead of providing the desired response (bmandan provided example above). Our Moveworks customer success manager suggested setting up this idea so the Creator Studio product team can incorporate a way to allow the exact message or direct quote. We did evaluate other Moveworks options and a knowledge article isn’t appropriate for this business requirement. Hope this provides more context. Please let us know if there are any further queries.

​@Ajay Merchia Here are some examples of AI rewording our inputs. We’ve noticed this for both long and short text.

1. When business was testing they found an instance where short text was being rewritten by My Assistant:

   

    

2. Issue was originally noticed with long text.
   Desired text that gets summarizes: An Statement of Work (SOW) documents the work products and services, including, but not limited to the work activities and deliverables to be supplied under a contract or as part of a project timeline. In contrast to a typical Staff Augmentation arrangement which is billed based on time worked, SOW agreements are usually billed based on fixed price deliverable or hitting specific milestones (Service Providers, Managed Services, Freelancer/Independent Contractor/1099 contracted via SOW) Is this what you are looking for?  For additional information, please visit the following link: https://murray.paramount.com/murray?page=non-staff-labor-management-new

   

   
   Reworded text that does not get summarized: Do you want to hire using Statement of Work (SOW) which is an agreement that outlines deliverables and milestones, and is billed based on fixed prices instead of time worked? For additional information, please visit the following link: https://murray.paramount.com/murray?page=non-staff-labor-management-new
   1. Result wording is only slightly changed:

      

      
       

It looks like these rephrases are very minor (e.g. “Will” changes to “Would”)

These types of customizations aren’t a good fit for Agentic AI & Generative AI – when building AI agents, you do need to “release” control to the AI agent to make the decisions that are best for each specific user (e.g. consider different dialects, multi-part questions, verifiable citations, clarifying questions, etc.).

However, I do see other cases when necessary information is being omitted. We do have ways to steer your AI agent to return information correctly to users. 

https://help.moveworks.com/docs/reasoning-instructions

This should be a good place to get started, but I also chatted with your account team more internally about your specific use case. Myself & our professional services team will be in touch!