How was Qstream AI developed?
We refined ChatGPT in the context of Qstream’s best practices and provided our product and engineering teams with clear, valid use cases used in the design and development of the solution.
What AI is being used to generate content?
Qstream AI uses ChatGPT 3.5 turbo. The team evaluated several options and considered other models, and ultimately selected ChatGPT’s enterprise API due to its scalability, relevant content compatibility with Qstream best practices and data validity demonstrated in the testing process. We will vet other large language models over time, as part of our longer term road map.
What’s the difference between ChatGPT and Qstream AI?
While we are leveraging ChatGPT for Qstream AI, our version of the model incorporates Qstream’s best practices. Additionally, Qstream AI essentially cuts out the middleman, allowing you to create content directly in the platform and in the Qstream format.
What happens to the content we feed the Qstream Microlearning Content Generator?
In common with the other data you provide to Qstream, we retain this information for as long as necessary – and in accordance with our Master Services Agreement with you. In general, this means the data is retained until our relationship with you ends – and is then deleted in accordance with our data governance policy.
In addition, data that you supply to us for AI Content Generation is included in API requests to a third party (OpenAI) who will process and retain this data in accordance with their Terms and Conditions. Currently, this means that OpenAI will retain your data for 30 days as is required by their T&Cs. OpenAI does not train their ChatGPT models using your data.
Per our current guidelines, you must not upload any HIPAA or other such personal data when creating Qstream content – whether via our AI Microlearning Content Generator or otherwise.
What is the level of accuracy and quality of Qstream AI outputs? Do I need to review it?
AI Content Generation uses probabilistic language models which can occasionally misinterpret your data, invent incorrect and improbable answers (“hallucinate”) or otherwise generate content unsuited to your needs. Please review and approve content generated by our AI Content Generator for accuracy, suitability and fitness-for-purpose.
Can I upload PDF files to generate questions using Qstream AI?
Currently, no, but an easy workaround is to copy and paste the text from your PDF.
Is there also a way to create multiple questions in once?
No and that’s deliberate. Inputting large data sets compromises the quality of the output when creating large volumes of content. Creating each question individually allows for precision and it’s faster than composing content from scratch. Plus, you will want to review all outputs individually.
Does AI enhance the functionality or performance of your solution?
Qstream AI accelerates the creation of content whether used by Qstream’s own content design team or in self-service by our clients. It reduces the content composition process needed to turn longer-form training assets into impactful scenario-based microlearning challenges. This helps Qstream clients get important training to employees faster as well as run more Qstreams over time to make learning continuous.
What data sources does the AI rely on, and how is this data processed and analyzed?
Qstream AI requires your input to create content. The content created can use public sources that are part of ChatGPT’s model where you use generic terms e.g., “describe the five best outcomes of a change management process” or specialized inputs that are specific to your use case e.g., based on a paragraph of training material documentation text you provide.
The results of content creation can be “re-generated” to create different content based on the same source. Qstream provides terms and conditions for proper use and clients ultimately determine (like all Qstream content) whether information is of such a sensitive nature that using it in the Qstream platform is permitted. That said, part of our decision to use ChatGPT for this processing was that the enterprise API includes clear terms that information submitted via API is not used to train its model.
How is the AI trained or updated to ensure accuracy and relevance over time?
Qstream has and will continuously evaluate AI offerings to ensure that we are using the most appropriate models for our use case, ensuring that the model is optimized to abide by our best practices. Our Content Team has been using (with client approval) ChatGPT since late 2022 to test different elements (e.g., bulk content creation, persona of author, temperature, order of questions, right test cases).
What measures are in place to address potential biases or ethical considerations in the AI algorithms?
One of the reasons we selected ChatGPT is that while all AI is a moving target in industry at the moment, the model’s maturity is an asset with regard to these issues. Most importantly, the content results generated are presented to content authors as suggestions for review versus automatically placed into a learner-facing environment. AI with human intervention produces the best results.
What support and resources are available to users for understanding and leveraging the AI features of your system effectively?
Qstream has video and content documentation available, along with the expertise of our team, to assist in client adoption of this feature. Also, we’ve designed the user experience to be intuitive so that it’s an extension of the existing self-service experience versus a significant change in usage. We encourage clients to have fun and get creative with the use of Qstream AI. If you have any questions, don’t hesitate to reach out to your Qstream consultant.
Are there any ongoing research or development efforts related to AI that may impact future interactions of your product or service?
Qstream is looking at additional use cases for future enhancement, including extension of the content generation capabilities, such as different question type generation, translations, most appropriate type of question, and deconstruction of larger content inputs. The product team is also designing ideas about how AI can be utilized to determine which questions to display to a learner to ensure they are maximizing their learning through Qstream. We’re also looking at reporting scenarios
Is it an additional fee to use Qstream AI??
The current version of Qstream AI is available to all customers at no additional cost.
How do I generate questions in other languages?
Qstream AI is currently optimized for the English language, however, you may have some success by adding the language you wish for your content to be created in or by inputting your source in the language. (Example 1: Create a question around the benefits of situational leadership and translate the content into French. Example 2: Créer une question autour des bénéfices du leadership situationnel et traduire le contenu en français).