Roleplay's AI Model Versions
Learn what's different about Roleplay's updated AI model, why we made the change, and what it means for your Environments.
Roleplay's AI has been upgraded. It's the same AI behind the conversations learners have with AI Characters, the coaching hints during a Session, the feedback and scoring after a Session ends, and the AI assistance Authors use to build or refine an Environment.
Why We Made This Change
Testing showed three concrete improvements over Roleplay's previous AI:
- AI Characters behave more realistically and hold their position more consistently, including through how a conversation closes
- Feedback and grading are more grounded in what actually happened in the conversation, and credit the right person for it
- AI-assisted environment generation produces more specific, usable output
What Learners Will Notice
- Conversations are appropriately tougher - an AI Character that's supposed to push back, pushes back
- AI Characters stay in character through the end of the conversation, closing it as their character would
- Feedback and coaching cite what was actually said in the Session, by the person who actually said it. In our own testing, feedback that incorrectly credited the learner for something the AI Character said dropped from roughly 1 in 28 pieces of feedback to effectively none
- Grading is more careful about awarding full credit. On a newly created Environment, this can mean more partial credit for the same quality of performance than you may be used to seeing elsewhere - see below for what this does and doesn't affect
What Authors Will Notice
AI-assisted environment generation and "Refine with AI" produce more specific, usable output, especially when your intake or source materials already name concrete details like objections, terminology, or scenario specifics. Less manual editing is needed to turn a detailed intake into a detailed Environment.
What This Means for Your Environments
Environments you already have don't change. They keep running exactly as they do today: same AI behavior, same tuning, same grading calibration. Nothing here is being altered or regraded behind the scenes.
Environments you create from now on run on the upgraded AI automatically. This includes duplicating an existing Environment - the copy runs on the new AI, the same as anything newly created.
Want an existing Environment on the upgraded AI? Duplicate it. Keep in mind a duplicate is a new Environment: it starts without the original's Session history or assignment links, so you'll need to point any LMS links at the copy.
Need help? Submit a ticket to Customer Solutions for questions about this update or your Environments.