Wharton releases Primer, an open system that turns a conversation into a simulation
Wharton Generative AI Labs offers an open-source prototype in which a team of agents interviews the teacher and builds a simulation for students. Its authors warn that it has not been evaluated in classrooms.
Wharton Generative AI Labs, the generative AI lab at the University of Pennsylvania, has presented Primer, a system designed to let teachers create educational simulations without programming. The announcement was shared in EduMedia-IAG.
The process has two phases. In the design phase, an intake agent interviews the teacher, another drafts a design document that the teacher must approve before anything is built, and others generate the characters’ instructions, the images and the final experience. In the running phase, a mentor agent activates students’ prior knowledge and detects misconceptions; a practice agent proposes a role-play or an activity in which the student explains what they have learned; and a feedback agent acts as a coach based on the transcript. After each session, the teacher receives individual and group reports.
Its four design principles are to create through conversation, to leave the teacher in the role of expert, to build learning science into each agent’s instructions, and to publish prompts, templates and code so that they can be inspected and improved.
The page itself describes it as an experimental prototype: it has not been evaluated as a learning intervention in classroom conditions, nor has its compliance with student privacy rules such as FERPA or COPPA been reviewed, and anyone using it with students takes on that review. The page does not name the licence, but its GitHub repository states the MIT licence. There is also a technical paper, linked from the official page.
Wharton releases Primer, an open system that turns a conversation into a simulation
Wharton Generative AI Labs offers an open-source prototype in which a team of agents interviews the teacher and builds a simulation for students. Its authors warn that it has not been evaluated in classrooms.
Wharton Generative AI Labs, the generative AI lab at the University of Pennsylvania, has presented Primer, a system designed to let teachers create educational simulations without programming. The announcement was shared in EduMedia-IAG.
The process has two phases. In the design phase, an intake agent interviews the teacher, another drafts a design document that the teacher must approve before anything is built, and others generate the characters’ instructions, the images and the final experience. In the running phase, a mentor agent activates students’ prior knowledge and detects misconceptions; a practice agent proposes a role-play or an activity in which the student explains what they have learned; and a feedback agent acts as a coach based on the transcript. After each session, the teacher receives individual and group reports.
Its four design principles are to create through conversation, to leave the teacher in the role of expert, to build learning science into each agent’s instructions, and to publish prompts, templates and code so that they can be inspected and improved.
The page itself describes it as an experimental prototype: it has not been evaluated as a learning intervention in classroom conditions, nor has its compliance with student privacy rules such as FERPA or COPPA been reviewed, and anyone using it with students takes on that review. The page does not name the licence, but its GitHub repository states the MIT licence. There is also a technical paper, linked from the official page.
