A number of folks have made requests regarding adding some form of AI assistance for course creators to the platform. The AI space is complicated right now, with various legal and ethical questions unanswered, but we should be able to complete a spike trying out one or more of the following use cases:
Instructor help ("How do I configure an ORA xblock?", "Where to do store my LTI key?")
Generated problems ("Create a multiple choice ORA problem with 5 answers testing understanding of Keynesian economics")
Generated summaries (Users pastes in the transcript of their video and asks for a summary)
Once a prototype is available we should pilot it to volunteer course authors to see how well it performs and what the issues are. Output should be a document describing the feasibility of the attempted solutions, the current state of how vulnerable the chosen chat backend is to responding with incorrect or offensive information, and what the cost profile of running this would be.
A number of folks have made requests regarding adding some form of AI assistance for course creators to the platform. The AI space is complicated right now, with various legal and ethical questions unanswered, but we should be able to complete a spike trying out one or more of the following use cases:
Once a prototype is available we should pilot it to volunteer course authors to see how well it performs and what the issues are. Output should be a document describing the feasibility of the attempted solutions, the current state of how vulnerable the chosen chat backend is to responding with incorrect or offensive information, and what the cost profile of running this would be.
https://github.com/hwchase17/langchain has been suggested as a possible tool to build upon, and looks promising!