Why does submitting the same answer sometimes result in different scores?
Last updated August 17, 2024
The underlying AI models Class Companion uses are based on OpenAI's GPT-4 . While GPT-4 is the most advanced Large Language Model currently available, there is still variability in the responses. This is the core reason Class Companion cannot ensure perfect consistency.
To contrast with non-AI alternatives, this is somewhat similar to how two different human graders may not provide the exact same feedback. Even a single grader may have variability in scores and feedback if they were to evaluate complex criteria on the same essay months apart.
We do, however, strive for high quality feedback with each submission, and if you're seeing major differences in how the AI graded the same criteria, we would love if you could share your particular examples so we could train the AI to avoid such mistakes. To do so, please contact support and include the assignment name, student, and the rubric.