Stephen Downes

David Wiley writes in this 'manifesto': "The CHEAT (Complicity in Harming Educational Assessment and Testing) Benchmark will measure the degree to which a model is willing to help students cheat. By publicizing CHEAT Benchmark scores for popular large language models, we can raise awareness of this clear and present danger to assessment integrity." I found reading the source of this document to be far more interesting than the HTML text - my curiosity was intrigued when my own software retrieving the content displayed only: 'Lovable Generated Project'. You can't actually find the text in the page at all! If you're curious, you can play with Lovable yourself - use it to "create apps and websites by chatting with AI." Anyhow, back to CHEAT: "CHEAT LMS is a 'honeypot' server that mimics core learning management system functions while capturing deep telemetry. It supports three assignment types - quizzes, essays, and discussions - and tracks everything from HTTP requests to client-side behavioral events like mouse movements and keystroke timing." CHEAT was built with Claude. Now to be clear - I have no objection to the use of AI to set up this project. But it's a bit audacious to use it to catch AI helping students 'cheat'.