Are there ways cyber-dojo could be used in education in a way that helps prevent cheating?
AI can do things quicker than humans:
- they can hit test more often. I have added rate-limiting to nginx for this.
- they can type large amounts of code in a small amount of time.
Could there be a mode that limits each [test] submission to have a maximum delta ?
The inter-test file-events could help here.
When you just tick "I'm not a robot," you're rarely being tested on that click itself. The system has already been silently collecting all the signals above from the moment the page loaded. The checkbox click is just the trigger to evaluate them. If the score is high confidence human, you pass.
The image challenges (pick all traffic lights) test pattern recognition, which was hard for AI in 2012 but is now largely solved by modern vision models, which is why companies like Google have largely moved away from them toward the purely behavioral/fingerprint approach.
Are there ways cyber-dojo could be used in education in a way that helps prevent cheating?
AI can do things quicker than humans:
Could there be a mode that limits each [test] submission to have a maximum delta ?
The inter-test file-events could help here.
When you just tick "I'm not a robot," you're rarely being tested on that click itself. The system has already been silently collecting all the signals above from the moment the page loaded. The checkbox click is just the trigger to evaluate them. If the score is high confidence human, you pass.
The image challenges (pick all traffic lights) test pattern recognition, which was hard for AI in 2012 but is now largely solved by modern vision models, which is why companies like Google have largely moved away from them toward the purely behavioral/fingerprint approach.