Loading…
26
Unpopular opinion: I got 41 wrong answers out of 100 letting an AI grade my team's safety quizzes
I built a little AI grader in March to score our weekly toolbox talk quizzes, and it marked 41 of 100 answers wrong that were actually right. Turned out it kept flagging any answer with the word 'never' because it read the word as a trick, even when the question asked for the safest option. Has anyone found a way to pin down what an AI grader is actually keying on, or do you just reread every score it gives?
1 comments
Log in to join the discussion
Log In1 Comment
finleyl5520d ago
Could you test it by feeding it the same quiz twice but swap "never" for "always" in the right answers and see if the score flips? That would tell you if it's really just keying on one word instead of reading the whole question. Sounds like your grader is pattern matching on keywords, not actually understanding what the question is asking.
1