> Like giving GPT a multiple-choice test on gibberish—it will pick something, and say it with its chest.
If I gave a classroom of under grad students a multiple choice test where no answers were correct, I can almost guarantee almost all the tests would be filled out.
Should GPT and other LLMs refuse to take a test?
In my experience it will answer with the closest answer, even if none of the options are even remotely correct.
Just on a practical level, adding a way for the LLM to bail if can detect that things are going wrong saves a lot of trouble. Especially if you are constraining the inference. You still get some false negatives and false positives, of course, but giving the option to say "something else" and explain can save you a lot of headaches when you accidentally send it down the wrong path entirely.
Not refuse, but remark that none of the answers seems correct. After all, we are only 2 days away from an AGI pro gamer researcher AI according to experts[1], so I would expect this behavior at least.
1: People who have a financial stake in the AI hype
In multiple choice if you don't know then a random guess is your best answer in most cases. In a few tests blank is scored better than wrong but that is rare and the professors will tell you.
as such I would expect students to but in something. However after class they would talk about how bad they think they did because they are all self aware enough to know where they guessed.
I would love to see someone try this. I would guess 85-90% of undergrads would fill out the whole test, but not everyone. There are some people who believe the things that they know.
I think the issue is the confidence with which it lies to you.
A good analogy would be if someone claimed to be a doctor and when I asked if I should eat lead or tin for my health they said “Tin because it’s good for your complexion”.
If I gave a classroom of under grad students a multiple choice test where no answers were correct, I can almost guarantee almost all the tests would be filled out.
Should GPT and other LLMs refuse to take a test?
In my experience it will answer with the closest answer, even if none of the options are even remotely correct.