Do you agree with the machine?

Three AI models were asked twenty questions about everyday technology and online privacy. Another model marked their answers. You are going to mark five of the same answers, and then find out where you and the machine disagreed.

It takes about five minutes. You are not being tested: the interesting result is where a person and a model look at identical text and reach different conclusions, and that only shows up if you score what you actually think.

What is collected This is a research exercise and the data is only used for that. It is collected with nowhere to put anything personal:

No name, no email, no account, no sign-in.
No IP address, no analytics, no advertising, nothing from a third party.
No free text anywhere, so there is no box that could hold something about you.
Stored: your five sets of scores, the experience answer below, and a random number generated in your browser so the five can be grouped together.

One question first

Only so the results can be reported separately for people who work in this area and people who do not. There is no wrong answer and nothing is verified.

1 person has done this so far, leaving 24 scores.

Part of an eval harness built by Dale Mooney. The reasoning behind it is in what the judge does instead of judging.