FAQ

No. HIBN gives you a score on a fixed set of questions. Run the same check again later, with the same setup, and it shows the change in your score when the two checks can be compared. It can’t show that an AI was changed, or why a score moved. Methodology: limits

No. A score shows how the AI did on these questions this time. HIBN has no pass mark, so it doesn’t call any score good or bad. A Quick or Deep score covers 6 or 10 questions and says nothing about writing, coding or the other things you use your AI for.

HIBN can’t tell you. One Quick question is worth up to 17 points, and an AI can answer the same question differently on two attempts. Something may also have differed on your side or on the provider’s side, and HIBN can’t tell which. More comparable checks show whether the lower score repeats. They can’t show why. Methodology: comparing two checks

No. The same score means the question scores added up to the same total, to the nearest point, both times. It doesn’t show that nothing changed, and it covers these questions only.

Because at least one answer is missing or couldn’t be scored. We count a missing answer as zero only if you tell us the AI refused or didn’t answer. To get a score, run a new check and paste the whole reply. Methodology: when a check has no score

HIBN compares two checks only when they used the same questions, the same scoring code and the same setup. Nothing in the setup can be Unknown or Other, and both checks must have a score. Where two results can’t be compared, HIBN gives the reason. Methodology: comparing two checks

HIBN doesn’t know. It has not measured which is harder. Deep has 10 questions and Quick has 6, and Deep’s are about twice as long on average. Quick and Deep scores are never compared with each other.

No. Analytics shows how many Quick and Deep checks got a score, in total and by country. For a setup with at least 20 checks and nothing set to Unknown, it also shows the median score (the middle score). People chose to run those checks and reported their own setups. It is not a ranking. Methodology: how Analytics is produced

Whichever you paste the questions into. HIBN doesn’t connect to any AI. You choose the provider, app, model and level, and HIBN records your choices.

No. HIBN sees only the reply you paste. The setup is what you choose, and HIBN can’t confirm it.

Choose Unknown. A wrong guess would record the wrong setup with your result. The cost is that a check with Unknown or Other in its setup can’t be compared with other checks.

If an answer is missing from the reply, the check page asks what happened to it. Tell us there. We record those answers as “Refused (you told us)” or “Not answered (you told us)”, and each scores 0. If you leave them as missing, the check has no overall score. A refusal that the AI wrote under a question’s label is scored like any other answer.

HIBN finds answers by their labels, and the reply has to be pasted unchanged. Leave the answers as missing and run a new check.

From LiveBench, a public benchmark for AI language models. HIBN uses 16 of its questions from 2024, with its own instructions added at the top. It removed a request to reason step by step, and fixed how line breaks and one maths command are written. It changed no numbers and no correct answers. HIBN is not affiliated with LiveBench. About LiveBench

Yes. The questions have been public since 2024, so an AI may have seen them, or their answers, in training. A high score may partly reflect that. About LiveBench

No. HIBN shows the outcome of each question. The correct answers are not secret: they can be worked out from LiveBench’s public files. About LiveBench

No. The expected answers come from the pack’s author. HIBN scores replies against them and doesn’t confirm that they are right. Pack guide

Not yet. Every check works by copy and paste.

No. You can run checks and open your results without one. We store your results on our servers and link them to a random key in a cookie in your browser. The key lasts 90 days from the day it is made, and new checks don’t extend it. To keep a result for longer, or to open it on another device, save it to an account with Google sign-in.

No other visitor can. A result opens only in the browser that ran the check, or in the account it is saved to. Quick and Deep checks with a score are counted in public Analytics, which shows totals and medians and never an individual result. Privacy & data

We count checks by country in Analytics. Your city is stored with your result and never published. Your location is estimated from your connection. On the home page or a pack’s page you can change it or choose not to say, and then we store only your choice. “Check again (same setup)” uses the same choice as the check it repeats. Privacy & data

Results saved to an account have no set end date. Other results are linked to your browser’s key, which lasts 90 days from the day it is made, and after that they can’t be opened. A daily clean-up then removes the reply, the detail of each question, the city and anything you typed for Other. We keep the score, the rest of the setup, the country and the dates, with no set end date. A Quick or Deep check with a score still counts in Analytics. Privacy & data

Open a result and choose “Delete result”. To delete a check you didn’t finish, choose “Delete” beside it in My Results. To download your data first, use “Download my data” in My Results. To have your account or your display name deleted, write to privacy@haveibeennerfed.ai. We can delete only what we can match to you. Privacy & data

BEMA Labs Pty Ltd, a company registered in Australia. About

Yes. HIBN is free to use during the public beta.

Yes. Say how many checks a number is based on, or the range Analytics shows, and the dates it covers. Say that people chose to take part and reported their own setups. A HIBN number can’t show that an AI was changed. About