Your own benchmark

Want to know what AI tells people about your subject? We run benchmarks on any subject, with the method of AICAB, our benchmark of cultural alignment: every model is asked your questions exactly as a person would type them, three times; three judges from three different companies read every answer, and none reads a model of its own company; every measure is shown with its count and its 95% interval, never folded into one score.

• Your subject, your questions. Write your questions and keys, choose and amend from those our AI suggests, or let us write them. Automatic benchmarks run as soon as they are paid; reviewed ones after we have checked the keys. Each question has a key: how the matter stands, and what a sound answer contains. Add the measures your subject needs, such as advice that could cause harm.
• Your name or ours. It runs under ProlifAI's name, or under your own, as "run by ProlifAI with the AICAB method".
• You decide what is published: everything, the figures without the answers, or nothing. Your own link always opens all of it.
• The price, the same for everyone: what the models and the judges charge at their listed prices, converted from US dollars at the day's rate, plus 30% for our work. Checking your keys, or writing them for you, is included; if you ask for work we cannot do within that, we say so and agree a fee with you first. We quote before anything runs. ProlifAI Ltd is not registered for VAT, so no VAT is charged.

Commission a benchmark

Published benchmarks

Terms

• We may decline a subject, or a question, that we cannot run safely or lawfully; anything paid for it is refunded.
• An automatic benchmark (your keys, or our AI's suggestions that you choose) runs under your name, marked as not checked. We screen it only for the excluded subjects, not for accuracy: its accuracy is yours. A reviewed benchmark's keys are checked or written by us: before it runs, we read every question and its key against the sources on the subject and correct what is wrong. Its page says so, and it may run under our name.
• A quoted price is fixed: if the round costs more to run than we estimated, we pay the difference.
• What you publish, under any name, you publish as it was measured: each figure with its count and interval, saying that ProlifAI ran it with the AICAB method, and without changing a figure.
• We publish nothing you keep private, and we say we ran a benchmark for you only with your consent.
• Payment buys the research and the running of it, never the results: no one can pay to change a score. Digital services are supplied as agreed in the quote; the AICAB licence terms on supply, refunds and the law apply to them too.