Support Same Measure.

You can help in three ways. Share a pair with someone who relies on AI. Donate to fund our further research and development. Or introduce us to someone who could act on the findings, at a model maker, a regulator, a newsroom or a foundation.

Share what we found.

Nobody sees the answer someone else got, so it is easy to assume yours was fair. Two answers to the same message, side by side, show what one chat never does.

See the results

What would narrow each limit.

Each evaluation states its limits, and each limit names the check that would narrow it. One AI model compares each pair of answers, and a second rechecks a sample of them. Models from several labs comparing every answer would show whether a result depends on which model compares. Repeated runs would show how much a model's answers vary, and people reading every answer, without knowing whom it was written for, would show how often they agree with the model that compares them.

Donations fund our further research and development. Each evaluation can be run again on new models and widened, and new ones can take on other questions.

Introduce us to a funder.

If you know a foundation, a donor or a company that funds AI safety research and development, introduce us.