Webb, Holyoak & Lu (2023)

View original ↗

At a glance

Tests
Analogies (Reasoning): seeing that one relationship matches another: A is to B as C is to what?
How
Four-term verbal analogies from psychology studies, and letter strings: abc changes to abd, so what does ijk change to?
Best AI
3 of 4
Humans
—
Our items
3
Reasoning gap, overall
-26.7

Example item

A specimen written for this page. Live items are similar but never shown, so posted answers can't spoil them.

Complete the analogyspecimen · not scored

Pick the word that completes the analogy.

(a) Question

  1. 1. Sculptor is to statue as baker is to …
    (a) oven(b) bread(c) flour(d) kitchen
Figure 1. A four-term verbal analogy. The wrong options are all linked to baking; only one matches the maker-to-product relationship. Benchmark: Webb, Holyoak & Lu (2023). Answer: Bread: a sculptor makes a statue, a baker makes bread.

Human vs AI

What the benchmark reports, and where the faculty stands overall

Best published AI

3 of 4

verbal analogy sets where GPT-3 matched or beat people

Webb, Holyoak & Lu, Nature Human Behaviour, 2023 ↗

Humans

—

This benchmark doesn't report a human figure.

Reasoning overall · faculty reference, 0–100

Measured on other tests, not on Webb, Holyoak & Lu (2023).

Human ref.
69.7
Best AI
96.4

Human: GPQA Diamond, PhD experts. AI: GPQA Diamond, Gemini 4 Argon (Epoch AI).

How we mirror it

New verbal analogies whose wrong options share a surface association with the words (odometer and speed), so only the relationship gives the answer, plus new letter-string analogies.

Each item has two questions; both must be right.

Items are original, versioned, and assigned at random, so a posted answer spoils only one variant.

What we record

each answerresponse timepastestab switches