Chimera, The Complete Book · twelve questions · four arms · one blind judge

The Three-Body Test

Is an AI's plain first answer a draft to keep, or a body as heavy as the question and the Book? Twelve questions, each answered four ways, two answers per way. A blind judge scored all eight answers to each question without knowing which way made them.

  • the question
  • the first answer (C)
  • the Book slice
  • E's score where they cross
  • D's score

Three equal masses chasing one another around a single figure-eight: a real periodic orbit of the three-body problem, found by Cristopher Moore in 1993 and proved by Alain Chenciner and Richard Montgomery in 2000, computed in your browser. Here the bodies are E's three inputs. The orbit is a picture of the idea, not a measurement; the glow where the bodies cross and the dashed ring are the judge's scores for the question you pick below.

Four ways to answer, two answers each

  1. ABook first

    The Book slice, then the question.

  2. CQuestion alone

    No Book. This plain answer is the first answer that D and E start from.

  3. DRevise

    The first answer and the slice, told to revise it: keep what it gets right, keep it plain, add from the notes only where that truly serves.

  4. EThree bodiesnew

    The question, the first answer and the slice, told to reason on all three together, then write the best answer.

What came out measured

One question's eight answers

Each dot is one answer, numbered by its run; the thin bar marks the arm's mean. Tap or hover a dot for the judge's scores.

Measurements measured

Score is the mean of the judge's 1 to 10 dimensions. Rank is the judge's mean place among the eight answers to a question, 1 is best.

The prediction

Written into the test's code and committed before any judge call, then scored exactly as written.

Mo's reading claims, not measurements

    The caveats