The ARC test does not address factuality, compositionality, or common sense.

Was it true for the period described?

Outcome: Mixed

Part of the claim holds and part does not, and the record says which part is which.

Scored on a plain reading. The speaker named no measure for this claim, so it was judged on the reading an informed reader would take, stated below under our interpretation. Both reviews had to find that reading fair.

Evidence as of

2026-10-28

Deadline state when assessed

No horizon

Assessment recorded

2026-10-29

Policy version

2026-10-28.7

Rationale

Factuality: supported. ARC Prize says ARC-AGI restricts itself to core knowledge priors and avoids learned or cultural knowledge, so facts about the world are not tested. Compositionality and common sense: the claim that they 'aren't even addressed' goes too far. Chollet's 2019 paper builds ARC on core-knowledge priors, including 'objectness and elementary physics', which are basic commonsense intuitions. The ARC-AGI-2 paper says ARC-AGI-2 targets 'deeper levels' of compositional generalization. That implies ARC-AGI-1 covered combining rules shallowly, though most of its tasks needed a single transformation. So ARC covers these two only thinly, not 'not at all'.

Limitations

• The result depends on how 'common sense' and 'compositionality' are defined. A narrower reading, such as broad everyday world knowledge, would favor Marcus.
• The ARC-AGI-2 paper was published after the post. It describes the earlier benchmark's design, which did not change.
• The ARC Prize design page is undated, so its available_at is set to the date it was read.

Evidence (3 items)

• Supports the claim: ARC Prize Foundation, 2019-11

  ARC-AGI avoids this by restricting itself to core knowledge priors, those cognitive building blocks that are either present at birth or acquired very early in human development with minimal explicit instruction

  Why it matters here: ARC deliberately avoids acquired and cultural knowledge, so factual knowledge is not tested. This supports the factuality part.

  Where

  section Core Knowledge Priors, paragraph 2

  Publisher

  ARC Prize Foundation

  Kind of source

  Company disclosure: the company describing itself, not an independent finding

  Describes events of

  2019-11

  Public from

  2026-10-28

• Goes against the claim: arXiv (François Chollet), 2019-11

  Objectness and elementary physics: humans assume that their environment should be parsed into “objects” characterized by principles of cohesion

  Why it matters here: ARC is built on core-knowledge priors such as objectness and elementary physics, which are basic commonsense intuitions. That cuts against 'common sense aren't even addressed', though only the most basic layer is covered.

  Where

  Section II.3.1, Core Knowledge list, first bullet (p. 47)

  Publisher

  arXiv (François Chollet)

  Kind of source

  Preprint, not peer reviewed

  Describes events of

  2019-11

  Public from

  2019-11-05

• Goes against the claim: arXiv (Chollet, Knoop, Kamradt, Landers, Pinkard), 2019-11

  Lastly, a major focus with ARC-AGI-2 is to investigate deeper levels of compositional generalization — the ability to combine known rules or concepts in novel ways.

  Why it matters here: The ARC authors describe ARC-AGI-2 as probing 'deeper levels' of compositional generalization than the 2019 ARC-AGI-1 design. That implies ARC-AGI-1 covered shallower levels, not none. The same section says most ARC-AGI-1 tasks needed only one high-level transformation, so coverage was weak.

  Where

  Section 7.1, final paragraph

  Publisher

  arXiv (Chollet, Knoop, Kamradt, Landers, Pinkard)

  Kind of source

  Preprint, not peer reviewed

  Describes events of

  2019-11

  Public from

  2025-05-17