What is ARC?
The Abstraction and Reasoning Corpus (ARC) is a collection of tasks designed to measure a system's or individual's intelligence, especially their ability to perform abstraction and reasoning. Each task in ARC consists of a few input-output examples, followed by one or more test inputs. It's up to a human or AI to predict the correct outputs.
Constants and Variables. Notice what stays the same across examples. Notice what changes across examples. This might be the shape, color, number of elements, arrangement, rotation, resizing, flipping, or repetition.
Generalize the Rules. Based on observations, hypothesize a set of rules that can transform the given inputs into their corresponding outputs. Sometimes, rules can conflict, or one rule needs to be applied before another. Determine the hierarchy or order of rule application if necessary.
Apply to Test Input. Use the hypothesized rules to transform the test input into the output. This is the critical test of whether the identified patterns and rules are correct. Check for overlooked patterns or incorrect assumptions. Double-check that the rules work consistently across all given examples and all the tests.