> The fact that scaled reasoning models are finally showing progress on ARC proves that what it measures really is relevant and important for reasoning. One mi

> The fact that scaled reasoning models are finally showing progress on ARC proves that what it measures really is relevant and important for reasoning. One might also interpret that as "the fact that models which are studying to the test are getting better at the test" (Goodhart's law), not that they're actually reasoning.