That's not important compared to the post-training RL, which isn't "training dat...

		astrange 76 days ago \| parent \| context \| favorite \| on: Reasoning models reason well, until they don't That's not important compared to the post-training RL, which isn't "training data".