Mitigating Overthinking in Large Reasoning Models via Manifold Steering

Open in new window