Why reasoning modes often make summaries less accurate
https://www.4shared.com/office/b9DVb_-fku/pdf-35982-5647.html
Why developers and editors see worse summaries when models "reason" People try to make language models smarter by asking them to show their work