The Decreasing Value of Chain of Thought in Prompting - Mollick

Appears that asking for direct answers generally decreased performance across all non-reasoning models.