Skeleton-of-Thought is a latency optimization technique where the model first generates an outline of the answer, then expands each section in parallel.
Technique to speed up long-form generation.
Reduces user wait time for complex reasoned responses.