When Not to Use DeepSeek-R1 for Simple Queries

Thoughtology experiments observed that when presented with simple control sentences, DeepSeek-R1 can exhibit non-humanlike over-deliberation, and SambaNova advises applying R1 only where its deeper holistic reasoning adds value. Route simple zero-shot sentiment or intent lookups to DeepSeek-V3.

DeepSeek-V3 Zero-Shot Prompt (temp=0.1): Classify the sentiment of the following customer feedback as Positive, Neutral, or Negative: "{{FEEDBACK}}"