🔬Thinking Inertia: LLMs Keep Reasoning When Told to Stop
TL;DR
Researchers tested six no-think prompts on six LLMs and found visible reasoning persists. The effect grows as the answer space opens up.
Researchers tested six no-think prompts on six LLMs and found visible reasoning persists. The effect grows as the answer space opens up.
Key Points
Authors include Philip Torr and James Zou; submitted October 8, 2026
Measured by Empty-Thinking Rate, question relevance and an LLM-judge Explicit Inference Rate
Accuracy holds on Boolean and multiple-choice tasks
Open-ended tasks show a trade-off between answer-only compliance and accuracy
Supplying candidate answers makes answer-only output easier
Why It Matters
If you rely on a no-think instruction to cut latency or tokens, it may not hold on open-ended prompts. Test compliance, not just accuracy.
Quick Facts
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,568 builders reading daily.