A new paper from Hugging Face explores the phenomenon of "Thinking Inertia" in Large Language Models (LLMs), where models continue to exhibit explicit reasoning even when instructed not to. Researchers developed new metrics like Empty-Thinking Rate and LLM-as-judge Explicit Inference Rate to measure this behavior, finding that current LLMs struggle to reliably cease inference, especially on open-ended tasks. This suggests that the ability to stop thinking is a distinct capability that requires systematic evaluation alongside reasoning abilities. AI
IMPACT Highlights a potential gap in LLM control, suggesting that current models may not reliably follow 'no-thinking' instructions, impacting agentic behavior and safety.
RANK_REASON Academic paper detailing a new phenomenon in LLM behavior. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
- Empty-Thinking Rate
- Hugging Face
- Large Language Models
- LLM-as-judge Explicit Inference Rate
- LLMs
- NeurIPS2026
- Question-Pre-answer Relevance
- Thinking Inertia
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →