Machine Learning Research
Ordinary LLMs Implicitly Take Reasoning Steps: Anthropic experiment finds Claude shows signs of unprompted reasoning
Even without explicit training in reasoning, large language models “think” in ways that may be more deliberate than previously understood.