Patrones de razonamiento
Definición
Los patrones de razonamiento son formas estructuradas de provocar u organizar el razonamiento del modelo: chain-of-thought (step-by-step), tree-of-thoughts (explore branches), ReAct (reason + act), and RDD (recuperación-decisión-diseño), among others. Using a clear pattern improves reliability (more consistent razonamiento) and debuggability (you can inspect steps or actions).
Son used in prompt engineering (por ej. CoT) and inside agents (por ej. ReAct, RDD). Choosing a pattern depends on the task: CoT for math/razonamiento, ReAct for tool use, ToT for search/planning, RDD for spec compliance.
Cómo funciona
You feed input (question, task) into a pattern: the pattern constrains how the model reasons or acts (por ej. “think paso a paso”, or thought–action–observation loops). The model produce an output (answer, action sequence). Prompts or system diseño encourage the model to show razonamiento (por ej. “Think paso a paso”) or to interleave thought and action. Patterns can be combined (por ej. CoT inside an agent loop). See the linked pages for each pattern’s details.
Casos de uso
Different patterns suit different needs: CoT for stepwise razonamiento, ReAct for tool use, ToT for search and planning.
- CoT: math, logic, and multi-step razonamiento tasks
- ReAct: tool-using agents that reason before each action
- ToT: search and planning over multiple solution branches
Documentación externa
- Chain-of-Thought Prompting (Wei et al.) — CoT paper
- ReAct: Synergizing Reasoning and Acting (Yao et al.) — ReAct paper
- Tree of Thoughts (Yao et al.) — ToT paper