Padrões de raciocínio
Definição
Padrões de raciocínio são formas estruturadas de provocar ou organizar o raciocínio do modelo: chain-of-thought (step-by-step), tree-of-thoughts (explore branches), ReAct (reason + act), and RDD (recuperação-decisão-projeto), among others. Using a clear pattern improves reliability (more consistent raciocínio) and debuggability (you can inspect steps or actions).
Eles são used in prompt engineering (por ex. CoT) and inside agents (por ex. ReAct, RDD). Choosing a pattern depends on the task: CoT for math/raciocínio, ReAct for tool use, ToT for search/planning, RDD for spec compliance.
Como funciona
You feed input (question, task) into a pattern: the pattern constrains how the model reasons or acts (por ex. “think passo a passo”, or thought–action–observation loops). The model produces an output (answer, action sequence). Prompts or system projeto encourage the model to show raciocínio (por ex. “Think passo a passo”) or to interleave thought and action. Patterns can be combined (por ex. 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 raciocínio, ReAct for tool use, ToT for search and planning.
- CoT: math, logic, and multi-step raciocínio tasks
- ReAct: tool-using agents that reason before each action
- ToT: search and planning over multiple solution branches
Documentação 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