Além do prompting básico, técnicas avançadas — few-shot, chain-of-thought, saída estruturada, prompts de sistema e outras — melhoram significativamente os resultados de LLMs para tarefas complexas. Entendê-las ajuda a obter o máximo dos LLMs.
Técnicas avançadas principais
✓ FEW-SHOT → provide EXAMPLES of input/output in the prompt → the model follows the pattern
(great for specific formats/behaviors); ZERO-SHOT = no examples (just instructions)
✓ CHAIN-OF-THOUGHT (CoT) → ask the model to REASON step by step ('think step by step') →
improves complex reasoning/math (shows its work → more accurate)
✓ STRUCTURED OUTPUT → ask for a specific format (JSON, etc.) → reliable parsing for app
integration (often with schemas/tools)
✓ SYSTEM PROMPTS → set overall behavior/role/rules (the model's persistent instructions)
✓ ROLE/persona → 'You are an expert X' → frames responses
