Home / Advanced techniques
Fig. 03 — Mechanisms

Advanced techniques

Beyond the basic structure, these six techniques improve reliability on long, ambiguous or high-stakes tasks.

01 — Chain of thought (CoT)

Reasoning step by step before concluding improves calculation, logic and multi-criteria analysis.

Think step by step before giving your final answer.

02 — In-context learning

One good example often beats a long paragraph. Start with zero or one; add more only if needed.

Input → expected output, then: your turn with this data.

03 — Self-consistency

Generate several independent reasoning paths and keep the conclusion that recurs most.

Give 3 distinct lines of reasoning, then say which converges.

04 — Prompt chaining

Split a task into steps, each reusing the previous output, to limit compounding errors.

Step 1: extract facts. Step 2: classify. Step 3: summarise.

05 — Meta-prompting

Describe the expected logical structure instead of piling up examples: fewer tokens, less bias.

Describe your structure (premises → rule → conclusion) before writing.

06 — Permission to be uncertain

Explicitly allowing “I don’t know” reduces hallucinations on incomplete data.

If the data is insufficient, say so instead of guessing.

Reasoning models: adapt the technique

Recent models reason natively and expose a reasoning-effort setting. There is no need to repeat “think step by step”: describe the success criteria instead, and tune effort to the task’s difficulty.