Advanced Techniques
The ten advanced techniques are designed for complex, multi-step, or high-stakes tasks where foundational techniques alone are insufficient. These require more careful prompt engineering, more context budget, and often more iterative refinement — but they unlock capabilities that simpler techniques cannot reach.
Who these are for: Practitioners building LLM-powered products, researchers evaluating model behaviour, and engineers designing prompt pipelines for production systems.
Techniques in This Section
- Self-Consistency — Sampling multiple reasoning paths and majority-voting the answer
- ReAct Prompting — Interleaving reasoning and action in an agentic loop
- Step-Back Prompting — Reasoning from first principles before tackling specifics
- Prompt Chaining — Breaking long tasks into sequential, dependent prompts
- Least-to-Most Decomposition — Solving the simplest sub-problem first, building to the hardest
- Generated Knowledge — Prompting the model to generate useful context before answering
- Calibration Prompts — Eliciting confidence estimates and surfacing uncertainty
- Directional Stimulus — Using hints or keywords to steer response direction
- Meta-Prompting — Using an LLM to write or improve prompts
- Evaluation Prompts — Using LLMs as judges to score or compare outputs
Advanced techniques consume more tokens, often require multiple API calls, and have more failure modes than foundational techniques. Always establish a foundational baseline before applying advanced methods.