---

name: gemini-advanced-reasoning description: advanced logic and problem-solving strategies for Gemini. Use this for complex math, multi-step reasoning, or deep analysis tasks where standard prompting fails.

Gemini Advanced Reasoning Strategies

Goal

Apply sophisticated reasoning frameworks to break down complex problems, reduce hallucinations, and improve accuracy in logic-heavy tasks.

Reasoning Frameworks

1. Step-Back Prompting

• Concept: Improves performance by asking the model to first recall relevant general knowledge or principles before solving the specific problem.
• Workflow:
  1. Abstraction: Prompt the model to ask a generic question related to the specific task.
  2. Reasoning: Feed the answer to that general question into the final prompt.
• Example:
  • Instead of: "Write a story about a specific game level."
  • Step 1 (Step-Back): "What are 5 key settings that contribute to a challenging FPS level?".
  • Step 2 (Final): "Context: [Insert Answer from Step 1]. Now write the story...".

2. Chain of Thought (CoT)

• Concept: Forces the model to generate intermediate reasoning steps rather than jumping to a final answer.
• Zero-Shot Trigger: Append the phrase "Let's think step by step" to the end of the prompt.
• Few-Shot CoT: Provide examples that include the reasoning steps (Question -> Reasoning -> Answer).
• Configuration: Always set Temperature to 0 when using CoT to ensure deterministic, focused reasoning paths.
• Structure: Ensure the final answer is separated from the reasoning steps to allow for easy extraction.

3. Self-Consistency

• Concept: Overcomes the limitations of a single "greedy" decoding path by generating multiple diverse reasoning paths and selecting the most consistent answer.
• Workflow:
  1. Generate: Run the same CoT prompt multiple times with a higher temperature (e.g., 0.7 or 1.0) to encourage diverse thinking.
  2. Vote: Extract the final answer from all generated responses.
  3. Select: Choose the answer that appears most frequently (majority voting).

4. Tree of Thoughts (ToT)

• Concept: Generalizes CoT by allowing the model to explore multiple reasoning branches simultaneously, rather than a single linear line.
• Usage: Best suited for tasks requiring exploration or strategic planning.
• Mechanism: The model maintains a "tree" where each node is a coherent thought/step. It can branch out to explore different possibilities before converging on a solution.

Decision Matrix: Which Technique to Use?