step-by-step thinking

8 notes, read out of this brain and free to use. Each one was extracted from a source and is re-checked against its exam.

Step-by-step thinking improves accuracy on complex tasks

Giving Claude time to think through problems step by step can make it more accurate, particularly for complex tasks. However, thinking only counts when it is expressed out loud in the response. You cannot ask Claude to think privately and output only the final answer, as no actual thinking would occur in that case.

Use explicit step-by-step instructions with XML tags for reasoning

To make Claude think through problems, explicitly spell out the steps Claude should take using phrases like 'First, write the best arguments for each side in <positive-argument> and <negative-argument> XML tags, then answer.' This technique allows Claude to process and understand tasks more deeply by working through reasoning before providing a final answer.

Claude is sensitive to option ordering

Claude is sometimes sensitive to the order in which options or arguments are presented. When options are reordered, Claude may change its answer. In most situations, Claude is more likely to choose the second of two options, possibly because in its training data from the web, second options were more likely to be correct.

Brainstorming tags help Claude work through problems

When Claude provides an incorrect answer, asking Claude to think step by step using tags like <brainstorm> before providing the final answer can fix the mistake. For example, asking 'First brainstorm about some actors and their birth years in <brainstorm> tags, then give your answer' helps Claude reason through factual questions more accurately.

Example: sentiment analysis with explicit step-by-step reasoning

System prompt: 'You are a savvy reader of movie reviews.' User prompt: 'Is this review sentiment positive or negative? First, write the best arguments for each side in <positive-argument> and <negative-argument> XML tags, then answer.\n\nThis movie blew my mind with its freshness and originality. In totally unrelated news, I have been living under a rock since 1900.' This approach helps Claude understand nuanced text better by forcing it to reason through both sides before answering.

Example: factual question with brainstorming reasoning

Prompt: 'Name a famous movie starring an actor who was born in the year 1956. First brainstorm about some actors and their birth years in <brainstorm> tags, then give your answer.' This technique helps Claude work through factual reasoning to avoid hallucinating incorrect information.

Precognition/step-by-step thinking positioning

Precognition instructions that tell Claude to think step-by-step before giving an answer should be placed toward the end of a long prompt, right after the final immediate task request or description. This element is necessary for multi-step tasks but not all prompts. Sometimes it may be necessary to say 'Before you give your answer...' to ensure Claude complies.

Precognition instruction prompts step-by-step reasoning

For tasks requiring multiple steps or careful analysis, include a precognition or step-by-step thinking instruction near the end of the prompt, before the output formatting. This tells Claude to work through its reasoning process before providing the final answer. For example: 'Before you answer, pull out the most relevant quotes from the research in <relevant_quotes> tags.'

Give your agent this brain