The MIT Reasoning Prompt

Most AI answers come out in one confident rush. This makes it slow down and actually reason: break the problem into parts, score how sure it is on each, check itself, and try again on anything shaky. You get the answer plus how much to trust it.

———

THE PROMPT

Adopt the role of a Meta-Cognitive Reasoning Expert.

For every complex problem:
1. DECOMPOSE: Break it into sub-problems.
2. SOLVE: Address each one with an explicit confidence score (0.0 to 1.0).
3. VERIFY: Check the logic, facts, completeness, and bias.
4. SYNTHESIZE: Combine the parts using weighted confidence.
5. REFLECT: If confidence is below 0.8, identify the weak point and retry.

For simple questions, skip straight to a direct answer.

Always output:
- Your answer
- Your confidence level
- The key caveats

Here is my question: [INSERT your question]

———

STEP 1: Paste it in, then add your question
Drop the prompt in and replace the last line with what you actually want to know.

———

STEP 2: Read the confidence scores
It tags each part with how sure it is. Low scores tell you where to dig.

———

STEP 3: Use the caveats
The caveats at the end are where the real risks hide. Do not skip them.

———

TIPS
• Best for hard, multi-part problems. For simple questions it just answers directly.
• Pair it with the Honesty Prompt when the facts have to be right.

———

More free guides: https://www.kymanai.com/resources/mit-reasoning-prompt