Post
10 Prompt Engineering Techniques for Peak SI Performance
Ten prompting habits that help, and the reminder that a clever prompt is not a performance test of superintelligence.
Disclosure. Further reading links to a sister publication under the same ownership. Those pages are commentary, not evidence for a score or a ranking on this site.
The short answer
State the job, the sources, the constraints, and how you will check the answer. Iterate on one change at a time. These habits improve drafts. They do not prove a model is superintelligent, and they do not replace tests.
How to read this list
This is an editorial map, not a tested ranking and not a score. BestSuperintelligent has not independently tested the products, discoveries, or headlines named here. Nothing below is a winner. Use it to decide what to verify, then check a primary source.
Prompt advice is cheap. A check you can rerun is the part that belongs on this site.
Which techniques are worth using?
A short sequence
- Name the reader and the jobSay who will use the output and what decision it supports.
- Attach the sourcePaste or point at the document. Do not ask the model to invent the file.
- Set constraintsLength, tone, tools it may not use, and facts it must not guess.
- Ask for the check firstCriteria, tests, or citations you will verify.
- Change one thingIf you rewrite the whole prompt, you will not know what helped.
- Show a good and a bad example when the format matters.
- Split research from the final draft.
- Ask it to list what it does not know.
- For code, require a test, then run the test yourself. Choosing a coding assistant covers that trial.
- Stop when the next prompt is hiding a missing source.
Evidence in this article Editorial interpretation Our reading of what the evidence means. Judgment, clearly separated from the evidence itself.
Frequently asked questions
What is the single most useful technique?
Give the success check before the draft: what must be true for you to accept the answer.
Do secret prompt templates create peak performance?
No. A template you cannot evaluate is folklore. Compare outputs on your own examples.
Will these techniques raise a benchmark score?
Not by themselves. A score needs a fixed task set. A better prompt on your own example is a workflow change, not a published result.
Sources
- How to read an evaluation BestSuperintelligent. Our own standard for what a result must show.
- Methodology BestSuperintelligent. Proposed weights and the rule that missing evidence is not a score.
Further reading
From a related publication under the same ownership. It is commentary, not evidence for anything on this site.
- Ten useful AI prompts, TheBestSI (sister site). Sister-site prompt examples. Try them on your material and check the result.
- Startup founder prompts, The Best Blog Ever (sister site). Sister-site prompt essay. Not a performance test.