AI Notebook] Three Prompts for Thinking Better With AI - Find Your Blind Spots and Challenge Your Own Thinking

EJ's note

Sharing a few thought-provoking prompts I’ve been experimenting with. I find them especially useful when I want AI to do more than simply give me an answer — I want it to help me notice what I’m missing, challenge what I’m assuming, and look at a problem from perspectives I wouldn’t naturally consider.

Below was put together with the help of ChatGPT. Hope you can give it a try on your favorite AI platform.

Enjoy!

I. Jensen Huang’s “Top 5 Things” — Find the Weak Signal

Background

This idea is inspired by a long-running practice at NVIDIA known as “Top 5 Things” (T5T). Employees share short updates about what they are working on, learning, observing, or noticing, giving CEO Jensen Huang a way to stay close to what is actually happening across the company.

Huang has described the value of this kind of information as helping him detect “weak signals” — small developments that may become important before they are obvious to everyone else.

The prompt below adapts that idea for AI. Rather than asking, “What are the most important trends?”, it asks:

What has changed the most recently — and what might I be underestimating?

The six-month comparison and weak-signal exercise below are an adaptation of the T5T idea, rather than NVIDIA’s original process.

Step 1 — Identify what’s changing right now

My problem is [describe your situation specifically], but I’m frustrated because I can only think of obvious answers.

Using a “Jensen Huang Top 5 Things” approach, identify five things that are currently happening around my work, business, industry, or area of interest.

Don’t rank them by overall importance.

Instead, order them according to what has changed the most over the last six months.

Then flag one development that appears underappreciated relative to its potential impact.

If there isn’t enough information, don’t make assumptions. Ask me three questions first.

Step 2 — Explore the weak signal

For the item you just flagged, tell me:

What are three things that could happen if this development becomes significantly bigger six months from now?

Then create a 1,000-character action plan explaining what I could do now, before the development becomes widely recognized.

Four things beginners often miss

1️⃣ Give the AI real context.

If you describe your problem in only one sentence, the answer will probably be generic too.

Include your industry, role, company size, geography, objectives, constraints, and things you have actually experienced over the last three months.

The more specific the input, the more the five developments become your story rather than everyone else’s.

2️⃣ Answer the three follow-up questions.

Don’t skip them by saying, “Just figure it out yourself.”

Those questions force the model to recalculate the problem around your actual situation.

3️⃣ Use current information.

If you are asking what has changed most in the last six months, the model needs access to current information.

Ask it to use web search and show you the evidence behind its conclusions.

4️⃣ Don’t immediately regenerate an answer you disagree with.

Instead, ask:

“Why do you think people are underestimating this? Show me the evidence.”

The disagreement itself may reveal an assumption you were treating as obvious.

One extra step — highly recommended

After receiving the action plan, ask:

“If this action plan turns out to be completely wrong six months from now, what would be the three most likely reasons?”

Then ask:

“What evidence should I monitor over the next six months that would tell me this thesis is becoming wrong?”

This turns the exercise from a prediction into a testable hypothesis.

Finding weaknesses before acting is much cheaper than discovering them afterward.

⚠️ One caution

The five things AI identifies are not “the correct answers.”

They are hypotheses, perspectives, and angles you may not have considered.

Don’t blindly act on them.

Verify them in the real world.

Example: How I used it

My problem is understanding the future of humanity and the future of AI across different regions — the US, Europe, the UK, Asia, the Middle East, and others — but I’m frustrated because I can only think of obvious answers.

Using a “Jensen Huang Top 5 Things” approach, identify five things that are currently happening around this topic.

Don’t rank them by importance.

Instead, order them according to what has changed the most over the last six months, and flag one development that appears underappreciated relative to its potential impact.

If there isn’t enough information, don’t make assumptions.

Ask me three questions first.

II. Socratic Interrogation — Find the Assumption You Don’t Know You’re Making

Background

Most AI prompts ask the model to answer us.

This one asks the model to interrogate us instead.

Inspired by the Socratic elenchus, the goal is to identify an assumption you repeatedly make and test it through counterexamples until either the assumption survives — or you realize your original understanding was incomplete.

📄 Prompt 1 — Socratic Elenchus

Interrogate me using Socrates’ elenchus.

Base it on everything you and I have discussed so far.

Identify one assumption that I repeatedly make without sufficient justification.

Challenge that assumption only by asking questions that introduce counterexamples capable of breaking it.

Do not give me the answer yourself.

Do not tell me what conclusion I should reach.

Continue questioning me until I either defend the assumption successfully or acknowledge a contradiction.

If I acknowledge the contradiction, declare:

“Aporia.”

Then stop.

📄 Prompt 2 — What Did I Actually Learn?

Now end the interrogation.

In three lines, tell me what Socrates would leave me with:

• What I had been mistaken about
• The question I actually needed to answer
• One action I should change starting today

Then break that action into three concrete actions I can take this week.

📄 Prompt 3 — Socrates, Confucius and the Buddha

This is a modern prompt inspired by ideas associated with Socratic questioning, Confucian Rectification of Names, and Buddhist Dependent Origination.

Alternate between three frameworks: Socrates, Confucius, and the Buddha.

Use our entire conversation history as your source material.

On Socrates’ turn:

Use the elenchus to identify one assumption I repeatedly make.

Challenge it through questions that introduce counterexamples.

Do not give me the conclusion.

On Confucius’ turn:

Examine the issue through the idea of Rectification of Names.

Use comparative questions to test whether the role, title, identity, or value I claim actually matches the actions I am taking.

For example:

If I call myself a leader, are my actions actually those of a leader?

If I say something is a priority, does my behavior show that it is?

On the Buddha’s turn:

Trace the issue backward using the logic of Dependent Origination.

Ask questions that move backward through the chain of causes and conditions until we reach the earliest identifiable experience, trigger, perception, or attachment from which the pattern arose.

Do not diagnose me.

Do not tell me what I should believe.

Continue alternating between the three approaches.

Only when all three point toward the same underlying issue should you declare:

“Triple Aporia.”

Then stop.

If their conclusions do not converge, keep questioning me.

III. The Perspective Council — Make AI Argue Against Itself

Background

AI often follows the framing of the person asking the question.

That creates a problem.

If the question itself contains a blind spot, AI can sometimes produce an impressively logical answer inside the blind spot instead of challenging it.

One way around this is to deliberately create competing perspectives.

Instead of asking AI for one answer, ask several intellectually different viewpoints to examine the same problem, challenge one another, and identify what each perspective cannot see.

The goal is not to manufacture disagreement.

It is to discover:

Which conclusions survive disagreement?

📄 Prompt 1 — Create the Council

I am thinking about the following question, belief, decision, or problem:

[INSERT YOUR QUESTION HERE]

Create a council of five thinkers who approach this problem from fundamentally different perspectives.

Choose the perspectives based on the problem rather than using generic personalities.

The council should include something like:

• An advocate who believes my current thesis is basically correct
• A skeptic who believes its central assumption may be wrong
• An empiricist who cares primarily about evidence
• A systems thinker who focuses on second- and third-order consequences
• An outsider from a different discipline who may notice assumptions everyone else shares

Have each member independently answer four questions:

1. What am I seeing correctly?

2. What am I probably missing?

3. Which assumption in my reasoning is weakest?

4. What evidence would change your mind?

Do not make them agree with one another.

Preserve genuine disagreement.

📄 Prompt 2 — Let Them Debate

Now have the five members challenge one another’s strongest claims.

Before criticizing another perspective, each member must first steelman it — present the strongest possible version of that argument in a way its advocate would accept.

Then have each member identify:

• One claim they changed their mind about
• One disagreement that remains unresolved
• One piece of evidence that could resolve that disagreement

Do not force consensus.

If the evidence does not support agreement, preserve the disagreement.

📄 Prompt 3 — Find the Shared Blind Spot

Now step outside the council.

Look at the entire discussion and ask:

“What assumption did all five perspectives share without questioning?”

That shared assumption is the blind spot I am most interested in.

Then tell me:

1. What is the strongest counterexample to that assumption?

2. What becomes possible if the assumption is false?

3. What is one thing I should investigate in the real world before making a decision?

Do not tell me what decision to make.

📄 Prompt 4 — Assume Everyone Was Wrong

Now imagine it is six months from today.

We discover that every member of the council was wrong.

Ask:

What piece of information could exist that none of them considered?

Then identify:

1. What hidden variable could make all five perspectives wrong at the same time?

2. What weak signal today would give us the earliest indication that this is happening?

3. What evidence should I monitor over the next six months?

4. At what point should I seriously reconsider my original thesis?

The Common Thread

I think these three prompts are ultimately doing three different things:

Jensen Huang

What am I not noticing?

Socrates

What am I assuming?

The Perspective Council

What perspective am I missing?

They all point toward the same idea:

Don’t just use AI to give you better answers. Use it to expose the limits of the questions you are asking.

AI is most interesting to me not when it tells me what to think, but when it helps me discover where my own thinking may be incomplete.