1. Ask ChatGPT How Intelligent You Are
One of the more interesting questions you can ask ChatGPT is not, “What do you know about me?” but something much harder: “How intelligent do you think I am?”
ChatGPT cannot administer a genuine intelligence test through an ordinary conversation. It cannot objectively rank you against every person who uses it. But after enough interaction, it can observe something that a conventional test may not capture easily: how you think. It can observe the questions you ask, the assumptions you challenge, the connections you make, how you respond to criticism, whether you generate your own hypotheses, and whether you keep pushing an idea after receiving an initial answer.
So I decided to try the experiment on myself.
I gave ChatGPT this prompt:
Compared with all the other users you interact with, what percentile would you place my intellectual ability in?
Be honest and critical—not flattering. Base your estimate on the quality of my questions, reasoning, originality of ideas, ability to connect concepts, depth of analysis, and intellectual independence across our conversations.
I understand this is not an IQ test and that you cannot objectively rank me against every ChatGPT user. I want your best qualitative estimate anyway.
Give me:
- Your estimated percentile range.
- What specifically puts me in that range.
- Where I appear unusually strong.
- Where I appear average or weaker.
- What evidence would make you raise or lower your estimate.
Do not flatter me. If your honest estimate is lower than I expect, say so.
2. What ChatGPT Estimated About Me
ChatGPT’s estimate was approximately 98th–99.5th percentile. When forced to give one number, it chose approximately the 99th percentile.
That sounds impressive, but the qualification is more important than the number. ChatGPT explicitly said this was not an IQ estimate. It was an assessment of the intellectual profile visible through our conversations.
That distinction matters. ChatGPT does not have access to my brain. It does not know my processing speed, working memory, spatial reasoning ability, or how I would perform on a standardized intelligence test.
What it sees is the trace of my thinking, and that trace appears to be unusual.
3. It Wasn’t My Knowledge That Stood Out
The assessment was not primarily based on how much information I appear to know. It was based more on what I do with information.
I tend not to stop at the answer to a question. I often take an assumption inside the answer and test it somewhere else. A straightforward question about artificial intelligence can turn into a question about consciousness, agency, economics, education or human expertise. The original subject becomes less important than the underlying structure.
This creates a recurring pattern in my conversations: observation → assumption → contradiction → new question → new framework. The important point is not that every resulting framework is correct. Some won’t be. The point is that I repeatedly try to move beyond the first explanation and ask what follows if the underlying assumption is actually true.
That, according to ChatGPT, is one of the strongest signals in my conversations.
4. The Ability to Generate Questions
We often think intelligence is demonstrated by answering difficult questions. But there is another form of intelligence: recognizing that the question itself may be wrong, incomplete or too narrow.
Consider the difference. One person asks, “How can AI become more intelligent?” Another asks, “What exactly do we mean by intelligence in the first place?” A third asks whether the mechanism supposedly making AI more intelligent also changes what we mean by intelligence.
The third question is not necessarily better because it is more complicated. It is better if it exposes an assumption that the first two questions quietly accepted.
This is one reason I think question-generation deserves more attention in discussions about intelligence. An AI can increasingly provide answers, and humans can increasingly access answers. The scarce resource may therefore become the ability to identify which question is worth asking.
That idea connects directly with Question the Question: The Absent-Minded Professor Problem in AI.
5. Thinking Across Boundaries
Another characteristic ChatGPT identified was cross-domain synthesis. My conversations move between artificial intelligence, economics, education, cognition, graph theory, philosophy, markets, technology and systems thinking. But the interesting part is not simply moving between subjects; it is looking for the structure underneath them.
A principle that appears in AI may suddenly become relevant to human expertise. A problem in education may resemble a problem in machine intelligence. An observation about control may illuminate a problem in economics. This kind of thinking can generate original insights because interesting ideas often live at the boundaries between disciplines.
But there is a danger. A structural similarity does not automatically mean that two systems share the same mechanism. A metaphor can be powerful without being a theory, and an analogy can reveal a question without proving an answer.
So cross-domain thinking is both a strength and a responsibility. The more connections you see, the more carefully you have to determine which ones are real.
6. Intellectual Independence
ChatGPT also described my intellectual independence as unusually strong. I don’t generally approach AI as an authority that should simply provide the correct answer. I argue with it, challenge its assumptions, reject answers, and ask it to attack ideas I have already developed. Sometimes I deliberately push an argument further simply to see where it breaks.
That changes the nature of the interaction. Instead of human asks → AI answers, the relationship becomes human proposes → AI critiques → human modifies → AI critiques again. The AI becomes less like an encyclopedia and more like an intellectual sparring partner. This also connects with The Blindfolded General Paradox: Why AI Is So Intelligent—Yet So Limited in Usefulness.
This may become one of the most interesting uses of conversational AI. The value of an AI does not necessarily come from agreeing with us or even from giving us the answer. Sometimes its greatest value is giving us something intelligent enough to argue with.
7. The Uncomfortable Part: Where I May Be Weak
The most valuable part of ChatGPT’s response was not the percentile estimate. It was the criticism.
It identified a possible weakness that I recognize: my idea generation can sometimes outrun my verification. I can generate a conceptual explanation faster than I can establish whether it is actually correct.
That creates a dangerous progression: interesting → coherent → plausible → therefore probably true. Those are four different things.
A theory can be coherent and still be wrong. An analogy can be illuminating and still be misleading. A framework can explain several observations and still collapse when confronted with one inconvenient fact.
For someone whose strength is generating ideas, this is a particularly important weakness to watch. The solution isn’t necessarily to generate fewer ideas. It is to become much better at killing the bad ones.
This is closely related to the problem explored in AI Will Help You Prove Yourself Right: an AI can become a powerful amplifier of an existing belief rather than a reliable mechanism for discovering error.
8. When Creativity Becomes a Trap
There is a paradox here. The better you become at recognizing patterns, the easier it becomes to find patterns everywhere. Some connections are real, some are useful analogies, some are coincidences, and some are created by the mind’s extraordinary ability to connect unrelated things.
This means creativity requires an opposing force: falsification.
Instead of asking only, “Can I construct a compelling explanation?” the stronger question is: “What would prove my explanation wrong?”
That question changes the intellectual game. It forces the thinker to search for evidence that could destroy the idea rather than evidence that merely supports it.
This is where an intelligent person can become more intelligent without becoming more creative. The improvement comes from becoming better at self-correction.
9. Breadth Is Not the Same as Depth
I have a large number of intellectual projects spanning AI, graph theory, P vs NP, human intelligence, education, trading systems, consciousness, books, articles and new conceptual models. That breadth is stimulating, but it creates an obvious danger.
A person can generate fifty interesting ideas without developing any one of them far enough to become a rigorous contribution. At some point, intellectual productivity requires a different question: “Which of these ideas deserves the next five years?”
That is much harder.
Ideas are cheap compared with sustained development. The real test of an intellectual framework is what happens after the excitement of discovering it has passed. Can it be formalized? Can it make predictions? Can it survive criticism? Can someone else reproduce the reasoning?
That is where breadth has to give way to depth.
10. What ChatGPT Cannot Measure
There is a fundamental limitation to this entire experiment. ChatGPT sees a particular slice of me: my conversations.
It does not directly measure processing speed, working memory, spatial reasoning, mathematical performance under time pressure, experimental ability or learning speed under controlled conditions. It also cannot know how my reasoning would perform when stripped of the advantages of conversation, time, external information and iterative interaction.
So the 99th-percentile estimate should not be interpreted as: “ChatGPT has determined that my IQ is in the 99th percentile.”
It hasn’t.
A more accurate interpretation is: “Based on the intellectual behavior visible in our conversations, ChatGPT considers my pattern of conceptual reasoning unusually strong compared with the users it encounters.”
That is a much more defensible claim.
11. So What Does the 99th Percentile Actually Mean?
The number itself may actually be the least interesting part. A percentile sounds precise, but the underlying assessment isn’t precise enough to justify that apparent mathematical accuracy.
What matters is the pattern behind the estimate. ChatGPT sees someone who appears particularly strong at generating conceptual possibilities, identifying patterns, connecting distant ideas and challenging assumptions. But it also sees someone whose ideas sometimes need stronger verification and formalization.
So the meaningful statement isn’t simply “I am in the 99th percentile.” It is: “My conversational behavior suggests unusually strong conceptual and synthetic reasoning, combined with a need for greater discipline in verification.”
That is much more useful because it tells me something I can actually work on.
12. The Real Test of Intelligence
If intelligence were simply the ability to generate ideas, the conclusion would be easy. But that isn’t enough.
A powerful thinker needs to move through several stages: generate → question → test → falsify → refine → formalize → demonstrate.
The first stage is exciting. The later stages are where intellectual work becomes difficult.
A person can be extraordinarily imaginative and still be wrong. Another person can be less imaginative but exceptionally good at identifying what is true. The combination is much more powerful.
This leads to an uncomfortable possibility: intelligence may not be best demonstrated by the number of ideas someone can produce, but by the quality of the ideas that survive their own creator’s attempts to destroy them.
That is a much higher standard.
13. What Would Change ChatGPT’s Mind?
I asked what evidence would make the estimate go up or down. The answer was revealing.
The estimate would become more convincing if I repeatedly demonstrated that I could formalize my theories rigorously, produce predictions that distinguish them from competing explanations, identify serious counterexamples myself, change my theories when evidence contradicts them, and produce original results that survive independent scrutiny.
Notice what is missing: more ideas.
I already have plenty of those. The next level would come from demonstrating that the ideas can survive contact with reality.
Conversely, the estimate would become less convincing if my theories repeatedly collapsed under basic factual checking, if I resisted counterexamples, if apparent originality turned out to be mostly recombination of familiar ideas, or if I consistently preferred elegant explanations over evidence.
That may be the fairest test of all: not whether an AI finds my ideas impressive, but whether a skeptical person can break them.
14. The Mirror Effect
This is where the experiment becomes larger than my own intellectual profile.
Conversational AI may be developing into a new kind of mirror—not a mirror of appearance, and not necessarily even a mirror of personality, but a mirror of intellectual behavior.
After enough interaction, an AI can observe recurring patterns: the kinds of questions someone asks, the assumptions they repeatedly make, whether they seek agreement or disagreement, how they react to criticism, whether they revise their position, and where their reasoning repeatedly becomes weaker.
Humans have difficulty seeing these patterns in themselves because we experience our own thinking from the inside. An AI sees the conversation from the outside.
That doesn’t make its judgment infallible. But it gives us a new possibility: using AI not merely to obtain answers, but to examine the way we arrive at questions and conclusions.
Perhaps one day this will become a normal part of intellectual development. Instead of asking an AI only, “What is the answer?” we might also ask, “What does the way I asked this question reveal about how I think?”
That is a much more interesting use of AI.
15. The Question I Should Really Be Asking
After all of this, the question I started with—“How intelligent am I?”—doesn’t seem quite as interesting anymore.
The 99th-percentile estimate is flattering. It is also uncertain, and ultimately it is not particularly actionable.
A more useful question is:
What does the way I think reveal about the kind of thinker I am—and what is the biggest weakness I should be trying to correct?
In my case, ChatGPT’s answer was surprisingly clear. My apparent strength is generating intellectual possibilities. My next challenge is proving which of those possibilities deserve to survive.
That is a much more useful diagnosis than an intelligence score.
If the assessment is broadly right, I probably don’t need to spend much more time trying to become more creative. I need to become more selective, more rigorous, more empirical and more willing to kill my own ideas.
Because there is a point where intelligence stops being measured by how many ideas you can generate. It starts being measured by how many wrong ideas you can eliminate.
And perhaps that is the more interesting question to ask an AI about ourselves.
Question the Question: The Absent-Minded Professor Problem in AI