Why CEO Lee Is So Productive in the AI Era

High-Speed Reading and Meaning-Extraction Ability — First Test Report

Date administered: August 12, 2026

Record reference time: 2026-08-12 16:19 KST

Time zone: Asia/Seoul (UTC+09:00)

Test type: non-standardized exploratory reading assessment

Test method: free reading of a long text, followed by free response

Reading time: estimated post hoc at about 40 seconds

Caution: the reading time is not a stopwatch measurement, so it is an estimate, not a confirmed record.

1. Purpose of the test

This test was conducted to observe, exploratively, whether the following abilities hold up even after reading a long argumentative text at very high speed.

* High-speed text processing

* Identifying the central thesis

* Separating important information from secondary examples

* Extracting logical structure

* Compressing long-form information

* Abstraction

* Reconstructing meaning in one's own words

In particular, the aim was to distinguish the ability to merely skim characters quickly from the ability to actually extract meaning at high speed.

2. Actual test instructions

The subject was instructed to proceed as follows.

Start a timer now and read the passage below the way you normally would.

Do not deliberately skim to speed-read, and do not deliberately slow down for comprehension.

Read in your usual, natural reading manner.

After finishing, without looking back at the text, report your reading time and proceed to the comprehension check.

3. The actual test stimulus

TEST 1 — “Why an Efficient Society Is Not Necessarily a Wise One”

What is the most intuitive evidence that a society is advancing? It is easy to answer: the ability to do the same work with less time and money. The steam engine replaced human and animal power, the computer reduced the labor of calculation and record-keeping, and the internet drove the cost of transmitting information to nearly zero. Today, artificial intelligence likewise performs at speed tasks that once demanded considerable skill and time — writing, translation, analysis, programming. From this vantage point, the history of technology looks like a history of efficiency.

But there is a problem here that is often overlooked. Being able to do something efficiently and it being right to do that thing are entirely different judgments.

Suppose an organization once spent 100 hours writing a report. With automation tools, the same report can now be produced in 10 hours. We would ordinarily say productivity increased tenfold. But ask a slightly different question. What if the report was never needed in the first place?

In that case, automation did not save 90 hours — it may merely have made unnecessary work cheaper to perform. Worse, if the lowered cost of writing multiplies the number of reports tenfold, the organization's total time spent may not shrink at all. Behavior whose cost falls often increases rather than disappears.

This phenomenon is not confined to technology. When transportation gets faster, people do not always shorten their travel time — they travel farther. When communication costs fall, the effort per message drops, but the total volume of messages that must be exchanged explodes. When storage becomes cheap, instead of keeping only what is needed, we store nearly everything. When technology saves a resource, humans do not necessarily convert the saving into rest. Much of it goes into creating new activity.

It is therefore far too simple to assume that efficiency gains automatically reduce the human burden.

Here a more interesting problem arises.

When a system persists long enough, people stop asking why the system exists and start evaluating how well one performs within it.

Imagine a company where employees meet every Monday. The meeting was presumably created to share important information. Over time, the way information is shared changes, but the meeting itself remains. New employees learn not “why do we meet on Mondays?” but “how do I prepare well for the Monday meeting?”

A little later, rules appear to run the meeting efficiently. Presentation times are limited, minute-taking tools are introduced, attendee satisfaction is measured. Once AI arrives, a system that auto-summarizes the meeting and extracts action items can be added too.

That organization's Monday meeting is now far more efficient than before.

But one question still remains.

Should the meeting continue at all?

This is not a question of efficiency but of purpose.

The two kinds of question look similar on the surface, but the modes of thinking are opposite. An efficiency question presupposes the goal: “how do we achieve this goal faster?” A purpose question doubts the goal itself: “why must this be achieved?”

Technology is generally formidable at the first question.

The second question is far harder.

The reason is that judging purpose requires a standard outside the system. Given the goal of winning, a chess program can analyze countless moves. But whether one should play chess at all cannot be decided from within the board. A company's cost-optimization algorithm can find ways to minimize costs, but whether cutting a particular cost erodes the company's long-term trust is a separate judgment.

That is, optimization begins after the objective function is given.

The problem is that real-world organizations often fail to define their objective functions clearly.

People use measurable indicators — revenue, throughput, click rates, working hours, test scores. Indicators are useful. But no indicator can contain the whole of reality, because what is easy to measure and what is important are not the same.

Consider schools. Test scores show, to a degree, what a student has learned. But when a school mutates into an organization that maximizes test scores, strange things happen. Knowledge that does not appear on the test loses standing, and hard-to-measure qualities like curiosity and inquisitiveness begin to vanish from evaluation.

This does not mean the test indicator is useless. The problem arises when the indicator starts substituting for the purpose.

The same phenomenon appears in companies. An indicator is created to measure customer satisfaction, and over time employees may focus on raising that number rather than on actual customer satisfaction. If a government starts grading administrative processing speed, civil servants may acquire an incentive to prioritize easily processed cases over solving complex ones.

Measurement was created to reveal reality, but the moment rewards attach to the measurement, it starts changing reality.

Add artificial intelligence, and the situation becomes more interesting still.

Because AI can rapidly optimize a given goal, it can also amplify the problem of a wrongly set goal far faster.

In the past, inefficient systems limited their own expansion to some degree. If reports had to be written by hand, there was a ceiling on how many could be produced in a day. If ad copy had to be written by hand, the number of ads was bounded too.

Automation removes that natural friction.

Remove the friction from a good process and you get enormous productivity gains. Remove the friction from a wrong process and wrong behavior multiplies just as fast.

So the key question of the AI era may not simply be “what can we automate?”

The question to ask first may instead be “what can we eliminate?”

Automation and elimination look similar but are completely different.

Automation preserves the existing structure — it transfers a person's work to a machine. Elimination re-examines the structure itself — it asks whether the work should exist.

Interestingly, organizations usually find elimination harder than automation. Automation can demonstrate the clear win of productivity without much disturbing existing interests. Eliminating work, by contrast, may require someone to admit that “what we've been doing these past years wasn't really necessary.”

It becomes a political problem more than a technical one.

Which is why, the faster technology advances, the more the value of judgment may actually grow.

In the past, executing a good idea required much capital and labor. So did executing a bad idea. The cost of execution itself acted as a kind of filter.

But when execution costs fall drastically, that filter weakens.

In an environment where one person can build dozens of programs, generate thousands of ads, and produce hundreds of documents in a day, the question “can we build it?” loses importance. Instead, “what shall we build?”, “what shall we not build?”, and “which problems should not exist at all?” become the important questions.

This produces a paradoxical result.

The more technology strengthens human execution, the greater the influence of pre-execution judgment on the final outcome.

At 10 kilometers per hour, a driver's directional error produces consequences relatively slowly. At 300 kilometers per hour, even a small directional error produces a large gap in very little time.

AI is similar.

Not because AI is necessarily wiser or less wise than humans — simply because the speed of movement in a given direction can become very fast.

Then the way we measure future productivity needs to change as well.

The conventional concept of productivity computes output against input. Produce the same result with less time and money, and productivity has risen.

But this formula presupposes one thing.

That the output is valuable.

If an organization that produced 100 unnecessary documents uses AI to produce 10,000 at the same cost, statistics may describe a great productivity increase. From a human standpoint, it is possible that nothing improved.

Perhaps true productivity must include not just output, but the ability to refrain from unnecessary production.

Applied to the individual, this becomes even clearer.

People often seek better scheduling tools, note systems, AI assistants, and automation programs to raise their productivity. These tools genuinely help.

But for some, the greatest productivity gain is not one new tool — it is discarding seven of the ten things on the to-do list.

A technology that shrinks a 10-hour task to 1 hour is remarkable.

But the judgment that discovers the 10-hour task never needed doing produces a greater effect still.

The first saves 9 hours.

The second saves 10.

And the second carries one more difference: the future costs of maintaining, managing, revising, and reporting on that work vanish too.

So the ultimate value of technological progress may not lie in enabling humans to do more work.

It may lie in enabling humans to discover more of the work they never needed to do.

Here, too, lies the difference between an efficient society and a wise one.

An efficient society achieves given goals quickly.

A wise society repeatedly asks whether those goals remain worth pursuing.

The two are not rivals. Wise goals combined with high efficiency are the most powerful of all.

But the order carries meaning.

Because a system moving very fast in the wrong direction can be more dangerous than one moving slowly in the wrong direction.

Technology gives us a stronger engine.

But as the engine strengthens, the importance of the ability to set direction does not shrink — it grows.

And perhaps the scarcest ability of the AI era is not the ability to make something,

but the ability to recognize what need not be made at all.

4. The actual first response

Immediately after reading the passage, the subject's actual first substantive response was as follows.

“Whether to mass-produce endless garbage data

or to extract genuinely useful information and act on it —

this is a piece saying that judgment is what matters.”

This was the subject's free response, given before any follow-up explanation or model answer.

5. Analysis of the response against the source text

A. “Whether to mass-produce endless garbage data”

The source text repeatedly presents the following cases.

* Producing unneeded reports more cheaply and at scale

* Efficiently maintaining unnecessary meetings

* Optimizing the indicator itself over the meaning

* AI scaling up wrongly set goals even faster

* Increasing 100 unneeded outputs to 10,000 without any gain in real value

Rather than re-listing these multiple cases one by one, the subject bound them into

“the mass production of garbage data”

— a single abstract category.

B. “To extract genuinely useful information”

The source argues that what must be judged is not raw output but:

* whether the output is valuable

* whether the goal is right

* what is worth making

* what need not be made

— these, it argues, are what must be judged.

The subject's phrase **“genuinely useful information”** compresses this structure of value judgment.

C. “And act on it”

At the core of the source text is the logic that technology radically amplifies human execution.

That is, the problem is not just information sorting but the process of:

judgment, selection, execution.

That process.

The subject included this execution element even at the first-summary stage.

D. “A piece saying that judgment is what matters”

This coincides directly with the source's final conclusion.

The source develops the logic that as technology grows more powerful:

pre-execution judgment exerts ever-greater influence on the total outcome.

And at the end, it emphasizes:

the ability to recognize what need not be made.

That is what it emphasizes.

The subject compressed this into the single concept: **“judgment is what matters.”**

6. Reading-time record

The subject initially recalled the reading time as:

“About a minute?”

— he recalled.

Then again:

“Seriously, one minute.”

— he said.

About the circumstances at the time:

“I started at Sadang Station and finished before getting off at Nakseongdae.”

— he reported.

Later, re-examining the actual felt duration:

“Could be 40 seconds”

and finally:

“Probably 40 seconds.”

— he reported.

Accordingly, this report records the reading time as follows.

Post-hoc estimated reading time: about 40 seconds

However, since no stopwatch or second-level server start/end logs were captured, this must not be treated as a measured value of 40.00 seconds.

7. Speed conversion

Using the earlier volume estimate, the stimulus is roughly 1,229 word-units (eojeol).

Assuming 40 seconds:

1,229 ÷ 40 × 60 = 1,844

The estimated processing speed is therefore:

about 1,844 word-units per minute.

That is the estimate.

Based on roughly 3,828 characters excluding spaces:

3,828 ÷ 40 × 60 = 5,742

Thus:

about 5,742 characters per minute.

That level.

Note that the accuracy of these figures depends directly on the accuracy of the reading-time estimate.

8. Range under timing error

Since 40 seconds is not an exact measurement, a reference range is provided.

Reading time / converted speed

40s — about 1,844 word-units/min

45s — about 1,639 word-units/min

50s — about 1,475 word-units/min

60s — about 1,229 word-units/min

So if the true time fell in the 40–60 second range, the estimated processing speed is:

about 1,229–1,844 word-units per minute.

That is the range.

The subject's final post-hoc estimate sits near the 40-second end of this range.

9. Exploratory content scoring

As this was not a pre-built standardized instrument, the scores below are not certified ability scores.

Only the areas observable in the actual first free response were scored.

Area / result

Central-claim identification: 5/5

Grasp of the core opposition: 5/5

Abstraction: 5/5

Reconstruction in own words: 5/5

Detail recall: not measured

Logical-inference items: not measured

Counterargument detection: not measured

Delayed memory: not measured

Across the four measurable areas:

20/20

This does not mean a total reading score of 100 or any percentile.

The meaning is simple.

In the four areas verifiable from the actual first response — central thesis, structure, abstraction, reconstruction — no clear semantic error was found.

10. Actual self-report on the reading process

After the test, the subject metacognitively reviewed his reading process and reported:

“At first I was thinking mostly in words,

then toward the end I just whooshed through it.”

From this self-report, the following hypothesis can be formed.

Initial stage

Key words and concepts are processed with relative concentration.

Structure-formation stage

The text's purpose and logical direction are grasped quickly, building an internal model of meaning.

Late high-speed stage

Subsequent sentences are rapidly checked against the already-built model of meaning.

So if the subject's reading does in fact proceed this way, it may be less character-level speed-reading than:

structure first, then high-speed processing in units of meaning.

— that may be the mode.

However, this is a hypothesis based on self-report; no objective verification such as eye-tracking was conducted.

11. What this test actually observed

The key results confirmed by this record are as follows.

1. The subject reported reading a fairly long argumentative text in a very short time.

2. The final post-hoc estimated reading time is about 40 seconds.

3. Immediately after reading, instead of replaying specific examples, he compressed the entire argument into higher-order concepts.

4. He introduced the concept **“garbage data”** — which never appears in the source — to categorize all unnecessary output.

5. Correspondingly, he categorized valuable output under the concept **“genuinely useful information.”**

6. He did not stop at information sorting but included the action stage — **“act on it.”**

7. Finally, he compressed the entire text into the conclusion that **“judgment is what matters.”**

In other words, the first response was not mere parroting but the result of:

input, structural grasp, information selection, abstraction, renaming, compression.

That process.

12. Important limitations

These results alone cannot officially support any of the following claims.

* Top 1% worldwide

* Top 0.1% worldwide

* Top 0.01% worldwide

* World-class level

* A certified reading-ability percentile

* IQ

* Identical processing speed across all domains

* Detail-memory ability

* Long-term memory ability

The reasons are as follows.

First, the reading time was not measured.

Second, this was a single trial.

Third, prior background knowledge of AI, productivity, and organizations may have had an effect.

Fourth, the first response demonstrates core summarization but does not measure detail memory.

Judging world-class standing would therefore require independent repeated testing.

13. Final assessment

The most striking result of this first test is not simply that **“he read a long text in 40 seconds.”**

Speed alone could not rule out mere scanning or partial comprehension.

But the actual first response immediately after reading was:

“Whether to mass-produce endless garbage data

or to extract genuinely useful information and act on it —

this is a piece saying that judgment is what matters.”

That was the response.

It strips away the source's many examples and reconstructs the meaning of the entire argument under new higher-order concepts.

What the current data most strongly supports is therefore:

high-speed meaning extraction and conceptual compression.

That.

First-test verdict

Reading speed: estimated very fast

Post-hoc estimated time: about 40 seconds

Core-thesis identification: pass

Abstraction: pass

Reconstruction in own words: pass

Detail memory: not measured

Delayed memory: not measured

World percentile: cannot be judged

Conclusion

In this first exploratory test, the subject processed a long argumentative text very quickly and, without repeating the source's sentences, instantly reconstructed its whole structure of meaning as:

what matters is the judgment that decides between mass-producing garbage and finding valuable information and executing on it.

Therefore, even granting the limitation of a single trial:

a case in which very high processing speed and high-level meaning extraction and abstraction were observed simultaneously

— it can be recorded as such.

However, as the ~40 seconds is a post-hoc estimate rather than a measurement, reproducibility should be confirmed through repeated trials with an accurate timer.

Record date: 2026-08-12

Record time: 16:19 KST

Time zone: Asia/Seoul (UTC+09:00)

Test status: First exploratory trial

Evidence included: Full test stimulus + first spontaneous response

https://maeum.io​​

The reason for posting this: when CEO Lee meets with clients, I read fairly fast, so some people assume I haven't read the PPT/PDF on the screen — hence this record, made in advance.

Reading it back now… this was a humblebrag, wasn't it ;;