This Is Boring But It Will Make You Scary Intelligent

At a glance

Seventeen minutes on the one skill Sandeep Swadia thinks is about to become the only scarce thing left. His opening move is a refusal: people keep saying AI is turning us into the stupidest generation in history, and he does not buy it. The machines are not the problem. The problem is that we are losing the ability to check anything at exactly the moment when a lie can be manufactured that looks, sounds and blinks exactly like the truth. He calls it the crisis of critical thinking.

The structure is tight and it does not waste a second. Five distortions that hijack your judgment, one practical tool for each. Three of the five come at you from outside and he packs them into an acronym he spells ASC, pronounced "ask," because asking is the entire point. Authority, which is what let some of the smartest investors on earth put $700 million into Theranos without ever asking how the box worked. The selling, which is what Apple and Tesla and every credit card company do with "up to" and "starting at" and "as low as," the true lies that are legally airtight and informationally empty. And the consensus, the crowd, which is what Solomon Asch proved in 1951 can make you say a short line is long because seven strangers said so first. The last two come from inside: what AI tells you, and what you want to be true.

The examples are real, dated and checkable. A Hong Kong finance employee at Arup who did everything right, insisted on a video call, watched his CFO and several colleagues walk him through the transfer, and sent $25 million to a room full of deepfakes. Elizabeth Holmes and a board holding two former Secretaries of State and a four star general and almost nobody who understood blood. An MIT Media Lab study where the ChatGPT group could not quote a single line from an essay they had finished writing minutes earlier. And, at the end, a coffee shop conversation about a friend in love, which turns out to be the sharpest thing in the video.

This page rebuilds the whole thing in his order, keeps every number, every trick, every prompt he actually types, and then goes and finds the underlying studies so you can see where his retelling is exactly right and where the popular version of a famous experiment is stronger than the experiment. The video ends with a monk's line delivered in a business voice: the fog is not lifting, so drive slower.

Chapters

Click any timestamp and the floating player jumps there and keeps playing. Swadia did not post chapter markers on this one, so these are derived from the video's own transitions, which he announces out loud each time.

• 0:00 The crisis of critical thinking
• 0:54 The most expensive phone call ever made
• 2:25 A haystack of fake needles
• 2:40 Five distortions, three of them from outside
• 2:56 A: Authority, and the question nobody asked
• 4:57 The three signs the investors missed
• 6:32 The tool: what needs to be true for this to be real
• 7:16 S: Selling, and the land of true lies
• 8:23 Up to, as low as, starting at
• 9:29 The tool: the question mark move
• 10:25 C: Consensus, and what everybody knows
• 12:13 The one thing that broke the spell
• 13:03 Distortion four: what AI knows
• 13:55 The tool: be precise, please verify
• 14:23 Distortion five: what you want to be true
• 15:51 The tool: what am I refusing to see
• 16:01 The fog, and driving slower

The refusal, and the credential (0:00)

The first sentence is a position, not a hook.

People say AI is turning us into the stupidest generation in history. I don't buy it. The real danger is what I call the crisis of critical thinking.

The reason he gives is compact: in the age of AI, anything can be faked, and a lie looks exactly like the truth. That is not a claim about intelligence declining. It is a claim about verification getting harder. Those are different problems with different fixes, and the whole video follows from choosing the second one.

Then, at 0:19, the shortest bio in the genre. He trained as a monk. Then he went to MIT. Then he spent many years running and advising companies as a CEO and as a board member. That is the entire self introduction, and it is doing real work, because the two halves of it are the two halves of the video. The monk supplies the internal distortions. The board member supplies Theranos and the marketing.

What he says he learned from all of it is the thesis:

When machines can outsmart all of us, the only ability that will matter is your own judgment.

The promise for the next sixteen minutes: five distortions that hijack your judgment every day, and one simple tool in each case. Learn to see all five, he says, and you will feel like a genius catching what everyone else is missing.

The most expensive phone call ever made (0:54)

He starts with a story rather than a framework, and the story is true.

In January of 2024, an employee in the finance department of the Hong Kong office of a company called Arup got a message from his CFO about transferring $25 million. Arup is the London headquartered engineering and design firm behind the Sydney Opera House structure and a long list of buildings you have seen, which makes the story sting more, not less.

And here is the part that matters: the employee thought it smelled fishy. He did not comply. He did the thing every corporate security training tells you to do. He asked for a live video call to confirm it.

No problem. The CFO got on a video conference along with a few other team members, and together they walked the finance employee through a step by step process to transfer the money. Swadia's line about what happened next is the whole point of the story:

Now, the employee was super proud of himself because he insisted on in-person instructions from his CFO on a video conference. So, he transferred the money, and days later, he and his company found out that every person on that call was a deepfake.

He then defines the term for anyone who needs it. A deepfake is a fake video or voice made by AI to look and sound exactly like the real person. The faces, the voices, all of it was generated by AI from clips of the CFO scraped from the internet. And of course, the money was gone forever.

A haystack of fake needles (2:10)

Then the image that gives the video its spine.

You know, we've all heard that saying, finding the needle in a haystack. But now, you have to find the real needle from the haystack of fake needles that look exactly like the real one.

It is a good reframe because it names what actually changed. The old problem was scarcity of signal: the needle was one thing among many things that were obviously not needles. The new problem is that the haystack is made of needles. Nothing in the pile disqualifies itself on sight. Every fake needle passes the surface test, which means surface tests are finished as a filter and something else has to do the work.

At 2:25 he takes fifteen seconds for the plug: subscribe to his newsletter, one idea every week that helps you think more critically, every Tuesday, one insight, one tool, one practice, free. Then straight back in.

Five distortions, three of them from outside (2:40)

What makes it hard to find the real needle, he says, are five distortions. Three of them come from outside you. Those three he collapses into an acronym:

I call them the asks. We spell it ASC and it sounds like ask because that's the whole point.

He names the first letter out loud, Authority, and then lets the other two land with their own sections rather than reciting them. Working backward from the sections themselves, the S is the selling, the engineered language of marketing, and the C is the consensus, the crowd. The two remaining distortions are internal and he introduces them in place: what AI knows, and what you want to be true.

A: Authority, and the question nobody asked (2:56)

He starts with authority because, in his words, that is the force that can fool the smartest people in the room.

Some of the smartest investors threw hundreds of millions of dollars at a company called Theranos, but forgot to ask one very important question.

What Theranos sold (3:13)

The pitch was genuinely beautiful, and he says so. Theranos was hailed as one of the pioneering companies that could run hundreds of blood tests on a single drop of blood. All the tests that would typically require someone poking your arm with a needle and taking several tubes of blood, done from a fingerstick. It was, he says flatly, a brilliant idea.

The numbers he gives:

• $700 million raised from marquee investors.
• $9 billion valuation at its peak.

Both check out. The SEC's 2018 complaint says Theranos raised more than $700 million from investors through what it charged was an elaborate, years long fraud, and the $9 billion peak valuation is the standard figure.

Elizabeth Holmes, the character (3:39)

Then the portrait, and he does not pretend she was unimpressive. Elizabeth Holmes was such a dynamic and charismatic woman. The details he lists, in order:

• She dropped out of Stanford at the age of 19.
• She would work 20 hour days.
• She wore a black turtleneck just like Steve Jobs.
• She had a deep baritone voice. "She talked like this all the time," he says, dropping his own register to demonstrate.
• People noticed that she barely blinked when she was giving answers. "I can't do that," he adds, which is the video's one real laugh line.
• She had a sense of noble purpose.
• For 12 years, she kept the technology secret.

And the summary sentence, which is the actual mechanism: she was running a medical devices company that felt like a tech company from Silicon Valley. Move fast, stay stealthy, do not show the box. Those are reasonable norms for a photo sharing app and catastrophic norms for a diagnostic that decides whether you have cancer.

Except that none of it was true. The core technology had never worked as promised.

The facade cracked when the Wall Street Journal published John Carreyrou's investigation in October 2015, and it turned out the miracle machine was just a box that could not do anything she had promised. The company collapsed. Holmes was sentenced to 11 years in prison, and Swadia notes, correctly at the time of recording, that is where she is right now. The precise sentence was 11 years and 3 months, handed down in November 2022; she reported to FPC Bryan in Texas in May 2024, and good conduct credits have since moved her projected release to August 2032.

Sign one: the halo effect (4:57)

So how does it happen? How do the sharpest people in finance skip the basics? He gives three signs where critical thinking would have saved the investors.

First, the halo effect. Theranos had one of the most impressive boards on the planet:

• Two former Secretaries of State
• A retired four star general
• Former CEOs from major companies

Brilliant people. Powerful people. Influential people. And then the question that unravels the whole thing:

But did they know anything about blood chemistry or medical devices? Absolutely not. But nobody asked.

The names behind that description are worth having, because the halo is much easier to feel when it has faces on it. The two former Secretaries of State were George Shultz and Henry Kissinger. The four star general was James Mattis. Also on the board at various points: former Secretary of Defense William Perry, former Senator Sam Nunn, Admiral Gary Roughead, Wells Fargo chief Richard Kovacevich, and Riley Bechtel of Bechtel. It is one of the most decorated boards ever assembled, and its collective expertise in laboratory diagnostics was close to nothing. The halo effect is the bias that lets credibility earned in one domain get spent in a completely unrelated one, and this board is the cleanest example of it in modern business history.

Sign two: FOMO (5:25)

Second, the fear of missing out.

Nobody wanted to miss the next Nvidia or Apple or Google. And Elizabeth Holmes made the investors feel like the train was leaving the station all the time. Nobody asked.

This is the sign that is hardest to argue with, because venture capital is structurally built on it. The whole asset class is a bet that one enormous winner pays for every loss, which means the psychological cost of passing on Nvidia is far worse than the cost of losing a few million on a fraud. Manufactured urgency exploits a real incentive, not an irrational one, which is exactly why it works on professionals.

Sign three: the veil of secrecy (5:39)

Third, the veil of secrecy. And here he produces the receipt.

A reporter asked Holmes six times how the company's technology actually worked, and he never got a straight answer.

The reporter was Ken Auletta, profiling Holmes for The New Yorker in December 2014. Auletta later described asking her, six different times, some version of "tell me what happens to the nanotainer of drops of blood when you put it in your machine," and getting, in his own phrase, gobbledygook six times. The answer Swadia plays is the famous one:

A chemistry is performed so that a chemical reaction occurs and generates a signal, which is translated into a result.

His verdict: that meant almost nothing. And then the question that should have ended the meeting:

Could a real medical breakthrough meant to diagnose your health ever avoid any scrutiny at all? Absolutely not. But nobody asked.

Notice the structure of the three signs. Every one of them ends with the same four words. The halo effect, the FOMO and the secrecy are not three separate failures. They are three different reasons the same question never got asked.

It keeps happening (6:12)

And it is not a one off. He rattles off the pattern:

That story keeps repeating itself. Enron, FTX, WeWork, the 2008 financial crisis. In every single one of those cases, people replaced their own judgment and let the charisma, the credentials, and the confidence of others distort their reality.

Three Cs, and none of them is evidence. That is the definition of the authority distortion: substituting a signal about the speaker for a signal about the claim.

The tool: what needs to be true for this to be real (6:32)

The action item is a single sentence, and it is deliberately unaggressive.

The first skill about critical thinking is just asking a simple disarming question that nobody else has bothered to ask. What needs to be true for this to be real?

The word doing the work is "disarming." He is not telling you to accuse anyone of lying. The question is structurally polite and logically brutal at the same time. It does not attack the claim, it asks the claimant to enumerate the preconditions of their own claim, and the moment they start listing them, each one becomes independently checkable. Run it on Theranos and you get: the chemistry has to work at microliter volumes, the dilution has to not destroy the signal, hundreds of assays have to run on one device, and the FDA has to have looked at it. Four checkable statements, all of which were false, from one polite question.

Then the AI extension, and he is careful to gate it:

This is where AI can help, too, by the way. But only if you use it correctly. So, if someone makes an impressive claim, go back to AI and ask: Show me the evidence that supports the claim. Show me all the evidence that challenges it. What needs to be true for this to be real?

That is a three part prompt and it is deliberately symmetrical. Asking a language model for evidence in favor of something is the single easiest way to manufacture false confidence, because the model will oblige. Asking for both sides in the same breath, and adding the preconditions question, forces it to build the case it would otherwise never volunteer.

Theranos was a $9 billion failure of critical thinking, an expensive lesson in bad judgment.

S: Selling, and the land of true lies (7:16)

The bridge is good: the same magic tricks are played by lawyers and marketers every single day all around us.

You know, big companies don't lie to you anymore. They don't have to. They have something better. The smartest companies on Earth have mastered one dark art. How to create true lies.

A true lie is a statement that would survive any legal challenge and still leave you with a false belief. It is not a lie in the sense a court cares about. It is a lie in the sense your brain cares about.

Word noodles (7:41)

He starts with the phone in your pocket, and he opens with a disclosure: he really likes Apple products, so he is not dissing them. But if you watch how Apple talks about their products at launch, it is a master class in marketing gymnastics. He plays the clip:

Today, we're raising the bar with an all-new iPhone that is unlike anything we have created.

Then he takes it apart in two sentences:

You know, when you listen to that line carefully, you realize that they didn't say anything. All they said was, "We've created something new that we hadn't created before." Well, that's what new means.

The sentence is a tautology wearing a suit. "Unlike anything we have created" is entailed by "all-new," and "raising the bar" has no referent, no unit and no comparison. It contains zero bits of information about the product, and it feels like an announcement. His name for it is the best coinage in the video:

Those are just word noodles. Perfectly edible, but if you're a critical thinker, you won't be able to digest them.

Trick one: up to (8:23)

One trick is "up to." The company will say, "The battery can last up to 36 hours."

And then the demolition, which he does on himself rather than on a company:

That's exactly like me saying this video will have up to 1 billion views. If only 12 of you show up, well, I still told the truth.

"Up to" is a ceiling, and a ceiling constrains nothing on the downside. Every number from zero to the ceiling satisfies it. The claim "up to 36 hours" and the claim "somewhere between 0 and 36 hours" are logically identical, and one of them sounds like a feature.

Trick two: as low as (8:38)

Another trick is "as low as." A credit card company will offer you as low as 0% interest, as long as you qualify perfectly and behave perfectly. Miss once and it's 24% compounding every month on everything.

Same shape, mirrored. "As low as" is a floor, so it also constrains nothing on the side you will actually land on. What makes this one nastier than "up to" is that the advertised number is not merely unlikely, it is conditional on a behavioral tripwire, and the penalty for tripping it can be a 24 point jump applied retroactively to the whole balance.

Trick three: starting at (8:54)

Another trick is the phrase "starting at."

And then the worked example, with real numbers he timestamps himself by saying "at the time I am recording this." The Tesla Model 3:

• Starting at $37,000 in the US
• Want all wheel drive? $47,000
• Want faster acceleration? $55,000
• Want the self driving activated? An extra $1,200 per year, forever

Those figures are close to the public list at the time. The Model 3 Standard carries a base price of about $36,990 and the Performance trim about $54,990, and Tesla's Full Self Driving has been offered as a $99 per month subscription, which is $1,188 a year. The trailing word "forever" is the sharpest part of his example, because a subscription is a price with no terminal value, and "starting at" quietly excludes it entirely.

His summary of the whole category:

Everything we hear is lawyered up, sanitized, manicured, curated, brilliantly engineered true lies.

The tool: the question mark move (9:29)

So, here's one actionable move, and it works on everything. I call it the question mark move. It has two steps.

Step one: train your ear for the shiny words, the word tricks. His list of empty calories: "up to," "starting at," "clinically proven," "recommended by experts."

Step two: when you hear one of those phrases, repeat the claim in your head with a question mark at the end.

So, when you hear "up to eight times faster," change that to "up to eight times faster?" What does that even mean, exactly? Faster than what?

What makes this a genuinely good technique rather than a slogan is that the punctuation change does the cognitive work for you. A declarative sentence is processed as an assertion and filed. The same words with an interrogative contour cannot be filed, because a question demands a completion, and the completion is exactly the missing information the marketer removed. "Faster than what" is not a clever retort you had to think of. It is the hole the question mark opens.

Once you start cutting through these true lies, you'll start seeing critical gaps between the promise and the product.

And the transition line, which is a nice piece of writing:

Now, the next skill is trickier. Creating a gap between what they're selling and what you're willing to buy.

C: Consensus, and what everybody knows (10:25)

Group think stops you from thinking on your own.

The experiment (10:28)

He describes it without naming it, which is a choice worth flagging, because it is one of the most famous studies in psychology. This is the Asch conformity experiment, run by Solomon Asch at Swarthmore College starting in 1951.

His account: researchers brought one real participant into a room with seven others already sitting there. Everyone was shown one line, then three comparison lines, and asked which comparison line matched the first one.

The answer was obvious. There was no optical illusion, there was no trick. Only one of those three lines matched the original one.

That detail is the entire design. Asch deliberately built a task with no ambiguity so that any wrong answer could only be social, never perceptual. In the control condition, with no group pressure at all, people got it right more than 99 percent of the time.

But one by one, the seven people gave the wrong answer. The eighth person, who was the participant, ended up giving the wrong answer, too.

And the reveal:

Those seven who were sitting there were hired actors. They were told to give the wrong answer, confidently. The experiment was never about the lines. It was about whether a room full of confident people could make you sway your own judgment on something so obvious.

What the participants confessed (11:31)

This is the part that makes the study land, and he keeps it:

Later on, the participants themselves confessed that they had known the right answer all along. They just couldn't stand to be the only one in the room with a different viewpoint.

That distinction is the whole thing. Asch's subjects did not see the wrong line. They saw the right line and said the wrong one. What the crowd overrode was not perception, it was the willingness to be alone.

When everyone around you believes something to be true, you stop asking for proof, because a thousand people can't all be wrong, can they? Turns out, they can be.

Crowds that have been wrong (11:57)

The examples he lists, fast:

Massive crowds can believe in things with no basis at all. The Salem witch trials, the tulip mania, the bank runs during the Great Depression, the dot-com frenzy.

And then the balanced sentence, which is the reason this section is not just cynicism:

The wisdom of the crowd is a real concept. But so is mass hysteria and groupthink.

Both are real, and they are not opposites of each other so much as different regimes of the same system. Crowds aggregate well when the individual estimates are independent. They aggregate catastrophically when everyone is looking at everyone else, which is what the Asch setup manufactures on purpose and what a bank run manufactures by accident.

The one thing that broke the spell (12:13)

The best finding in the video, and he is right to save it:

And here's another surprise from the same experiment. When just one of those actors started giving the correct answer, most participants also started giving the correct answer. It just took one person to disagree. That's all.

Asch ran exactly this variant, and the effect is enormous. Conformity in the unanimous condition ran around a third of critical trials; with a single ally who gave the correct answer, it collapsed to roughly 5 percent. Unanimity, not majority size, was the load bearing element. One dissenter removes most of the pressure, and that has an operational consequence: if you are the dissenter in a room, you may not persuade anyone, but you are almost certainly freeing someone.

The tool: go find the person who disagrees (12:27)

So, the action item is this. When everyone around you agrees, go find that person who disagrees.

And then the AI version, for the increasingly common situation where there is no such person available:

When everyone around you agrees, make AI your voice of dissent, your devil's advocate. Tell it to give you an opposite argument. Make the strongest case that this consensus is wrong.

The word "strongest" is the operative one. A weak counterargument is worse than none, because knocking it down manufactures exactly the confidence you were trying to test. Asking for the strongest case is asking for the steel man, and a language model is unusually well suited to producing one on demand, since it has no ego invested in the consensus you are questioning.

Then the guardrail, which a lot of critical thinking content forgets to include:

And by the way, you don't have to become the person who doubts everything. Just remember that a thousand people believing in something doesn't automatically make it true.

That is the correct calibration. The goal is not universal skepticism, which is its own failure mode and a very tiring personality. The goal is to stop treating headcount as evidence.

Distortion four: what AI knows (13:03)

Those were the three outside distortions. But we are in an age where AI is the biggest generator of ideas and biggest generator of distortions.

The study he cites:

MIT researchers asked people to write essays in two ways. One group used their own brain and outside research, and the other used ChatGPT. The brains of ChatGPT users showed the least amount of brain activity compared to the other group. In just minutes after both groups finished, they were asked to quote at least one line from what they had written. The ChatGPT group couldn't quote a single line of what they had just created.

This is Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task, by Nataliya Kosmyna and seven co-authors at the MIT Media Lab, released in June 2025. There were 54 participants and three groups rather than two: LLM, Search Engine, and Brain only. Each wrote essays under EEG in three sessions, with a fourth session in which the LLM and Brain only groups swapped conditions. The headline neural result is that brain connectivity scaled down with tool use: the Brain only group showed the strongest and most distributed networks, the Search Engine group moderate engagement, and the LLM group the weakest.

And the quoting result, which is the one he uses, is real and is the most visceral thing in the paper.

His reading of it:

AI tools give us an easy and powerful way to create infinite content. Instead of thinking creatively or thinking critically, we outsource all of it to AI.

The word "outsource" is precise, and the quoting failure is what makes it more than a metaphor. You cannot quote what you never encoded, and you do not encode what you did not have to construct. The essay existed. The author's memory of it did not.

The tool: be precise, please verify (13:55)

Two phrases, and he says he uses them more than anything else in a prompt:

When I'm working with AI, I use two lines way more often than anything else in the prompt. Be precise and please verify.

Then the cross examination habit:

I even take the output from ChatGPT, for example, to Claude or Gemini to verify it. Ask one of your AI engines to review the other one's work.

And the framing line that justifies all of it:

They're just mathematical beings, and it is your job to constantly verify their output.

There is a real reason cross model checking works better than asking a model to check itself. A single model asked to review its own answer tends to ratify it, because the same weights that produced the error also evaluate it, and the error is inside the region of the space the model finds plausible. A different model, trained on a different mixture with different failure modes, is not systematically wrong in the same places. It is not a proof procedure, but it is genuinely closer to independent sampling than a self check is, and it costs one paste.

Notice too that this is the same instrument he used in section one and section three. The AI is not the authority in any of the three prompts. It is the adversary, the researcher, the second pair of eyes. He never once tells you to ask it what is true.

Distortion five: what you want to be true (14:23)

And now we go to the fifth and the most difficult critical thinking skill: fighting what you want to be true.

The framing sentence is the best in the video:

There is one source that you will never fact-check because you trust it completely: yourself.

And the mechanism, stated without moralizing:

We don't lie to ourselves on purpose, we just want something to be true so badly enough, so we stop checking.

That is a better description of motivated reasoning than most textbook ones, because it locates the failure in the stopping rule rather than in the conclusion. Nobody decides to believe something false. They decide, without noticing, that they have checked enough.

The two friends (14:44)

Then he abandons business examples entirely and tells a personal story, and it is the right call, because self deception does not feel like Theranos. It feels like this.

Years ago he had two friends, a man and a woman. One day the man sat him down over coffee and told him how deeply he was in love with her, and from that day on he would not stop talking about her. He truly believed that they were made for each other.

Eventually Swadia gave up and had a heart to heart with the woman. He told her everything he had heard. He told her how deeply this guy was in love with her, and confessed that he had not seen such devotion anywhere else in his life.

And she smiled and said something that I have remembered to this day: His love for me is not enough of a reason for me to fall in love with him.

That was very wise.

It is the sharpest line in seventeen minutes, and it lands because it is a category error being corrected in real time. The man had an enormous quantity of one thing, his own feeling, and had silently converted it into evidence about a different thing, her feeling. The intensity was completely real. It was just evidence about him.

Now, what does this have to do with critical thinking? It's got everything to do with critical thinking, because we want something to be true just because we want it so deeply, but that doesn't make it true. We all get stuck in our own self-deception. We believe in what we want to believe.

The tool: what am I refusing to see (15:51)

The most important question for the critical thinker is this: What am I refusing to see because I need this story to be true?

Look at the construction. It presupposes that there is something you are refusing to see, which means the sentence cannot be answered with "nothing." It also names the mechanism, "because I need this story to be true," which forces you to identify which story you are protecting before you can identify what it is costing you. It is the internal version of "what needs to be true for this to be real," pointed the other way: the first question audits someone else's claim, the last one audits your own attachment.

He calls it the most difficult of the five, and he is right, for a reason he does not spell out. The first four distortions have an outside surface you can inspect. The fifth one is running the inspection.

The fog, and driving slower (16:01)

The close is a monk's ending delivered by a board member, and he lets it run.

Today, what's real and what's fake, what matters and what doesn't, it's all stirred together so completely that you can barely tell it apart. And when the world gets this foggy, your judgment becomes that much more crucial.

His answer is deliberately unglamorous. No tool, no app, no framework.

I think the answer lies in staying close to reality and investing in your own judgment, in your own critical skills. That's what this video was about.

And then the image that pays off the title. This is boring, and boring is the point.

And sure, the fog won't lift anytime soon and it is hard to see at a distance, but that doesn't mean that you're lost. It just means you have to drive slower. The road can be long and it can wind, but it still leads you home, always.

I'll see you next week. Thank you. And I love you.

The metaphor is doing more than it looks. Fog does not make a route unknowable, it makes speed unsafe. Nothing about the destination changed. What changed is how much stopping distance you need, which is exactly what every one of the five tools buys you: a small deliberate delay between the claim arriving and the claim being filed as true.

The five tools, on one card

Everything actionable in the video, in the order he gives it and in the words he uses:

Best quotes

When machines can outsmart all of us, the only ability that will matter is your own judgment. (0:31)

The employee was super proud of himself because he insisted on in-person instructions from his CFO on a video conference. So, he transferred the money, and days later, he and his company found out that every person on that call was a deepfake. (1:31)

We've all heard that saying, finding the needle in a haystack. But now, you have to find the real needle from the haystack of fake needles that look exactly like the real one. (2:12)

Brilliant people, powerful people, influential people. But did they know anything about blood chemistry or medical devices? Absolutely not. But nobody asked. (5:15)

Theranos was a $9 billion failure of critical thinking, an expensive lesson in bad judgment. (7:12)

Big companies don't lie to you anymore. They don't have to. They have something better. The smartest companies on Earth have mastered one dark art: how to create true lies. (7:30)

Those are just word noodles. Perfectly edible, but if you're a critical thinker, you won't be able to digest them. (8:13)

That's exactly like me saying this video will have up to 1 billion views. If only 12 of you show up, well, I still told the truth. (8:30)

The experiment was never about the lines. It was about whether a room full of confident people could make you sway your own judgment on something so obvious. (11:22)

They had known the right answer all along. They just couldn't stand to be the only one in the room with a different viewpoint. (11:37)

A thousand people believing in something doesn't automatically make it true. (12:56)

They're just mathematical beings, and it is your job to constantly verify their output. (14:18)

There is one source that you will never fact-check because you trust it completely: yourself. (14:33)

His love for me is not enough of a reason for me to fall in love with him. (15:27)

What am I refusing to see because I need this story to be true? (15:55)

The fog won't lift anytime soon and it is hard to see at a distance, but that doesn't mean that you're lost. It just means you have to drive slower. (16:33)

Where it stands

An honest ledger, because a video about checking claims deserves to have its own claims checked. The verdict overall is that his facts hold up unusually well for the format. Four notes.

The Arup story is accurate, and one detail in the telling is a slip. He says the employee "asked the CEO to get on a video call," then immediately continues with the CFO, which is who it actually was. Everything else matches the reporting: January 2024, Hong Kong office, an initial phishing email the employee correctly distrusted, a video conference populated by AI generated versions of the CFO and colleagues, HK$200 million across 15 transfers to 5 accounts, about US$25.6 million. Arup's global CIO confirmed publicly that the number and quantity of attacks against them had risen sharply. The money has not been recovered.

The Asch retelling is directionally right and slightly stronger than the study. "Three out of every four participants went along with the wrong group answer" is true in the sense that about 75 percent of participants conformed at least once across the twelve critical trials. But the per response rate was 36.8 percent, and about a quarter of participants never conformed on a single trial. So the crowd wins roughly a third of the time, not three quarters of the time, and a meaningful minority is completely immune. The dissenter finding he gives is, if anything, understated: conformity did not merely improve, it collapsed to roughly 5 percent. Worth knowing too that Asch's participants were male American college students in the early 1950s, and that the Bond and Smith meta analysis of 133 studies across 17 countries found conformity in the United States has declined since then and varies by culture. The mechanism is robust. The exact number is a product of its era.

The MIT study is real, is a preprint, and its authors have asked people not to say what almost everyone said about it. Your Brain on ChatGPT has 54 participants, only 18 of whom completed the fourth session, tested only ChatGPT, drew from a narrow geographic pool near Boston universities, and had not been peer reviewed at release. The authors published an FAQ asking specifically that coverage avoid "stupid," "dumb," "brain rot," "harm," "damage," and the claim that LLMs make you stop thinking, and they note the paper measures neural connectivity rather than making claims about intelligence. To Swadia's credit, he does not use any of the forbidden vocabulary and does not say AI makes you dumber. His phrase "the least amount of brain activity" is looser than the paper's "weakest connectivity," and his "couldn't quote a single line" is a compression of 83 percent failing in session one. Neither distortion changes the conclusion he draws from it, which is a practical one about verification habits.

The Theranos board did contain two physicians, which slightly softens the halo point without breaking it. Bill Frist, a heart and lung transplant surgeon, and William Foege, the epidemiologist who led the smallpox eradication campaign, both served. So "did they know anything about blood chemistry" is not literally zero. But neither was a laboratory diagnostics specialist, neither ran the assay validation the company never did, and the board's overwhelming center of gravity was diplomatic and military rather than scientific. The Wall Street Journal reporting and John Carreyrou's book both make the same structural argument he makes here.

The Tesla numbers are a snapshot and he says so. "At the time I am recording this" is an honest hedge, and the figures were close to Tesla's public list. Model 3 pricing and the Full Self Driving subscription have moved repeatedly and will move again, which does not weaken the point at all, since the point is about the shape of the phrase "starting at" rather than about any particular price.

One thing the video does not claim, and it is worth noticing: nowhere does he say AI is bad, or that you should use less of it. Three of his five tools are AI prompts. The argument is not about the tool, it is about which seat the tool sits in. Adversary, researcher, second reader, yes. Oracle, no.

Resources

The video and the creator

• This Is Boring But It Will Make You Scary Intelligent, Sandeep Swadia, 2 July 2026
• Sandeep Swadia on YouTube
• sandeepswadia.com and the about page
• The newsletter, the one he plugs at 2:25: one insight, one tool, one practice, every Tuesday
• Massachusetts Institute of Technology, where he studied

The deepfake

• Arup, the engineering and design firm targeted
• CNN: Arup revealed as victim of $25 million deepfake scam involving Hong Kong employee
• Financial Times: Arup lost $25mn in deepfake video conference scam
• Deepfake, the term he defines at 1:48

Theranos and the authority distortion

• Theranos
• Elizabeth Holmes
• Hot Startup Theranos Has Struggled With Its Blood-Test Technology, John Carreyrou, The Wall Street Journal, 15 October 2015
• Bad Blood: Secrets and Lies in a Silicon Valley Startup, John Carreyrou, the full account
• Blood, Simpler, Ken Auletta, The New Yorker, December 2014, the profile containing the "a chemistry is performed" answer
• Ken Auletta on getting gobbledygook six times
• SEC charges Theranos, Holmes and Balwani with massive fraud, the source of the $700 million figure
• Holmes's release date moved to August 2032
• The Inventor: Out for Blood in Silicon Valley, Alex Gibney's documentary
• Board members named in this page: George Shultz, Henry Kissinger, James Mattis, William Perry, Sam Nunn, Gary Roughead, Richard Kovacevich, Riley Bechtel, Bill Frist, William Foege
• Halo effect and fear of missing out
• The other collapses he lists: Enron, FTX, WeWork, the 2008 financial crisis

The selling distortion

• Apple and the Apple Events archive, source of the keynote line he plays
• Tesla Model 3 and Full Self Driving subscriptions
• Weasel word, the general name for the category he calls word noodles
• Federal Trade Commission on truth in advertising, which is why these phrases are built the way they are

The consensus distortion

• Asch conformity experiments and Solomon Asch
• Simply Psychology's writeup of the Asch line study, including the 36.8 percent, 75 percent and dissenter figures
• Bond and Smith, Culture and conformity: a meta-analysis of studies using Asch's line judgment task, 1996
• Groupthink, wisdom of the crowd, mass psychogenic illness
• The manias he lists: Salem witch trials, tulip mania, bank runs, the dot-com bubble
• Steelmanning, what "make the strongest case this consensus is wrong" is asking for

The AI distortion

• Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task, Kosmyna et al., MIT Media Lab, June 2025
• The project page and the authors' FAQ on how the study should and should not be described
• Nataliya Kosmyna at the MIT Media Lab
• The three engines he names: ChatGPT, Claude, Gemini
• Electroencephalography, the measurement behind the connectivity result

The self distortion

• Motivated reasoning
• Self-deception
• Confirmation bias, the mechanism that keeps the checking from restarting