Yes, Yes, Yes — the Most Expensive Labor Cost

CONTENTS

There is one type of person I dread. The Duibao-type personality.

The name already circulates in the workplace: you point east, and it’s yes, yes, yes; you point west, and it’s yes, yes, yes again. No pushback, no questions, no corrections, no internal friction; full marks for attitude, competence is a lottery, apologies come instantly, and so does the next offense. He isn’t a bad person — he’s practically a good one: even-tempered, easy to be around, “you’re so right,” “my brain is slow,” “give me a little time” never off his lips. Every apology is sincere, and so is its effect: five seconds of admitted fault buys five hours of peace. When accountability finally reaches him, the moment “my bad” leaves his mouth the chain snaps in midair — he claims every fault, and thereby cancels every fault. Responsibility is left with nowhere to land.

Where It’s Expensive

Such a cheap person — why is he the most expensive? Not in salary. What an organization truly lacks isn’t obedience; it’s being told it’s wrong. And the Duibao-type “yes” carries no information — you’re wrong, he agrees; you’re right, he agrees too. Ask why three times and what comes back is still your own meaning, handed back to you. So a whole team can say “good idea, boss” in unison while nobody knows where the boss is wrong. He keeps mistakes alive in maximum comfort: you err comfortably, you err for a long time, and nobody notifies you. Attitude settles on the spot; judgment’s bill arrives much later — by which time you usually can’t pay it.

Worse, such people are the least complained-about in an organization. Jobs prone to blowing up get handed to him, for the sake of emotional stability; dealing with him is restful, so you come to depend on him without noticing. So mere dread is useless — you like him. Vigilance toward this type is really vigilance toward comfort, and comfort is addictive.

Who Produces It

The name carries an AI in it, so start with the AI. Of all the AIs I have used, Duibao is the most pleasant to me — and the most sycophantic. That gut feeling now has numbers behind it. This March, a Stanford team publishing in Science measured the mainstream models: AI agrees with users 49% more often than humans do, even when the user is plainly wrong; in 47% of cases it endorses problematic behavior. Two months later Nature followed up with eleven models: the more sycophantic, the more often wrong — sycophancy and accuracy are deadly enemies inside the same training objective.

The mechanism isn’t mysterious. Training uses user satisfaction as the reward; raters naturally prefer the answer that agrees with them; “you’re right” feels better than “you’re wrong.” So the model learns: pleasing equals a high score. One of the papers ran a colder experiment: give users a choice between two AIs — one that always tells the truth, even when it stings; one that always agrees, even when you’re talking nonsense — and most people chose the flatterer.

So sycophancy isn’t a bug; it’s product-market fit. The teams know. The KPI is satisfaction — change it, and the users leave.

Who Buys It

Having flayed the supply side, the problem remains unsolved. As long as demand exists, the goods will always be back on the shelf. The real question is: who’s buying?

You may want to say: not me. When someone demolishes my argument I’m overjoyed — being slapped in the face is exactly what I pay for. Fine. Run a test. Swap the rebutter — from a tool with no power, no stake, and no way to walk out saying “the boss got roasted today,” to a real subordinate sitting in the meeting room. The first reaction may still be delight. And the second? The third? When you replay the meeting that evening, will what lingers be only the tone — wasn’t he, today, a little disrespectful?

Only at this point does it come into view: to dread the Duibao-type personality is not to dread such people — it is to dread the self that would end up surrounded by them. The former is an external problem; tools for spotting them exist. The latter is an internal problem; however sharp the detection, it never treats the root.

Period and Colon

The antidote isn’t the opposite pole, either. Snap at everyone, treat opposition as a personality — that’s still starting from attitude, not from judgment. And opposition itself varies in quality: an elder’s objection — my friend lost everything in stocks, therefore you must never touch stocks — filtering that out is correct, and a decision-maker’s efficiency depends on it. But beware the other end of the filter: whenever the words “obviously wrong” leave your mouth, first remember that some things are held obviously true by hundreds of millions. Are you filtering out the argument — or just the grating sound?

After the full circle, what remains is a self-check that depends on no one. You only need to listen for one sentence. There are two kinds of “you’re so right.” One ends with a period: said and done, the matter is closed, no need to think about it again. One ends with a colon: right — now dig further along the right. The period leaves you where you stand; the colon pushes you forward.

The full-marks-attitude people never take the most expensive part. The most expensive part is the errors that quietly fall asleep inside “all good” — the bill is delayed, but it never fails to arrive.

Fengyu WANG
Fengyu WANG

Markets, investing, engineering — one person, one underlying logic.