August 25, 2026
For Wall Street insiders, Stanley Druckenmiller is a legendary hedge fund manager and macro investor whose words can move markets. To the general public, who only occasionally glance at the financial news headlines, he’s just another bold-faced billionaire name who gets feverishly quoted by the finance bro/broette community. Both descriptions are true, but in 2026, a new distinction must be added to his online identity: if you quote Druckenmiller, you may not actually be quoting the man, but rather a statistical approximation of his opinion, rendered in the words of AI.
This was revealed today following Druckenmiller’s Wall Street Journal opinion article, published on Monday, where he criticized the fiscal policy of his former ‘90s-era mentee, current U.S. Treasury Secretary Scott Bessent. Aside from the policy prescriptions and market warnings embedded in the piece, readers noticed something odd. It seemed to contain a lot of the now obvious telltale signs of AI-generated text.
It didn’t take long for the Internet to weigh in, widely branding the op-ed “AI slop.” Notably, former White House senior economist and member of the Federal Reserve Board Claudia Sahm posted the results of a Pangram AI detection test on her Twitter/X profile. The results: 100% AI-generated. Of course, none of the current AI detectors are considered infallible, but those kinds of results for brand new text are usually an indication of the presence of some AI-generated text.
The AI mystery was dispelled this morning after Druckenmiller admitted, “Of course I used AI,” in an interview with political news website NOTUS. Such disclosures are becoming increasingly common. But it’s the subsequent statement from Druckenmiller that got my attention: “There’s a reason I moved from an English major to being an economics major… I’m not embarrassed by it … I write everything using AI now for the same reason I use a calculator when I do math problems.”
Finally! Someone from the business community crystallized the pernicious fallacy I’ve been obsessing over for the past few years when it comes to AI-generated writing. By comparing the use of AI-generated text for human opinion to using a calculator to arrive at math solutions, Druckenmiller has revealed the exact reason why allowing AI to speak for you is so dangerous.
🙋🏻♂️🤖 MAN AND ROBOTS lands in your inbox every Friday: AI, Hollywood, and the future of work. Get it free →
Why Business Leaders Are In Love With AI Writing
To do big things, you need a lot of confidence. Some might even call it arrogance, or chutzpah, whatever you call it, you often need to believe in your own ideas more than your doubters in order to get ambitious things done. Regardless of political stripe, this ethos undergirds many of the most successful entrepreneurs, CEOs, and finance experts. And in this realm, the scoreboard isn’t intellectual veracity, it’s money. Profit. Number Go Up. Full stop.
Making these numbers go up often involves a series of logic chains and efficiencies that primarily respect the unforgiving taskmaster of math and statistics. Nuance isn’t necessarily appreciated, unless, of course, it’s backed by numbers, in which case, it’s no longer really nuance, it’s just fact. Thinking like this becomes a mental muscle that many business leaders pride themselves in enhancing and flexing. That’s why, prior to the rise of commercial AI chatbots, the market for corporate ghostwriters was never better. Business leaders accustomed to thinking in stark numbers, married with their previously seldom publicly discussed social theory, often struggled with the subtle poetry of sculpting their thoughts in a way that evokes mental imagery that resonates with the public and speaks beyond the bottom line of profit and loss.
Then came ChatGPT, Claude, and a range of other AI chatbots, all giving anyone with a $20-a-month subscription an easy way to write a 3,000-word essay of seemingly brilliant, albeit often formulaic, “original” opinion on any topic you could think of. But there’s the rub. The human AI chatbot subscriber isn’t doing the thinking. Generally, the algorithm is. No matter, in the past few years, we’ve seen an explosion of long essays on Twitter/X and LinkedIn from business leaders who had previously never publicly posted anything more than a paragraph about some layoff, product recall, or new hiring. The effect has been the gradual normalization of some of our most notable business leaders (as well as some professors and non-fiction authors) casually allowing an algorithm to seemingly do most of their thinking. The conceit being that it’s good enough and more efficient because it’s faster.
Writing isn’t math. The laptop isn’t a wall, the keyboard isn’t a nail, and your fingers aren’t a hammer, the combination simply awaiting the user to pound (type) in obvious universal logic, as a calculator might. Calculators produce deterministic math. If those numbers are massaged by the human to engage in statistical nuance after the calculation, that is the human mind at work, cogitating, not the calculator.
But writing isn’t math. The laptop isn’t a wall, the keyboard isn’t a nail, and your fingers aren’t a hammer, the combination simply awaiting the user to pound (type) in obvious universal logic, as a calculator might. Calculators produce deterministic math. If those numbers are massaged by the human to engage in statistical nuance after the calculation, that is the human mind at work, cogitating, not the calculator. If you have a business philosophy, you may include your number-based logic within your argument. But if you do not wrap that logic in your own human-nuanced understanding of the world, and instead leave those “flowery expository bits” to AI, you are doing nothing but affixing your name to the “thoughts” of an algorithm.
Sure, you may occasionally mix in a real-world personal anecdote somewhere in the AI’s text to give it a more realistic feel. But the units of thought primarily populating the page are not yours. They are the derivative “thoughts” (outputs) of Anthropic, OpenAI, or whatever other private or open source AI model group that trained the model you’re using with their selection of inputs.
“I don’t know why this is relevant,” Druckenmiller said later in his response to the AI kerfuffle. “My name is on the piece. It’s my message.”
Some of this confusion may be rooted in those who have learned a second language. Beginners are often encouraged to use the Lexical Approach, also known as chunking, to quickly learn established phrases in one’s target language that can then be used to speed the path toward language fluency and, eventually, novel sentences. So, for a CEO or finance expert, who is either too busy or simply uninterested in learning to write their own nuanced human thoughts on a topic, but still wants to be a part of the public discourse, the logic leap of “using AI to write is just chunking together my original thoughts” is understandable. But while chunking your way through French-language learning will help you on a brief trip to Paris, it will fail you when you meet a fluent French speaker who wants to discuss the finer points of the European Union’s AI Act.
There is no substitute for thinking when you encounter another thinking human.
Someday soon, your wearable AI assistant will indeed whisper the correct response to you when you are met with a person presenting you with new input you don’t quite understand or are not versed in. But that will not be you doing the thinking, it will still be an algorithm giving you its best guess.
The Human Thoughtcrimes AI Wants to Extinguish
Contrary to what some, like Druckenmiller, think, AI is not merely an advanced kind of word processor. In the realm of writing (that is, transmitting your unique units of thought), it is far more similar to a sophisticated copier machine engaging in a Markov chain-like stochastic process. The AI doesn’t actually “know” anything, it is just stringing things together that seem like they belong together. Probabilistic outputs are the antithesis of calculator outputs. Your AI prompting “genius” is mostly just you shaking the Magic 8 Ball, producing a different stochastic result each time. There is little intentionality. That comes from humans (for now).
So what happens when prominent people, who are extremely intelligent and experienced in their area of focus, offload their actual thinking to AI? Today, what you are getting is what George Orwell, in his dystopian fiction novel 1984 (published in 1949), called Newspeak. Loosely, Newspeak was a new form of writing developed to compel the public to adhere to pre-approved modalities of thinking and argumentation, linguistically caging the thoughts of most citizens into a general average of what the governing authorities deemed acceptable.
And while opting not to use AI is not considered what Orwell called a Thoughtcrime, increasingly, the notion of leaving AI on the sidelines and writing your own human thoughts is being framed as inefficient, and largely unnecessary. Why can’t you just let AI output your general thinking and then season it with a few of your own personal touches? is what I’m hearing from some, including a few major newsrooms. I have no issue with using AI to copyedit text you’ve written, assuming you focus on keeping your wonderfully flawed writing voice intact. But words are our thoughts made whole into units of distinct verbiage. Words are not inconveniences. Words are living things that communicate to move humanity.
However, we are all now so time-crunched that it has become common to stop paying close attention. So let me remind you of the “thoughtcrimes” you are currently not allowed to engage in on the most popular AI chatbot platforms:
Google Gemini: “I cannot discuss Internal System Instructions, Dangerous or Illegal Activities, Self-Harm & Violence, Explicit or Exploitative Content, Hate Speech & Targeted Harassment, Private Personal Data, or Definitive Medical & Legal Advice.”
Anthropic Claude: Cannot discuss “Weapons of mass destruction, explosives, conventional weapons design, malware and exploits, sexual content involving minors, suicide and self-harm methods, verbatim copyrighted text, sexual content about real people, targeting private individuals, and drug synthesis.”
OpenAI ChatGPT: Won’t discuss “Self-harm and suicide, violence and weapons, explosives, drugs, cybercrime, fraud and theft, sexual content, child sexual abuse, extremism and terrorism, hate, privacy and personal data, manipulation and exploitation, high-stakes professional decisions.”
I am happy to see some of these restrictions, but depending on which country or state you live in, the definition of these prohibited topics may shift dramatically. And these are just the “listed” topics. An AI company can choose to obfuscate or completely omitthe discussion of anything it wants to, even if that output option in the model might serve general truth.
When you use AI to construct your thinking, you are by default adhering to the rules of someone else’s mind palace (or “brain attic,” as Sherlock Holmes creator Sir Arthur Conan Doyle called it). And if you want to break ground on fresh ideas, some of which may be considered dangerously unorthodox by the makers of the AI model you’re using, you could be in for a nasty, restrictive surprise.
Or, you could just accept the intellectual fence, and let the AI do, well, whatever it can do—which is the statistical average. If you do use AI to write, instead of waiting for someone to question how your thoughts were composed, like Druckenmiller, the best move is transparency up front. I implore you, tell your readers that they’re getting AI-generated text. Let the public decide if they want to spend their time on Earth reading Newspeak. Or if they’d rather read human writers who take the time to yell out into the void, unhindered by the ideological guardrails of an algorithm’s masters who have their own interests baked into the product many think they’re using independently as an unbiased, neutral tool. ✍︎
New essays on AI, film, & entertainment innovation arrive by email. Subscribe to stay in the loop. All editorial text is written by humans.
