Why Tech CEOs Sound Like Prophets on TikTok

A former OpenAI executive made a philosophical comment about AI in a New Yorker investigation. By the time it reached TikTok, he was summoning aliens through interdimensional portals.

The quote is real. The interpretation is not. And the gap between those two things tells you something worth understanding about how a large number of people now process information about technology.

This is not a niche problem. The clip has hundreds of thousands of views. The comments beneath it connect it to the Book of Genesis, to CERN, to Elon Musk quotes that do not exist. It is a live example of a pattern that repeats constantly, and mostly goes unremarked on.

What Was Actually Said

The line comes from Ronan Farrow and Andrew Marantz's investigation into Sam Altman, published in The New Yorker on April 13, 2026. In the piece, a former OpenAI executive describes the expansion of AI data centre infrastructure into the Middle East as "portals from which we're genuinely summoning aliens." The full context is a critical commentary about the pace of that expansion and what it means to build compute at that scale. The executive is calling it reckless. The "portals" are data centres. The "aliens" are a metaphor for a new kind of intelligence.

Clipped to eight seconds with a red graphic and Sam Altman's photograph, it becomes something else entirely.

The TikTok that spread the clip is not primarily about the New Yorker article. It is about the idea that powerful people are literally opening portals and summoning extraterrestrials. The original reporting, one of the more serious investigative pieces published about the AI industry in years, does not feature in the framing at all.

The Format Problem

TikTok, Reels, Shorts. These formats are optimised for attention. Not comprehension. Not context.

When your primary source of news is a 30-second clip, you are not receiving information. You are receiving a signal. Something that tells you how to feel about a topic before you have the background to evaluate it.

The scale of this is worth pausing on. Pew Research found in August 2025 that 1 in 5 American adults now regularly get news from TikTok, up from just 3% in 2020. Among adults under 30, that figure is 43%. No social media platform studied by Pew has seen faster growth in news consumption over that period. And of the TikTok accounts those users follow, only 0.4% are journalists or media outlets. The overwhelming majority of news consumed on the platform comes from influencers and content creators with no editorial accountability whatsoever.

This is not a character flaw in the people watching. Humans are fast pattern-matchers by design. Short-form video is engineered to exploit that efficiently. The platform rewards engagement, and engagement rewards alarm. A calm explanation of what a New Yorker profile actually argues does not perform. A clip implying a tech CEO is building alien portals does.

So the information environment selects for the most alarming interpretation of any given statement. Not necessarily through malice. Through mechanics. The algorithm is indifferent to whether content is true. It is very attentive to whether content produces a strong reaction.

The result is that the version of a story that spreads is almost always the version stripped of the context that would make it less frightening, less conspiratorial, less shareable.

The Metaphor Problem

Tech leaders speak in mythological language constantly. This is not new, and it is not always cynical.

Elon Musk said AI is like "summoning the demon" at an MIT symposium in 2014. Robert Oppenheimer quoted the Bhagavad Gita after Trinity. Scientists and engineers reach for religious and dramatic register when they are describing things that feel genuinely new and threatening. It is a way of making the abstract legible. A shorthand for: this is serious, this could go wrong, we are in territory without maps.

The people in those rooms understand that register. "Summoning the demon" means we may be creating something we cannot control. It is a warning about AI safety. Not a literal theological claim.

An audience that primarily consumes short clips does not always carry the vocabulary to decode that register. The metaphor lands as a statement of fact. The safety warning becomes a revelation. The person raising a legitimate concern about AI risk becomes, on a For You page, someone confessing to interdimensional summoning.

This is a failure of interpretation, not intelligence. But those two things are not as separate as we would like to think, and the data on where we are heading is uncomfortable.

The Intelligence Problem Nobody Wants to Name

The Flynn effect is the name given to the well-documented rise in IQ scores across the 20th century, typically measured at gains of two to four points per decade. It was, broadly, good news: successive generations were getting measurably better at abstract reasoning.

It has reversed.

A Northwestern University study published in the journal Intelligence in 2023 examined 13 years of cognitive data across 394,378 US adults and found consistent declining scores in three out of four cognitive domains between 2006 and 2018. The one domain that improved was spatial reasoning. Verbal reasoning, matrix reasoning, and letter and number series all declined. This is not a small sample or a contested finding. It is nearly 400,000 people over more than a decade.

The reversal has been observed across multiple countries. Norwegian researchers Bratsberg and Rogeberg documented declining IQ scores among men born after 1975, and found the decline appeared within families, pointing to environmental rather than genetic causes. Similar reversals have been recorded in France, the UK, Finland, and German-speaking countries.

What is declining is not narrow test-taking skill. It is abstract reasoning. The ability to interpret unfamiliar information, to hold a chain of logic together, to recognise when language is being used metaphorically rather than literally. The exact cognitive toolkit you need to watch a 9-second clip of a tech executive and understand that they are speaking philosophically, not prophetically.

The relationship between short-form video consumption and cognitive function is also increasingly well-documented. A 2024 EEG study published in Frontiers in Human Neuroscience found significant impairment to attentional functions in high-frequency short-form video users. A 2025 systematic review of 17 studies found that high-frequency use was consistently associated with attentional disruption, reduced executive functioning, and emotional dysregulation. These are not marginal effects. They are consistent across methodologies and populations.

The causal direction is contested. Do people with lower sustained attention gravitate toward short-form video, or does short-form video reduce sustained attention? Probably both. The point is that the environment and the cognitive decline are not unrelated, and the platforms scaling that environment have no incentive to slow down.

How Misinformation Gets Cited

There is a specific mechanism in comment sections worth naming.

Someone posts a misremembered fact. The creator of the clip replies: I remember hearing that too. Four hundred people like the comment. That number is visible to everyone who scrolls past.

Social proof functions as a proxy for truth. If 434 people agreed, it presumably corresponds to something real. Nobody checks. Nobody sources. The thread becomes the citation.

In the comments beneath the Sam Altman clip, a commenter states with confidence that Elon Musk called AI "the fallen Nephilim" in a TED talk, can't remember if it was 2011 or 2013. This talk does not exist. The clip's author replies: "I remember listening to him say that somewhere too." Four hundred and thirty-four people have now liked the exchange. For everyone encountering it downstream, that exchange is the source material.

This is not how misinformation used to spread. It used to require someone to write something false and distribute it. Now it requires someone to half-remember something in a comment, and someone else to say they remember it too. The infrastructure does the rest.

Who Is Actually Responsible

The people creating these clips know what they are doing. The deliberate framing, the sensational headline, the photograph selected for maximum unease, the real quote surrounded by fabricated context. That is craft applied with intent.

The viewers are a different matter.

If your media diet is primarily short-form video, and you rarely encounter long-form journalism, you do not have a working reference point for what a New Yorker investigation reads like, or why the paragraph before and after a quote changes what the quote means. That is not stupidity. It is a literacy gap produced by an information environment that has been systematically optimised against depth.

Short-form video is economically successful precisely because it holds attention without requiring comprehension. The business model works against the conditions that would let people distinguish a philosophical metaphor from a literal claim about alien portals. That is not incidental. It is structural.

And it compounds. People who form their understanding of a topic through short clips develop a framework for that topic. New information gets filtered through that framework. A long-form article explaining the actual context is less likely to be encountered, and if encountered, less likely to displace the mental model already formed by the clip. The News Literacy Project's 2024 survey of 1,110 US teenagers found that fewer than 2 in 10 could correctly distinguish between a news article, an advertisement, an opinion piece, and entertainment content. These are not difficult distinctions. They are foundational ones. And most young people currently cannot make them.

What This Means Practically

For anyone who communicates about technology: assume your words will be decontextualised. Anything that sounds remotely dramatic, supernatural, or ambiguous will be extracted and reframed. This is not a remote risk. It is the default outcome for anything that generates attention.

The AI industry has particular exposure here because it genuinely is building things that warrant serious philosophical language. The concerns about alignment are real. The concerns about power concentration are real. The people raising those concerns use language that is easy to misread, and the platforms carrying that language have every incentive to amplify the misreading.

For readers encountering this kind of content: the red graphic is a tell. Dramatic overlay text on a serious photograph is a format, not a source. A real quote does not validate the interpretation placed on it. These are separable things, and separating them requires going back to the original.

The original New Yorker article is one search away. It is a 16,000-word piece by a Pulitzer-winning journalist, based on 18 months of reporting and more than 100 interviews. It raises serious, substantive questions about OpenAI and how it operates. That story is considerably more interesting than anything a 30-second clip can convey.

That search rarely gets done. And the platforms benefiting from it not getting done are not going to start encouraging it.