AI Industry Goes Extreme

TThe story progresses very quickly.that it may all appear, at first glance, as if it were destined. After moving to Columbia last fall, as Chungin "Roy" Lee recounts, he utilized AI to bypass his academic requirements, and also used AI to cheat during internship interviews at Amazon and Meta—receiving offers from both—and in the winter shared his tool on social media. He was put on probation, suspended, and, more interested in AI than in his studies, dropped out this spring to launch a start-up. The start-up, Cluely, promotes the capability to “cheat on everythingUtilizing an AI assistant that operates in the background during meetings or sales calls. Last month, it concluded a $15 million funding round headed by Andreessen Horowitz, the renowned venture-capital company. (Columbia, Meta, and Amazon chose not to comment publicly regarding Lee's situation.)

Lee unreservedly holds the view that the emergence of all-knowing AI is unavoidable, with bots soon taking over every job. The use of the term "cheating" is merely a bold method to get everyone on board with this concept, Lee mentioned during our recent conversation. "We have no choice but to keep spreading the message: Do not consider it cheating," he said. ("Every time technology makes us smarter, the world panics. Then it adapts. Then it forgets. And suddenly, it becomes standard," states Cluely on its website.) Lee pointed out that while some may find it unjust if others can utilize AI to become "1,000 times better or more efficient," this will eventually just be the norm. Even if ChatGPT doesn't become any more capable than it is now, "every single white-collar job in America should already be gone," Lee said (or at least "conservatively," 20 to 30 percent of them). And "I would bet my entire life on AI becoming exponentially better."

As we discussed over Zoom, Lee was occasionally eating corn chips while talking about superintelligence, and his tone started to sound familiar. He seemed very similar to OpenAI CEO Sam Altman. Both founders approach selling a product as if they're spreading a belief. In arecent essay, Altman stated that the singularity—the era following which technology surpasses human control and understanding—has already started. "The pace of technological advancement will keep increasing, and it will remain true that people can adjust to nearly anything," Altman wrote. "There will be significant challenges, such as entire categories of jobs disappearing, but at the same time, the world will become vastly wealthier at an accelerated rate, allowing us to seriously consider new policy ideas that were previously unimaginable."

AI enthusiasts are prevalent throughout the Bay Area. I've encountered dozens of them: individuals who hold the belief thatAI’s rapid ascensionis unavoidable and by farthe the most significant event occurring on this planet. (Some have mentioned it's the sole thing worth paying attention to.) Their perspective is somewhat hopeful—the notion, though perhaps simplistic, is that superintelligence will ultimately enhance life for all—which enables them to quickly overlook the immediate negative consequences (such as)job loss and resource guzzling). AI start-ups promise"complete automation of the economy," "limitless connectivity" with millions ofAI personas, “limitless” memory, a solution to “all diseaseRecently, several AI researchers and founders have shared with me that they are reconsidering the importance of formal schooling: One entrepreneur mentioned that current bots may already be more academically proficient than his teenage son will ever become, causing him to question the worth of a conventional education.

[Read: The tech hub prepares for turmoil]

However, the radicalizing impact of AI extends beyond its supporters. In response to the growing rhetoric from Silicon Valley, AI skeptics have intensified their own arguments, similar to atheists challenging the priest during a church service. They view AI as exaggerated and largely ineffective, predicting its inevitable failure. One of the industry's main critics, computational linguist Emily Bender, recently co-wrote a book titledThe AI Conand promotes using the term "a racist pile of linear algebra" when talking about chatbots—a nod to well-established algorithmic prejudices against individuals of color—or "stochastic parrots." Gary Marcus, another key critic of the AI field and a cognitive scientist at NYU, recently shared one of his main arguments with me. Do chatbots possess intelligence? "I mean, you could argue your calculator thinks, depending on how you interpret the word"thinking,” he said.

The two groups are increasingly engaging in direct confrontation. A few days prior to our conversation, Marcus had initiated another online dispute with the AI industry by posting an altered version of...imagedisplaying Altman's face superimposed on a photo of the notorious Elizabeth Holmes. "True performance art," Altman remarked with a joke. Ed Zitron, a well-known AI critic, recentlywrotea nearly 7,000-word essay arguing that he is "sick and tired of everyone acting like generative AI is the next big thing," according to the political analyst Nate Silverdescribedlike "old man yells at cloud vibes" and "out of touch with reality."

This conflict has gone beyond reality, and maybe even proof, into a battle between different worldviews. There are now two separate AI realities, and most of us find ourselves in the space between them.

There have been disagreementsthroughout the history of AI, there have been debates between supporters and critics. However, in recent months, this discussion has become more heated as the industry rapidly spreads across the digital world. Billions of people are now expected to come into contact with generative AI daily via platforms like Google, Facebook, Instagram, X, their iPhones, Amazon product summaries, various voice assistants, and other services—not necessarily because they desire it, but because it's unavoidable. Many individuals are also actively looking for these tools. ChatGPT is now thefifth-most-visited website in the world, and OpenAI’s new image generatorwas reportedly utilized by over 130 million individuals in its initial week, placing amassive strainon the company's servers. (Anyone who controls the White House X account was one of those individuals, sharing anMeme created by AI showing an immigrant crying being detained by ICE.)

As technology and its results become widespread, AI leaders have become more vocal and bold regarding the significance of the technology. Two weeks ago, Jack Clark, a co-founder of Anthropic,warnedCongress that there may be around 18 months until the emergence of "truly transformative technology"—AI systems that surpass any current chatbot or cognitive ability. The day following Donald Trump's second inauguration, Alexandr Wang, the newly appointed chief AI officer at Meta,wroteto the president that the United States and China are engaged in an "AI competition."

The harsh language is accompanied byextreme spending. The technology sector has collectively spent hundreds of billions of dollars since the emergence of ChatGPT to develop more advanced AI systems and establish the necessary physical infrastructure, and there's no indication it will slow down. In recent weeks, Meta CEO Mark Zuckerberg, seemingly eager to keep pace in the AI competition, has been actively hiring in a recruitment drive that has seen him...reportedlyProvided nine-digit compensation packages to leading researchers. (Meta claims the figures have been overstated or misunderstood.) It remains entirely uncertain how generative AI will generate revenue, yet technology companies appear confident that funds will come in once the technology has fundamentally transformed the world. Regarding the skeptics: “When the AI bubble pops, I don't think the tech industry is prepared for how many people will truly enjoy it,” Zitron said.wrote last week.

There might be no more accurate example of the division than the reaction to a recentpaper, published by a group at Apple, titled "The Illusion of Thinking." The researchers provided advanced AI systems, referred to as "large reasoning models," from OpenAI, Anthropic, and DeepSeek with various tasks to complete: arranging checkers based on a specific pattern, for example, or restacking blocks in the fewest moves possible. All the puzzles could be solved using the same fundamental logic, regardless of their size—there was no change in the process for rearranging the blocks, even if more blocks needed to be moved. However, these "reasoning" AI models completely failed once the puzzles became large enough. "That's kind of like a young child claiming, I'm actually an excellent mathematician, but I can't add these numbers you're asking me to add because I don't have enough fingers and toes," Subbarao Kambhampati, a computer scientist at Arizona State University who wasn't part of the study, told me.

[The time of the GPT era has already passed.]

Kambhampati has been leading the investigation into the capabilities and shortcomings of "reasoning" models, and for him and others with similar views, including Marcus, the Apple paper confirmed longstanding concerns. "What I've been warning about as a critical weakness in the field for 30 years is real," Marcus said to me. "I can't deny there's some validation in that." According to this perspective, generative-AI models are not "thinking" beings but statistical approximators, highly skilled at reusing patterns from their training data but little else. The original ChatGPT had difficulty counting, and even today's version of ChatGPT struggles with certain basic puzzles.

Yet many supporters of AI reacted to the Apple paper with delighted mockery. In onememeShared in a large AI discussion group on Facebook, giant robots burn a city down while a group of humans nearby comment, "But they aren't really 'reasoning.'" Why does it matter if AI "thinks" like a person if it's better than you at your job? In fact, some critics of the paper suggested, the results just showed how human-like AI models have become through their flaws. (Who among us hasn't struggled with a long, complicated problem at times?)

Marcus's boasting about the paper on X made him a target for those who recognize AI's capabilities, including Altman, whowrote, “We deliver, he keeps telling us to get off his lawn.” Kevin Roose, a tech journalist atThe New York Times, took his own shot at Marcus, replying to Altman's post: “A man predicts 85 out of the last 0 AI failures, and this is how you respond?”

Roose's remark seemed especially insightful to me; he doesn't fully love the technology as Altman does, but he does see it as strong and here. His latest work for theTimes has concentrated on topics such aswhat actions to take if artificial intelligence systems gain awarenessand if artificial intelligence could present a fundamental threat to humanityin a few years. He is composing a book about thecompetition to develop artificial general intelligence," a version of the technology that is equal to or surpasses human abilities. More recently, he haslikenedSome AI skeptics have been compared to "an antinuclear movement that refused to acknowledge fission was real." When I contacted Roose to inquire about this strong position, he said, "Lately, I think those who are dismissing the abilities of these models are just sharing comforting tales to people who don't want to accept that change is on the way."

Tthe struggle between proponents of AI and atheistsPerhaps it will continue for a while. Generative AI is complex, and the terminology used to explain it is unclear—does it qualify as "intelligent" or "conscious," or both, or neither, and does that distinction matter? The companies developing this technology also avoid offering clear definitions or specific benchmarks for "general" or "super" intelligence. "We don't even know how to formulate the right questions to grasp these concepts," Kambhampati stated. Without proper questions, let alone answers, belief steps in to fill the gap. Everything can be interpreted to back either perspective in the discussion.

Autonomous and industry studies—conducted by Kambhampati, Bender, Apple researchers, and many more—have consistently demonstrated chatbots struggling with different tasks:basic arithmetic, logic, conceptual reasoning, you name it. However, technology companies also frequently createchatbotsthose that are better, occasionally by a large margin, at the same tasks. Is there a fundamental, widespread issue with generative AI, or is the technology moving along a trajectory toward endless progress? You could present a case for either perspective, using the very same data, and individuals do this regularly.

[Read: The future plans of Big Tech in artificial intelligence are becoming clearer]

The issue with the radicalization of AI is that it encourages individuals to consider possibilities outside the current material realities of the world. In truth, AI models are acceleratingscientific discovery and software engineering while also fabricating information and pushing people into mental breakdowns. Disregarding the chatbot era or claiming the technology is ineffective diverts attention from more complex conversations regarding its impact on jobs, the environment, education, personal relationships, and other areas. Perhaps even more concerning, believing that superintelligence is imminent leads to downplaying nearly any issue related to the technology as it exists today.

Underneath countless layers of online hostility, there could still be common ground between the two groups. For all his loud rhetoric online, for example, Marcus has acknowledged that current chatbots represent a genuine advancement, albeit one that is still far from...themajor advance; despite Altman's frustration, OpenAI's most recent large-scale reasoning models utilize new methods that are not significantly different from Marcus's concepts developed decades ago. Artificial intelligence can be extremely powerful yet also highly problematic, Kevin Roose mentioned to me. "What I am not implying is: We should accept the industry's claims at face value," he stated. If OpenAI is genuinely "confident we know how to create AGI," as Altman wrote this year, he needs to demonstrate it.

Ultimately, the current form of generative AI was not predetermined. When the concept of "artificial intelligence" began in the 1950s, there were two primary approaches: "Connectionists" believed that digital "neural networks" learning from data would be enough to create intelligence. "Symbolists" argued that intelligence could only arise from pre-programmed rules, logic, and knowledge. Neural networks prevailed: They serve as the basis for today's chatbots and underpin much of the modern technology sector.

Firms like Meta and Google invested heavily during the 2010s in expanding neural networks and data centers to support digital ads, social media platforms, search engines, and shopping algorithms, among other services. As users became increasingly engaged with these offerings, technology companies gathered massive amounts of data, which they leveraged for significant financial gains. These extensive datasets have now become a valuable resource for developing chatbots.

In 2023, scientists from MITfoundthat 70 percent of individuals with Ph.D.s in artificial intelligence enter the private sector, and that nearly all of the most significant, and consequently influential, AI models are developed by companies. With hundreds of billions of dollars already poured into generative-AI products andprofitabilityStill seemingly many years off, these companies cannot afford to display any signs of vulnerability. They have become more extreme, at least in part, because they need their vision to materialize. Even Lee, toward the end of our discussion about Cluely, acknowledged some skepticism: "Sure, it's a strategy to attract venture capitalists, but that's only after gaining the attention of hundreds of millions of everyday people." He once again reminded me of Altman, whose skill in crafting and leveraging a narrative has turned OpenAI from a research lab into a producer of new AI products.

As we discussed radicalization, Lee brought up another idea that caught my attention. He suggested, what if "half of America had moralized against the internet and technology, while the other half fully embraced it." One half would "be living as though electricity had never been discovered," while the other would be illuminated with success. "There would be a huge disparity in results," Lee explained. "This is experiencing a dystopian society. Such inequality is absurd."

Of course, half the country did not reject the internet, let alone electricity. And a "crazy" inequality will have been present long before the hypothetical emergence of superintelligence, much of it fueled by technology. Automation isresponsibleAt least half of the nation's increasing wage gap over the last 40 years, as stated by an economist. Tens of millions of Americans and billions of people worldwide.lackHigh-speed internet connectivity. Companies like Amazon, Uber, and Airbnb have eliminated whole categories of businesses without providing viable, fairly compensated alternatives. The ten wealthiest technology billionaires are valued at almost $2 trillion, exceeding the gross domestic product of all but 11 nations globally. Whether or not a technological singularity occurs, Silicon Valley has already created an alternate reality.