Bridging the Gap Between Hype and Profit
▼ Summary
– The article uses the “Underpants Gnomes” meme from South Park to describe AI’s uncertain path, where companies have built the technology (Step 1) and promise transformation (Step 3) but lack a clear plan for Step 2.
– Activist group Pause AI argues that Step 2 must involve regulation, though specifics on what it will call for and who will enforce it remain debated.
– AI boosters, like OpenAI’s chief scientist, envision Step 3 as salvation through economically transformative technology, but they disagree on the routes to get there.
– A study from Anthropic predicted job impacts from LLMs, but its findings are based on guesses about task suitability rather than real workplace performance.
– Another study by Mercor tested AI agents on workplace tasks for bankers, consultants, and lawyers, finding that every agent failed to complete most duties.
Produced by Pause AI, an international activist group that helped organize the protest, the message ended with a direct appeal: “Pause AI until we know what the hell Step 2 is.”
That line echoes a classic bit of pop culture. In the 1998 South Park episode “Gnomes,” the boys discover a band of nocturnal creatures stealing underpants. When asked why, the gnomes reveal their business plan: “Phase 1: Collect underpants. Phase 2: ? Phase 3: Profit.” The trio’s pitch deck has since become a staple internet meme, used to mock everything from startup ambitions to policy blueprints. Even Elon Musk, the meme lord in chief, once referenced it while discussing how he would fund a Mars mission.
Today, that same question mark defines the state of artificial intelligence. Companies have built the technology (Step 1) and promised sweeping transformation (Step 3). But the path from here to there remains a glaring unknown.
For Pause AI, Step 2 must involve some form of regulation. What that regulation looks like, and who will enforce it, are still open questions.
On the other side of the debate, AI boosters are convinced that Step 3 means salvation, and they tend to gloss over the messy middle. They envision a future of sunny uplands powered by an “economically transformative technology,” as OpenAI’s chief scientist, Jakub Pachocki, told me recently. The destination is hazy and still far off, but everyone is taking a different route. Whether any of them will actually arrive is anyone’s guess.
For every grand prediction about what’s coming, there is a more grounded assessment of how the rubber meets the road,one that deflates the hype. Consider two recent studies. One from Anthropic tried to forecast which jobs will be most affected by large language models. The takeaway: managers, architects, and media professionals should brace for change; groundskeepers, construction workers, and hospitality staff, not so much. But those predictions are essentially educated guesses, based on what LLMs seem capable of in theory rather than how they actually perform in the workplace.
Another study, published in February by researchers at Mercor, an AI hiring startup, tested several AI agents powered by top-tier models from OpenAI, Anthropic, and Google DeepMind. The agents were given 480 workplace tasks commonly performed by human bankers, consultants, and lawyers. Every single agent failed to complete most of its duties.
Why such stark disagreement? Several factors are at play. First, it matters who is making the claims and why. Anthropic has skin in the game. Moreover, most of the people telling us that something big is imminent have reached that conclusion largely based on how fast AI coding tools are improving. But not every task can be cracked with code. Other studies have found that LLMs are poor at making strategic judgment calls, for example. The gap between hype and profit remains wide, and Step 2 is still anyone’s guess.
(Source: MIT Technology Review)
