LLMs are not nothing.

 Given the state of things in 2026 and the two "next big things" that preceded generative AI, this is by no means a trivial point.

The metaverse was next to nothing. Web3 effectively was nothing. In both cases, an astounding and profoundly embarrassing level of hype was supported by nearly non-existent substance. 

Large language models are not like the other two. They are a powerful set of tools and represent a fantastic leap forward in natural language processing. There is plenty here to be legitimately excited about without resorting to exaggeration. Unfortunately, bullshitting is habit-forming, particularly when you have the same lack of guardrails and even worse incentives.

With almost 20 years having passed since the debut of the last world-changing piece of consumer technology, boosters have grown extraordinarily adept at building economic mountains out of technological molehills, and the press has gone beyond mere complicity to become active collaborators in the construction of these imaginary peaks.

Many have given themselves over completely to fanboy gush, but the more respectable press, represented as always by the New York Times, adopted a veneer of sobriety that, if anything, made matters worse. No matter how obviously absurd and self-serving the crap Silicon Valley tried to sell was, writers like Kevin Roose would pass it along with a framing of just asking questions/we need to keep an open mind/important if true. When faced with obviously absurd and self-serving crap, we do not need to keep an open mind; we need to be ready to call bullshit and to recognize the person behind it as a grifter.

For the hyperloop generation of tech and business reporters, hyping non-stories into major innovations was their job,  what readers wanted, what editors demanded. When handed a major innovation, perhaps it's not surprising they instinctively hyped it into something epochal.