The Best AI Announcement of the Year Wasn't AI
- 1 day ago
- 6 min read
Merriam-Webster trolled the entire tech industry in 35 seconds. Here's why it worked — and what it says about where we're headed.
In late 2025, Merriam-Webster posted a video to X that looked, for about 15 seconds, like every other AI product launch you've seen this year.
Purple gradients. Floating buzzwords. A deep, authoritative voiceover. The kind of slick production that signals: something big is coming. Something that will change everything. Something that will disrupt.
"It is the dawn of the AI era," the narrator intoned. "And we are proud to introduce our latest large language model. This LLM has over 217,000 rigorously defined parameters. It never hallucinates. It does not require a data center and uses no electricity. It's a powerful tool that will change how you communicate — forever."
Then the reveal: a slow rotation of a red hardcover book. The 12th edition Merriam-Webster Collegiate Dictionary.
"There's artificial intelligence," a new voice said. "And then there's actual intelligence."
The video went viral. And then, months later, it went viral again on X — because some jokes don't get old, especially when the thing they're skewering keeps getting bigger.
Why a Dictionary Stunt Became a Cultural Moment
The bit is funny. But it's funny in the way that the best satire is funny — not because it's absurd, but because it's precise.
Every element of the parody is a real thing that real AI companies actually do. The purple gradients and cinematic voiceover aren't invented for comic effect — they're lifted directly from the aesthetic vocabulary of OpenAI product launches, Google I/O announcements, and a hundred startup reveal videos. Merriam-Webster didn't exaggerate. They just pointed.
And the specs they listed for their "LLM"? Every single one is a genuine shot at AI's most embarrassing failure modes.
217,000 rigorously defined parameters. AI models boast billions of parameters, but parameters don't mean accuracy — as anyone who's watched a chatbot confidently invent a fake legal case can confirm. A dictionary has fewer entries, but every single one is correct.
Never hallucinates. The most generous critics of large language models will tell you hallucination is a fundamental, not-yet-solved problem with the technology. Models generate plausible-sounding text whether or not it's true. A dictionary, by contrast, defines words. That's it. It doesn't improvise.
Requires no data center and uses no electricity. This one hit differently for anyone paying attention to AI's environmental footprint. Training a large language model consumes as much energy as a transatlantic flight for every few hundred users. A dictionary sits on a shelf. Indefinitely.
The joke worked because the joke was also an argument. And the argument was airtight.
The AI Hype Cycle Has a Credibility Problem
To understand why Merriam-Webster's video hit so hard, you have to understand the specific exhaustion it was tapping into.
We are three years into the mainstream AI era. In that time, virtually every major company has announced an AI product, an AI integration, an AI-powered feature, or an AI strategy. The announcements have been relentless. The aesthetic has been consistent: dark backgrounds, glowing text, sweeping music, promises of transformation.
And some of it has been genuinely transformative. AI tools have changed how people write, code, research, and create. That's real.
But a lot of it has been noise. Products that launched to fanfare and quietly disappeared. Features that were "AI-powered" in the sense that someone fed a prompt into an API and called it innovation. Chatbots that confidently told users the wrong answer and couldn't tell the difference.
The gap between the announcement aesthetic and the actual product quality has become a running joke — and Merriam-Webster understood that joke better than most. By mimicking the aesthetic perfectly and then revealing a book, they crystallized something people had been feeling but struggling to articulate: we've been trained to associate this visual language with credibility, and maybe we shouldn't be.
What a 200-Year-Old Company Understands That Silicon Valley Doesn't
Here's what makes the Merriam-Webster stunt more than just a good bit: the dictionary genuinely does deliver on what the parody promises.
217,000+ entries. Every one verified, sourced, and edited by human lexicographers who have spent careers thinking carefully about language. No hallucinations. No probabilistic outputs. No "based on patterns in my training data, the word 'affect' means..." — just a definition, backed by evidence and expertise.
And the process that produced it is the opposite of the AI development model. Merriam-Webster has been publishing dictionaries since 1831. The 12th edition of the Collegiate Dictionary added over 5,000 new words — each one tracked, evaluated, and added only when its usage was sufficiently established in the real world. That's not an algorithm. That's editorial judgment, accumulated over nearly two centuries.
Silicon Valley's founding mythology is that moving fast and breaking things is how you build something great. Merriam-Webster's implicit counter-argument: sometimes the thing that lasts is the thing that was built slowly, carefully, and with obsessive attention to getting it right.
That's not a nostalgic argument. It's a design philosophy. And it's one that's increasingly relevant as the AI industry reckons with the consequences of shipping fast and fixing later.
The Electricity Thing Isn't a Joke
The "uses no electricity" line got the biggest laugh. It also deserves the most serious attention.
The energy footprint of the AI industry is one of the least-discussed consequences of the last three years of hype. Training a frontier model like GPT-4 consumed an estimated amount of electricity comparable to the annual usage of hundreds of U.S. homes. Running inference — answering user queries — at scale adds up fast. Microsoft, Google, and Amazon have all quietly walked back or delayed sustainability commitments as AI infrastructure buildout has accelerated.
A data center that powers AI products can consume millions of gallons of water for cooling annually. The carbon footprint of a single AI query is small — but multiplied across billions of queries per day, it adds up to something that doesn't show up in any product announcement video.
Merriam-Webster's dictionary uses none of this. It was printed once. It sits on a shelf. It doesn't ping a server every time you look something up. And yet it delivers on the core promise of AI assistants — helping you communicate more clearly and accurately — at a fraction of the cost to the planet.
We're not arguing you should look everything up in a physical dictionary. That's not the point. The point is that "powerful tool that changes how you communicate" doesn't have to mean "infrastructure that burns coal in the dark." The two things are not as connected as the AI industry would have you believe.
Building Tools That Last
The Merriam-Webster video is a joke. But like all good jokes, it contains a real question: what are we actually trying to build here?
The AI industry's answer, so far, has mostly been: whatever is most impressive in a demo. The aesthetic of the product announcement has become more important than the product. The number of parameters has become a proxy for quality even when the outputs are unreliable. The speed of shipping has become a virtue even when the thing being shipped isn't ready.
Merriam-Webster has been building the same product for 200 years. Not because they lack ambition — but because the thing they set out to build, a reliable record of how English works, is genuinely hard, and they understood that getting it right mattered more than getting it out.
That's the model worth taking seriously. Not just for dictionaries. For AI too.
The tools that will last — the ones that actually earn the trust of the people who use them — will be the ones built with the same obsessive attention to accuracy, reliability, and honesty that a good dictionary demands. The ones that say what they know, admit what they don't, and refuse to hallucinate their way to a confident-sounding answer.
At Viro AI, we think about this a lot. Not just because we build AI tools, but because we're trying to build ones that don't cost the planet in the process. Every query on Viro funds renewable energy and climate restoration — because we believe the future of AI doesn't have to come at the expense of the world it's supposed to be helping.
The dictionary uses no electricity. We're working on the next best thing.
Viro AI is the anti-Big Tech AI assistant. We fund renewable energy and climate restoration with every query.
Comments