OpenAI and Anthropic spent two years convincing venture capitalists that businesses would pay infinite money to make GPT-4 summarize emails. Companies bought in. They hired prompt engineers. They built workflows around something called "tokenmaxxing," which is what happens when a CFO learns a new word and decides to make it everyone else's problem.
Now those same companies looked at their bills. They saw six figures in API costs. They asked what they got for it. The answer was a chatbot that writes mediocre product descriptions and a Slack bot nobody uses. So they tightened their budgets. They started caring about efficiency. They stopped tokenmaxxing.
This is called a business cycle. You spend money on something stupid. You realize it's stupid. You stop spending money on it. OpenAI calls this "dampened growth rates." Normal people call it "we're not f*cking doing that anymore."
Anthropic built an entire company on being the safe, responsible AI lab. Their pitch was ChatGPT but with a conscience. Their customers were supposed to pay a premium for models that wouldn't say slurs or tell you how to make a bomb. Turns out nobody wanted to pay extra for that either.
The funniest part is watching AI companies pretend this wasn't obvious from the start. They sold infinite scale to people with finite budgets. They promised ROI to executives who couldn't define what the I even was. They built a market on hype and acted surprised when the hype required follow-through.
Retail traders saw this headline and bought the dip anyway. They think "efficiency" means the stock goes up. They think "dampened growth" is temporary. They're wrong, but they'll figure that out around the same time they learn what tokenmaxxing actually meant, which is never.
Photo by on Unsplash
Leave a Comment