Developers Are Making Claude Talk Like a Caveman to Slash Token Bills That Ate Uber's Whole AI Budget in Four Months
Engineers at OpenAI, Nvidia, and GitHub are installing a plugin that gags their AI into caveman grunts to cut output tokens by roughly 65%, because the industry's flagship product costs so much that Uber torched its entire AI budget in four months.
"less like a polite chatbot and more like a terse tool"
404 Media (Joseph Cox)
The most advanced software ever built is so ruinously expensive to run that engineers at OpenAI, Nvidia, and GitHub now bolt on a plugin that orders it to shut up and grunt, cutting output tokens by roughly 65% by stripping the chatbot’s groveling filler down to “Hulk smash.” Companies are deliberately making tools like Claude Code and Codex speak like cavemen to stop burning through AI tokens, turning the usually verbose output into something much more to the point. Uber blew through its entire AI budget in four months, and Legrand’s internal memo now points staff at the caveman plugin. A revolution in intelligence that has to be told to use fewer words so it does not bankrupt the f*ucking company is the whole bubble in one bit. 404 Media on the caveman plugin
Source: 404 Media (Joseph Cox) · Joseph Cox
More from The AI Bubble
OpenAI Spent $34 Billion to Earn $13 Billion While Leaked Books Hid the Worst Numbers
Zitron's follow-up shows OpenAI burned $34 billion to make $13.07 billion, with 6.6% of revenue ($867 million) coming from backer SoftBank and billions of costs vanished into a line item nobody can explain.
Every New AI Model Lands the Same, While Power Bills File Their Own Taxes
The miracle curve flatlined while the invoice kept right on sprinting.