1 of 1
Story summary
- OpenAI, Nvidia and GitHub use the Caveman plugin to rewrite LLM output into terse language while keeping code details.
- Creator Julius Brussee says Caveman reduces tokens by about 65 % to 75 %.
- A test saved roughly 5,800 tokens, a 65 % cut.
- Legrand memo urges staff to use Caveman to stay within AI budgets after Uber’s CTO reported the budget was exhausted in four months.
