Caveman: Why use many token when few token do trick

A tiny repo, a big question

A new GitHub project called "caveman" — title blunt and charming — has been circulating on Hacker News and sparking a bit of a debate. It has been reported that JuliusBrussee posted the repository under that name with the tagline you see above: “Why use many token when few token do trick.” The idea is simple on the surface: use fewer tokens to get the same result. Simple, but provocative. Who doesn’t love a good shortcut?

Minimalism meets prompt engineering

The broader pitch ties into two hot threads in AI right now: prompt-efficiency and cost control. Allegedly, caveman demonstrates techniques for making prompts or interactions with large models more token-efficient — less verbosity, smarter framing, maybe compact encodings. That’s appealing because tokens cost money and time. Think Occam’s razor for prompts: shave off the fluff and keep the useful bit. Sounds like KISS, right?

Community reaction and caveats

Hacker News users have been split. Some cheer the elegance — a return to basics, a caveman grunting his way to savings — while others warn that fewer tokens can mean brittle prompts or lost context. It has been reported that commenters flagged trade-offs: efficiency versus robustness, short prompts versus nuanced outputs. If nothing else, caveman taps into a real pain point: teams that pay for model calls are hungry for tricks that cut bills without cutting results. Will this be a practical tool or a clever thought experiment? Time (and tests) will tell.

Sources: github.com/juliusbrussee, Hacker News