A few years back a company had an ad campaign with a discouraged caveman who was angry because the company claimed their website was “so easy, even a caveman could do it.” Maybe that inspired [JuliusBrussee] to create caveman, a tool for reducing costs when using Claude Code.
The trick is that Claude, like other LLMs, operates on tokens. Tokens aren’t quite words, but they are essentially words or word fragments. Most LLM plans also charge you by the token. So fewer tokens means lower costs. However, LLMs can be quite verbose, unless you make them talk like a caveman.
For example, here is some normal output from Claude:
Sure! I’d be happy to help you with that. The issue you’re experiencing is most likely caused by your authentication middleware not properly validating the token expiry. Let me take a look and suggest a fix.
After Caveman that is reduced to:
Bug in auth middleware. Token expiry check use < not <=. Fix:
Turns out, you can understand the caveman output just fine.
There are a few rules. Caveman removes filler words, but knows to keep technical terms, code blocks, and error messages intact. It also outputs normal messages for things like commits.
As the website says: “Caveman not dumb. Caveman efficient. Caveman say what need saying. Then stop. If caveman save you mass token, mass money — leave mass star.”
If you want to code with an LLM, make sure you understand the terms of service. If you want to learn more about how LLMs work with tokens and other details, but want to skip the math, we can help with that.
Why Use Many Word When Few Do Trick?
Username checks out.
Corporate greed and the cobra effect usually seem to go hand in hand, the more the arms race goes back and forth, the worse the user experience gets.
The Grug Brained Developer was onto something!
https://grugbrain.dev/
The tool is basically an extensive prompt in a skill.md file so the LLM has this capability available when needed. It’s often interesting to look at the prompt:
Grammar
Drop articles (a, an, the)
Drop filler (just, really, basically, actually, simply)
Drop pleasantries (sure, certainly, of course, happy to)
Short synonyms (big not extensive, fix not “implement a solution for”)
No hedging (skip “it might be worth considering”)
Fragments fine. No need full sentence
Technical terms stay exact. “Polymorphism” stays “polymorphism”
Code blocks unchanged. Caveman speak around code, not in code
Error messages quoted exact. Caveman only for explanation
The complete skill can be viewed here:
https://github.com/JuliusBrussee/caveman/blob/main/caveman/SKILL.md
And after using this for a while, folks will start talking the same way. I can’t wait…
exactly , and by time it will get simpler and simpler , and humans and their ability to express themselves abstract will evolve backwards and with that culture ,skills and knowledge is gone .
no know you mean
make more easy
Just like in the movie Idiocracy (2006)
“I’m just a simple caveman, Your Honor. Your modern technology like phones and cars frighten me! But even I can see that my client isn’t guilty of what he have been accused of.”
Why are you dancing around the company name that did the original ad? It is Geico. Jeebus, people…
https://www.youtube.com/watch?v=e8aj1AlYvxI
Why harp on the play on words without giving proper credit/citation?
“Why harp on the play on words without giving proper credit/citation?”
Uhhh… maybe because the source was so obvious even a caveman would get it?
Hugo hungry. Hugo gather. Haku, paku. Infoberries, nyam, nyam.
Raugh!
Fire…. mediocre
As an idea guy with no game—I can at least come up with some names
THAL
HABILIS
Frankenbaud