That's friken hilarious.

It's likely that you'll get reduced performance from this, as blowing up the token count is part of getting a marginal amount of increased performance out of the models.

Funny though, as is most discoveries related to emergent LLM properties

can't you just append "make your reply as concise as possible" work?

I didn't realize you could ask for the 10 peso version.

Finally - the Grug brained LLM.

1. Full English so you understand and trust the output
2. Terse English to increase bandwidth <- you are here
3. A “native language” optimized for LLMs, similar to the language of the whispering earring

Finally - the Grug brained LLM.