It's by default so you use all those tasty tokens.

Kinda wish there was a deterministic, mostly terse, language to interact with computers

It's tied to the design. With humans, you have a train of thought which you can choose to represent in various ways--or not reveal them at all. In contrast, LLMs are make-document-longer machines being run over and over on alternating revisions of the document. Insofar as one might try arguing they have a "train of thought", it's made of the words/tokens.

Everything they (don't-)emit is partly for the benefit of the next run, a clue or signpost (not-)present. Documents may be wordy as a form of concept-emphasis and consistent direction as opposed to a form of communication to the human.

So a terse effect may require a layer of indirection and trickery: There's a verbose document (you'll still be charged for the tokens) with portions that are not "acted out" to the end-user. Imagine a film-noir movie script, where AI Detective's "I know Mickey couldn't have done it because" monologue is hidden, versus their terse dialogue "Too early to say."

> IMHO the overly-verbose default style of LLMs is the most annoying part of interacting with them, and I wish their masters would just tell them to be terse by default.

They don't know how to e terse. I've tried that a few months ago and gave up because the responses were almost incomprehensible!

I want to see more operators try https://github.com/juliusbrussee/caveman

How does it affect agent accuracy?

They ramble on because those words are for them, not for you. There is some amount of hiding this through "thinking" modes that are hidden by default, but still you have to remember that ALL THEY ARE are complex statistical machines for predicting the next symbol.

Caveman mode legitimately works

Produce pre-compressed output in the harness?

No thank you. I want information when it’s working on things and what (atleast codex) does right now works for me.

[dead]