LLMs Can Write Like It's 1866 and Cut Your Token Bill in Half
Write Like It's 1866: LLMs Relearn Telegraphese
A benchmark of 50 passages and ~1,300 questions shows that instructing LLMs to write in telegraphese—dropping articles and filler while keeping every fact—cuts token usage by 40–49% with no loss in downstream accuracy. Across four model families, compressed records were read at parity or better than plaintext, with recovery ratios of 0.99–1.10. The technique works because the register is already latent in training data, but it only pays off when compression happens after content is settled and for models whose reasoning can be disabled.
The expensive half of every LLM bill is optional overhead for machine-consumed text.
- OtherShrezzing
This page is (somewhat ironically) so extremely laden with Claude-speak that it's difficult to find the information in all the noise. But once you've waded through everything, you see these facts:
>What the test measures: A model is given a passage and a fixed set of questions with short, checkable answers — a date, a name, a count.
So, a model is given content which is especially amenable to compression, and asked to reproduce it under certain constraints, like...
>Why isn’t the plaintext baseline 100%? Answering questions about an uncompressed passage in plaintext scores ~91%.... a correct answer worded differently scores as a [failure]
Models can (and do) give objectively correct answers, but are penalised for not having some kind of omniscient knowledge of the implementer's phrasing preferences.
If this phenomenon is emergent in models, this benchmark is not proof of it in any meaningful way.
- Solomet
Newest LLM writing tell: Concepts are described in terms normally more appropriate for physical object.
> A lab that suppresses it in a frontier model just moves the advantage to open models that still _carry_ it
> they carry no signal about which is better
> where your workload _sits_ on that frontier should pick the point
> and no model _sits_ in the judge’s seat
> every ratio _sits_ at 0.99–1.10
Many many more examples of "sit"
> Every comparison in this post "holds" the questions
I have been seeing this a lot in my recent work with LLMs and it is quite frustrating. Even more frustrating is how frequently it uses low-signal terms for things unnecessarily. These 'physical object' terms are one example but at times it really seems that they 'preserve effort' by choosing a less descriptive term because it 'fits'
I have also caught it replacing descriptive terms with more vague ones for no discernible reason other than laziness.
"Minimize ambiguity" has been my go-to instruction as of late when the agent drifts back towards vague terms and lack of specificity.
- netsharc
The Cablese/Telegraphese is more interesting than the use in LLM. DuckDuckGo'ed "paromella":
https://en.wikipedia.org/wiki/Commercial_code_(communication...
Some codes I found interesting:
> INSANE - at what price, free on board and freight, can you offer us cotton for shipment by steamer sailing this week?
> COGNOSCO - dining out this evening, send my dress clothes here
Useful codeword!
> ANNOSUS — Confined yesterday, Twins, both dead, Mother not expected to live
How often did that one come into use??
- bilater
Actually, I think the lesson from this article is that a lot of these little hacks we’re doing around context compaction, AGENTS.md, system guides, and other harnessy ways of saving tokens are going to go away fairly quickly, just as all the tricks of the telegraph era went away once communication got cheap enough that it was easier to just speak in plain language.
- z2
From recent ChatGPT (GPT5.6) conversations where I've seen occasional reasoning leaks into the UI, it's clear that something like this is already implemented, and I'd speculate that this is the majority of recent claims of less token usage. Not sure if they are literally prompting for cablese of course.
"Need check output vs prev. Ran script, results fine, need prep next step. Ready? Go."