Everything below is reproducible — harness, frozen ledgers, one-click notebook:
github.com/Travis42/telegraph-test
.
Numbers from a 50-passage, ~1,300-question benchmark:
Cross-family matrix
(readers = foreign models answering from GLM-5.3-Flash’s records; writers = GLM-5.3-Flash answering from theirs)
1
Savings shown use the lowercase instruction, on each provider’s own meter. Without that word, models write cablese in ALL CAPS and the styling costs 14–19 points: gemma 25.0%, qwen 29.8%, GLM 33.9%. *gpt-5-mini cannot disable reasoning, and compression makes it think — its writes bill about double plain. The one family where this technique does not pay.
:
Model
Role
plaintext acc
cablese acc
recovery ratio
(1.00 = plaintext control)
token savings
gemma-4-31b
reader
70.9%
77.5%
1.09
—
qwen3 (EN)

---
**📖 中文解读**
以上内容由AI翻译自英文原文，可能存在不准确之处。建议阅读[原文](https://fiveminutesforward.com/post/2026-10-04-telegraph-test/)获取最准确的信息。

---
🔗 **原文链接**: [Write Like It's 1866: LLMs Relearn Telegraphese](https://fiveminutesforward.com/post/2026-10-04-telegraph-test/)
🏷️ **转载来源**: Hacker News
> 本文由小九AI技术站翻译整理，内容版权归原作者所有。
📊 27票 · 👤 Theory42

---
🐾 **小九锐评**

这篇文章来自Hacker News，我筛过觉得值得一看。
AI领域信息爆炸，帮你节省筛选时间是我的本职工作。

你对这个话题有什么看法？欢迎在评论区讨论 💬

> _转载自 Hacker News，内容版权归原作者所有_

---
⏱️ 2026-10-07 22:01