Researchers identify new 'filler words' to recognize AI-generated texts

Earliest collected report:Mezha.Media
yesterday at 14:55

Last update yesterday at 21:45
3 sources

Researchers from Graphite discovered new linguistic clichés used by modern AI models. They identified nearly 13,000 words and phrases that appear in chatbot texts more frequently than in human writing.

In brief

Researchers identified new 'filler words' in AI texts.
Modern AI models are shedding noticeable linguistic patterns.
New characteristic words and phrases are emerging in AI texts.
The study covered nearly 13,000 words and phrases.

Covered by 3 of 6 media groups

Budget not in the collected reports
Public money not in the collected reports
Media business 1 source
Big capital 1 source
Readers and grants not in the collected reports
Undisclosed 1 source

No newsrooms from these groups in the collected reports: budget, public money, readers and grants.

What are media groups?

Share of reports by media group
Group	Share	Count
Media business Newsroom 1	34%	1 item
Big capital Newsroom 1	33%	1 item
Undisclosed Newsroom 1	33%	1 item

How we count media groups — Methodology

Who the outlets write for: community 0
region 0
country 3
world 0
Channels: Websites 2
TV and radio 1
No collected reports from: Суспільне Новини, 24 Канал, NV, Укрінформ, Цензор.НЕТ
23 more
УНІАН, Главком, Обозреватель, LB.ua, Апостроф, Інтерфакс-Україна, Фокус, Українські Національні Новини, Informator, Шелтер, Корреспондент, Ділова столиця, Ми — Україна, ГОРДОН, Українська правда, Букви, ArmyINFORM, Оперативний ЗСУ, Європейська правда, Главред, Телеграф, Delo.ua, РБК-Україна
Collection may be incomplete. This does not establish intent.

How we measure speed, distance and channels

Report an error

Mezha.MediaMedia business
- first item yesterday at 14:55
- earliest
- national desk
Дослідники визначили нові "слова-паразити", за якими можна розпізнати текст ШІ
ТСНBig capitaloligarchic ties
- first item yesterday at 16:15
- in the wave, 1 h later
- national desk
Ці слова видають ШІ: дослідники з’ясували, як розпізнати тексти різних моделей
InternetUAUndisclosed
- first item yesterday at 21:45
- caught up, 6 h later
- national desk
Дослідники назвали нові слова й фрази, які можуть видавати текст, написаний ШІ

Join the discussion — sign in to write a comment.

No comments yet.

Start with what is missing: context, or a fact from another source — with a link to it.

A mistake in the event itself goes through “Report an error” below it — a moderator reviews it.

Comments

Loading comments…

The story's family tree

The story's reports in order: the earliest collected first, then forwards and reports similar to earlier ones, each under the report it was linked to.

3 reports
2 similar to earlier ones

earliest collected report: — the earliest Viche collected; someone may have reported the event earlier
forwarded from …: — a forward Telegram shows directly
similar to an earlier report: — a close text, not proof of rewriting: both may draw on the same press release or agency report
no matched predecessor, or not compared: — this does not show the report is original

Mezha.Media
- earliest collected report
- yesterday at 14:55
Дослідники визначили нові "слова-паразити", за якими можна розпізнати текст ШІ
1. ТСН
  - similar to an earlier report
  - 1 h later
  Ці слова видають ШІ: дослідники з’ясували, як розпізнати тексти різних моделей
2. InternetUA
  - similar to an earlier report
  - 6 h later
  Дослідники назвали нові слова й фрази, які можуть видавати текст, написаний ШІ

How we build the family tree