News for hackers. Stuff that matters.
Re:Check if a file was made with Claude by qurren on Wednesday September 02, 2026 @12:59PM (#49539166)
Up next:

1. Generate a bunch of responses with both Claude and various non-Claude LLMs (ChatGPT, Gemini, Kimi)

2. Train a discriminator model that can differentiate Claude vs. non-Claude

3. Train a de-watermarking model using the discriminator model as loss

Re:Check if a file was made with Claude by hughw on Wednesday September 02, 2026 @01:08PM (#49539273)
Or just write a 6 line program to remove the metainfo from the file?
Re:Check if a file was made with Claude by rcktmrtn on Wednesday September 02, 2026 @01:53PM (#49539900)
Your observation is sharp, but it's not just metainfo—it's load-bearing text. To remove it, you need to delve deep and alter the tapestry of carefully selected words.

edit: almost forgot the "—"

Re:Check if a file was made with Claude by Lockal on Thursday September 03, 2026 @02:25AM (#49546549)
s/text/prose/ s/'/’/

The honest seam is obvious: claudish works; imitation does not. That’s not nothing.

Re:Check if a file was made with Claude by aero-glide2 on Wednesday September 02, 2026 @01:15PM (#49539355)
very difficult to remove text watermarking
Re:Check if a file was made with Claude by HoldOnAMinute on Wednesday September 02, 2026 @03:44PM (#49541385)
There must be some set of outputs that is completely resistant to watermarking. For example if I told Claude to output "Hello, World"
Re:Check if a file was made with Claude by ProfessorLayton on Wednesday September 02, 2026 @02:06PM (#49540084)
Outputs can be easily translated from the watermarked language to another, either by just knowing two+ languages, or via non-LLM translation.
Re:Check if a file was made with Claude by DonsDiscountGas on Thursday September 03, 2026 @06:33AM (#49548267)
You just described how Pangram works.