AI Watermark - Detection & Removal? Worth Exploring?

Hi,
I was reading about AI content detection and came across AI watermarking where specific patterns of text are injected by the model in its writing as an identifier. Does anyone have any experience detecting it for specific models such as open ai or deep seek, etc? And around removing the AI watermark to pass AI text as human?

You might find this article interesting:

https://openai.com/index/understanding-the-source-of-what-we-see-and-hear-online/

I think manual humanize the text will work in this case to avoid Ai detection

Alexander301 said:

Hi,
I was reading about AI content detection and came across AI watermarking where specific patterns of text are injected by the model in its writing as an identifier. Does anyone have any experience detecting it for specific models such as open ai or deep seek, etc? And around removing the AI watermark to pass AI text as human?

View attachment 430734

Click to expand...

I don't bother with any AI content detectors - simply because you are only getting the opinions of a tool that has nothing to do with Google or any other entity that will be sending you traffic and/or commissions. However, I do have several filters that I pass my AI-generated content through before posting, and also have a few modifiers added to my prompts to limit each LLM in a few ways. If you're posting AI-generated content at scale, looking into these supposed 'watermarks' is a huge waste of time.

i saw that some tools had publicly released the watermark removal feature but later discontinued it

Steptoe said:

I don't bother with any AI content detectors - simply because you are only getting the opinions of a tool that has nothing to do with Google or any other entity that will be sending you traffic and/or commissions. However, I do have several filters that I pass my AI-generated content through before posting, and also have a few modifiers added to my prompts to limit each LLM in a few ways. If you're posting AI-generated content at scale, looking into these supposed 'watermarks' is a huge waste of time.

Click to expand...

I am glad to know that before I spent any more time investigating watermarks.

About the content filters and modifiers, are you at liberty to share some guidelines on those? I am also working on some AI-content and I even tried fine-tuning a few times to get a unique author tone but it didn't work as expected.

Alexander301 said:

About the content filters and modifiers, are you at liberty to share some guidelines on those? I am also working on some AI-content and I even tried fine-tuning a few times to get a unique author tone but it didn't work as expected.

Click to expand...

I've set up some scripts that parse my AI-generated content before posting - they filter out a lot of patterns/words that I personally don't like being used (lots of examples have been shared on the forum), whilst another script will paraphrase some sentences.

Some things you can add to your prompts are instructions to limit the language used in output, and I find that simpler content is better. So you can ask to limit the number of syllables used in each word, ask to limit to a certain Flesch reading ease score, etc.