Hacker News Reader

Parent Full thread

briandear

Add prompt to ChatGPT

Get answer.

Rewrite in your own words.

Feed back to chatGpT to check for errors.

Done. Watermarking really doesn’t solve any problem a clever person can’t trivially circumvent.

View on HN

Or easier: Use Llama or some other open weights model with a non-watermarking sampler.

Well spammers will probably skip the manual “ Rewrite in your own words” step.

So it’s still useful is reducing spam.

they will use google translate to translate to chinese and back and feed it back into chatgpt to fix grammar mistakes yielded :D (sorry. ur right its still useful, but spammers be spammers!)

Until it gets blocked as spam and then they will get a watermark stripping agent and bam. They are back in business.

Reply on news.ycombinator.com