Ask HN: Is there a browser plugin that will fade AI generated text to gray?
I've seen some that will annotate text but nothing that will fade AI generated text out to the background color depending on the level of confidence that a text was human written or AI written. Does anybody have something like this or have it in the works? If so is it local or does it require some service?
13 comments
Are there really apps than can distinguish AI vs human? I suspect that at best they can only id a small fraction of what is being written.
There are ones that purport to, but none actually do. The false positive rate is crazy high.
If that's the case then they have little use. We'll be reading AI created text from now on and we won't even know it. I'm not sure how I feel about that.
This does a very good job with everything from GPT afaik, better than the official one: https://openai-openai-detector--f4d78.hf.space
But as we have more non-GPT LLMs, we'll probably need more detectors.
But as we have more non-GPT LLMs, we'll probably need more detectors.
It's pretty awful actually....
I sent a one liner prompt to Chat GPT(3.5) to generate an article about some technical topic, sent it to this detector it said 99.98% real...
there is no way to detect decent LLMs at the moment
I sent a one liner prompt to Chat GPT(3.5) to generate an article about some technical topic, sent it to this detector it said 99.98% real...
there is no way to detect decent LLMs at the moment
Zerogpt works fairly well!
nope, it's crap
I got "Your Text contains mixed signals, with some parts generated by AI/GPT" for a Wikipedia article
and "Your Text is Likely Human written, may include parts generated by AI/GPT" for a GPT Generated article
there is ZERO ways to detect gpt
I got "Your Text contains mixed signals, with some parts generated by AI/GPT" for a Wikipedia article
and "Your Text is Likely Human written, may include parts generated by AI/GPT" for a GPT Generated article
there is ZERO ways to detect gpt
I wonder what % of wikipedia articles do have substantial portions of AI-generated text by now (and is it really a problem, providing the information is accurate - in fact in most cases I can't even imagine caring too much either way if something's written by a bot, but I do care if it's hallucinated BS).
it's an old article it's not ai generated
Nice concept Jacques.
I’m reminded of the (apocryphal?) story, about the Ranger at Yellowstone National Park who was asked why they didn’t just make all the bins bear-proof. His answer:
“There is a surprisingly large overlap between the smartest bears and the least smart humans.”
From everything I’ve read, outside those models that have plagiarism seeds that their own tools can detect, there is a surprisingly large overlap between the most humanlike AI and the most robotlike human writing.
I’m reminded of the (apocryphal?) story, about the Ranger at Yellowstone National Park who was asked why they didn’t just make all the bins bear-proof. His answer:
“There is a surprisingly large overlap between the smartest bears and the least smart humans.”
From everything I’ve read, outside those models that have plagiarism seeds that their own tools can detect, there is a surprisingly large overlap between the most humanlike AI and the most robotlike human writing.
There is no circumstance where I'd want text on a website to be faded out to the background color.
Fading text is a highly ill-advised channel for conveying information
Fading text is a highly ill-advised channel for conveying information
Isn't that the point? The AI generated text has no information so by hiding it into the background you aren't losing anything.
This person doesn't want the information being conveyed by AI.
This person doesn't want the information being conveyed by AI.