ChatGPT Detection App Built by Princeton Student

I'm getting an error over and over again.

MatrixR00T said:

Mh.... So the war between AI vs AI finally begins.... Make a AI that can't be detected by AI

Click to expand...

If it's really an AI the creating AI can always observe how the detecting AI detects and adjust accordingly. In Sartre's example above: Just add more perplexity (whatever that is). But for me, detecting AI makes no sense. Google should concentrate on detecting good content (for humans) and discard the crap, no matter if it's human or AI-generated.

really I dont know who cares, what is the reason to detect if content is AI I dont care at all. From SEO content perspective it is all made for Google because they force us to make tons of usless content that noone reads and needs. So whats the difference, personally I dont care at all

Well goggle have more problem than just detecting of AI content etc... I do see future where people don't even go to Google any more and just ask AI for what they want or looking for

qwerty13 said:

really I dont know who cares, what is the reason to detect if content is AI I dont care at all. From SEO content perspective it is all made for Google because they force us to make tons of usless content that noone reads and needs. So whats the difference, personally I dont care at all

Click to expand...

If a random student can make AI detection tool in a month imagine what google can do.

We keep talking about this detection tools from seo point of view, when most likely they are not built with that in mind. What if those tools are built for teachers to check their students work, website owners to check their writers works, etc.

Those tools are designed to check one piece of work one by one. Anyone ever tried to paste 100K words to these tools and see what happen?

And what about billions of world google have to check if they want to detect AI?

I always have mine getting lower scores around 10-20 (high chances of AI). I guess my GPT implementation let room for detection.

Now the quick fix days must have come to a hard fix months.

I wrote a python script to bulk-analyze the inputs scrapped from different blogs. So far I've analyzed 92 articles dated from 2018 - 2021. The false positive detection rate is quite high and it's around 62%. I would not bother too much with this app IMO.

User12121 said:

I'm almost sure that google already has some sort of tool to find which contact is AI-generated.
but it is nice to know there is such a tool for public users, It could help to test "copywriters"

Click to expand...

Exactly. If this guy is releasing it, huge chance Google already has their own. I'm thinking about testing this on a grey market site but don't want to ruin it in the future. Maybe I'll test on a new domain. I'm sure Google sees this as a huge problem and has been working on detection for a long time now.

The problem is, most generic content is on par with AI anyways. I have two writers I use. One that just rewrites stuff and one that creates much larger researched content.
I have a cheap writer I use and all she is doing is re-writing from the dozens if not hundreds of similar articles out there and AI does the same thing. Whether it's AI or cheaply reworded, I don't see it having much value. Maybe now but I doubt in the future. Though AI is a lot cheaper than me spending $5 per article.

We have to first see if the tool he made really works or not. He will also have to spend a lot of hosting resources which is hard to do as a single guy.

Straight out of GPT3 playground after tweaking the settings. No spinning/rewriting from third party apps or services.