Post format example:


Username: cryptodude

Post #1:


Hivemoderation: 99.9% likely to contain AI
Writer: 28% human


Post #2:


Hivemoderation: 99.8% likely to contain AI
Contentatscale: 28% human


Post #3:


Hivemoderation: 99.9% likely to contain AI
Contentatscale: 25% human

@nutildah
Is there a particular reason you suggest using those AI detectors and not some others? Are they more likely to detect automated content compared with some competitors.
I wanted to suggest two more AI classifiers that can get the job done and don't require you to register an account, pay for the service, etc. If you have any comments on their quality, don't hesitate to speak up.

https://gptzero.me/
https://platform.openai.com/ai-text-classifier

Here are the report results for the above examples, each post was reported as "AI spam":



Would they have been deleted without the additional link reference appended to each report? Hard to say. Just to make it clear what I'm putting in each report, it looks like this:



with the link being to the post that contains the results of the AI analysis.

---

I chose them for a few reasons:

- in my testing they seem to provide the most reliable results, although they're still not perfect
- they give a % as a result, which is more granular than something like "Your text is likely to be written entirely by AI"
- they are free of charge and don't require registration


gptzero.me seems to do a fairly good job but the problem is the results are hard to compare to analyzers that return percentages; same thing with platform.openai.com which additionally requires creation of an account

---

Also, for posts like these:


I don't think its necessary to run them through the analyzers -- hopefully the mod will see this giveaway text written in the post. I will, however, be using this post as a reference for a neutral feedback... not sure if I'm gonna do that for every account mentioned here but I might.

I'm thinking of lowering the standards for what can be included as AI-assisted because both writer.com and writefull.com are extremely lenient compared to hivemoderation.com.

I want to mention this poster because I am 100% certain that not only is it a bought account but that they are using AI to write posts -- but they are a bit more clever than the other examples because they remove punctuation from their posts, which tricks the detectors. It ends up looking like spam gibberish and nobody will ever read them but they get paid for them anyway.

---

First of all, nlovric has a registration date of February 03, 2012, yet they wrote this:


 

---

Post #1: Here's an example of a post with the punctuation removed:


It comes back as 32.4% likely to contain AI Generated Text, according to hivemoderation.com.

Here it is with proper punctuation:


This version comes back as 99.1% likely to contain AI Generated Text. However, all the other AI text analyzers suggest it is human-written, with the exception of GPTZero which says it has a "moderate likelihood of being written by AI."

---

Post #2: Here's another example where both commas and apostrophes ('s) were removed:


By itself, hivemoderation.com says it is 99.1% likely to contain AI Generated Text.

---

Post #3: They are also using ChatGPT to re-write quoted posts, which is maddeningly stupid:


Clearly AI didn't know that Strikers was in reference to a football position so it re-wrote the text as "department that is on strike," lol.

---

Post #4: Here's one more example where commas and apostrophes are removed, and he tacked on his own intro sentence at the beginning (hivemoderation.com says the rest of it is 99.9% likely to contain AI written text):


In the original version, GPTZero says its "likely to be written by a human," but after adding in the proper punctuation it goes up to a "moderate likelihood of being written by AI."

OK I'll give it a look next time. I'm thinking that the initial standards I propose need to change b/c they are too rigid... Perhaps it will be something like "any 2 AI detectors that report post text with a strong likelihood of being AI generated" in the future. Ultimately it depends on what the moderators think suffices, and speaking of that, here is the update on the 4 nlovric posts from above that were reported as "AI spam" with a reference link to this thread:

It would appear he's been temp banned as was suggested by lovesmayfamilis... he doesn't seem like the type to just stop posting during a campaign.

I don't see where you are getting these percentages from for Copyleaks. I even signed up to use the Enhanced version and it just says "AI Content Detected".

But I do see what you're doing by combining the two posts into one for hivemoderation. Its better to just copy and paste the same short post twice (or three times) to meet the 750 char minimum. Because if I combine one human post and one AI post it may report that both or neither of the posts are AI (less accuracy).

Yes, I also use copyleaks for AI text detection. In addition to everything, this application has a convenient extension that allows you to check texts on the fly. I would like to pay attention to the posts of this person, they are very similar to the use of an AI tool, but apparently he edits the posts a little, and therefore copyleaks does not always manage to guess up to 99% accuracy.

https://bitcointalk.org/index.php?action=profile;u=1128239

Yes, it should. Technically all ChatGPT posts are plagiarism because they are being copy/pasted from another source uncredited, but those ones are especially... Its the equivalent of using a fancy text-spinner, which is also considered plagiarism.

---

I'm starting to like this Copyleaks utility.

Found another one who quoted a nlovric post.

Post #1:


hivemoderation: 96.1% likely to contain AI Generated Text
copyleaks: 75.6% probability for AI

Post #2:


hivemoderation: 93.7% likely to contain AI Generated Text
copyleaks: 74.6% probability for AI

Post #3:


hivemoderation: 99.9% likely to contain AI Generated Text
copyleaks: 75.2% probability for AI


I think this user is already temp banned. If reports for these posts are all marked as "Good" I will update the OP with the new criteria.

It seems that not entirely, because there has been another forum member who wanted to take merit for the success. I have linked to this thread before without thinking that nutildah could have discovered the cheater before. I don't know what his motives may have been but I'll keep an eye on him.

I have updated the OP with 2 additional AI detector utilities that return results as %s.

---

Yeah I dunno, could be, people are weird as hell when it comes to merits sometimes. He could even have been trying to earn back his own pay  but he did give me credit so who knows.


OK well let's start with the last one on this list.


Username: petulino

Post #1 (quoting an nlovric post):


This post does the same trick used by nlovric, removing punctuation. Even as is this post comes back 99.9% likely to contain AI Generated Text according to hivemoderation but 98.3% human for copyleaks. Let's put in the punctuation and remove the last sentence:


hivemoderation: 99.9% likely to contain AI Generated Text
copyleaks: 77.6% probability for AI


Post #2:


hivemoderation: 99.9% likely to contain AI Generated Text
copyleaks: 60.1% probability for AI

Let's see what happens when we add the punctuation:


hivemoderation: 99.9% likely to contain AI Generated Text
copyleaks: 75% probability for AI


Post #3:


hivemoderation: 99.8% likely to contain AI Generated Text
copyleaks: 99.9% human

After adding in the appropriate punctuation:


hivemoderation: 99.9% likely to contain AI Generated Text
copyleaks: 82.3% probability for AI


All these posts appear to be a text-spin of the post they are quoting as well, which means this is almost certainly an alt account of nlovric.