Bitcointalk AI diarrhea auto-report | Bitcointalksearch.org

Lucius

legendary

Activity: 3234

Merit: 5637

Blackjack.fun-Free Raffle-Join&Win $50🎲

I am for everything that can help us fight more effectively against all those who think that AI is the solution to all their problems when it comes to participating in forums like this one. However, I think that automatic tagging should be avoided and that regardless of the accuracy of the results, the human factor should still be included in everything. I had one case where I accused a member of using AI, which in the end turned out to be incorrect, regardless of the fact that two AI detectors positively detected it.

Quote from: examplens on October 28, 2024, 05:39:44 AM

DT membership has been manipulated too many times, DT2 rank is not so unattainable.
Perhaps it is better to limit it to Newbie up to Full member rank or possibly only to users who have earned less than 100 merit. Somehow I got the impression that these are the users who most often decide on AI help when writing posts.

It is true that I personally detect beginners as those who use AI the most, but I think that is only because they are the most suspicious of us. So I agree that the focus should be on the ranks you mention, although I have no doubt that there are those with the highest ranks who use AI, although they are probably much more cunning at it.

ABCbits

legendary

Activity: 2870

Merit: 7490

Crypto Swap Exchange

Quote from: Joel_Jantsen on October 26, 2024, 03:49:14 PM

Quote from: aliveNFT on October 24, 2024, 03:12:04 AM

There’s also a feature that automatically assigns a neutral tag to a user, indicating they post AI-generated content, along with a reference link to the post, if the content matches by 95% or more. And report it here of course AI Spam Report Reference Thread

In general, I’m conflicted about this whole idea, so I’d appreciate your opinions. Maybe I’ll drop it and focus on something more useful.

I don't see this working out in the long run. Could you argue with a user who'd claim that their post is not AI-generated and your content detection LLM could be wrong? 95% accuracy with the warning "Don't trust our results as we could be wrong" can have long discussions about whether a post is AI-generated or written by the user. If you had to report any of my posts as AI-generated I could logically prove your LLM & text classification algorithm could be just giving false positives.

In addition, such statement already used by both user who use chatbot and user who doesn't.

Quote from: LoyceV on October 28, 2024, 05:22:53 AM

Quote from: examplens on October 25, 2024, 06:04:32 AM

Using AI to fight against AI posters. smart... Wink

Sell AI diarrhea and sell AI diarrhea detection. Create a problem, sell a solution. Smart indeed.

Your statement reminds me of OpenAI who release AI text detector, but stopped quickly due to poor performance. I wonder whether they'd make it paid feature if it have acceptable performance.

LoyceV

legendary

Activity: 3290

Merit: 16489

Thick-Skinned Gang Leader and Golden Feather 2021

Quote from: examplens on October 28, 2024, 05:39:44 AM

Perhaps it is better to limit it to Newbie up to Full member rank or possibly only to users who have earned less than 100 merit.

To get OP started:

Quote from: LoyceV on October 27, 2024, 01:45:29 PM

UserIDs, sent Merit and earned Merit machine readable
Sample

Code:

0:569:0
3:0:6922
4:0:752
10:0:1
11:0:5
12:0:1
13:3:26
14:0:11
21:0:2
23:0:1
24:0:9
26:0:18
27:0:4
28:0:13
29:0:4
30:380:722
31:0:1
33:0:27
34:0:4
35:13669:12345
37:0:6
40:0:4
42:0:69
49:0:5
97:0:2
101:0:2
143:0:2284
145:0:1
146:0:4
163:0:21
182:1:0
183:9:1
198:2:80
203:68:209
206:0:14
217:3:33
224:0:1301
237:0:5
241:0:9
244:0:1
251:0:1
262:0:1
267:0:2
269:0:1
270:0:51
271:0:1
274:0:42
284:0:6
288:0:8
292:0:5
.......
.......
.......
3655689:3:15
3656257:0:1
3656381:0:12
3656392:0:3
3656520:0:6
3656592:0:1
3656662:0:7
3656759:0:4
3656820:0:1
3656829:0:1
3656898:0:1
3656941:0:1
3657002:0:2
3657017:0:4
3657084:0:3
3657200:0:8
3657206:0:1
3657326:0:1
3657329:0:1
3657420:0:3
3657635:0:2
3657771:2:8
3657819:0:18
3657821:0:8
3658066:4:12
3658068:0:2
3658089:0:1
3658553:0:3
3658569:0:2
3658661:0:4
3658832:0:21
3658877:0:2
3659444:0:2
3659448:0:3
3659543:0:2
3659556:0:2
3659796:6:28
3659817:0:1
3659844:0:6
3659995:0:6
3660256:0:4
3660388:0:1
3660553:0:2
3660746:0:1
3660834:0:4
3661310:0:4
3661882:0:4
3662018:0:1
3662068:0:1
3662093:0:1

Full list (1 MB)

examplens

legendary

Activity: 3472

Merit: 3507

Crypto Swap Exchange

Quote from: aliveNFT on October 28, 2024, 05:27:25 AM

Will add all DT1/DT2 members to the ignore list. But I'm still thinking about how it's all beautiful and not expensive to implement, I'm looking for some triggers.

DT membership has been manipulated too many times, DT2 rank is not so unattainable.
Perhaps it is better to limit it to Newbie up to Full member rank or possibly only to users who have earned less than 100 merit. Somehow I got the impression that these are the users who most often decide on AI help when writing posts.

aliveNFT

full member

Activity: 147

Merit: 83

aliveNFT.github.io | Track your love.

Quote from: LoyceV on October 28, 2024, 05:22:53 AM

Anywhere from 1 to about 10,000 I guess. In my last 20 posts, I count a total of 9689 words including headers, quotes and lists. With 23,455 posts in 7 days, 300,000 words per months won't be enough.

I am also thinking about this problem now.
I think I'll ignore the list of users like you so that the messages are not read. (of course in code)

And in general, with the exception of bounty and other obvious things, I will add a threshold of 60-70 words on the detector.
Which immediately cuts off a very large amount of unnecessary text.

Will add all DT1/DT2 members to the ignore list. But I'm still thinking about how it's all beautiful and not expensive to implement, I'm looking for some triggers.

LoyceV

legendary

Activity: 3290

Merit: 16489

Thick-Skinned Gang Leader and Golden Feather 2021

Quote from: examplens on October 25, 2024, 06:04:32 AM

Using AI to fight against AI posters. smart... Wink

Sell AI diarrhea and sell AI diarrhea detection. Create a problem, sell a solution. Smart indeed.

Quote from: aliveNFT on October 25, 2024, 08:58:01 AM

also, does it make sense to use my main account for auto-posting and how secure will it be?

Why risk it, if you can just as easily create "aliveBOT"?

Quote from: aliveNFT on October 24, 2024, 03:12:04 AM

Also, how many words are typically posted on average in forum messages? If anyone could provide an approximate number, that would be helpful.

Anywhere from 1 to about 10,000 I guess. In my last 20 posts, I count a total of 9689 words including headers, quotes and lists. With 23,455 posts in 7 days, 300,000 words per months won't be enough.

Joel_Jantsen

legendary

Activity: 1988

Merit: 1317

Get your game girl

Quote from: aliveNFT on October 24, 2024, 03:12:04 AM

There’s also a feature that automatically assigns a neutral tag to a user, indicating they post AI-generated content, along with a reference link to the post, if the content matches by 95% or more. And report it here of course AI Spam Report Reference Thread

In general, I’m conflicted about this whole idea, so I’d appreciate your opinions. Maybe I’ll drop it and focus on something more useful.

I don't see this working out in the long run. Could you argue with a user who'd claim that their post is not AI-generated and your content detection LLM could be wrong? 95% accuracy with the warning "Don't trust our results as we could be wrong" can have long discussions about whether a post is AI-generated or written by the user. If you had to report any of my posts as AI-generated I could logically prove your LLM & text classification algorithm could be just giving false positives.

Also, from their official docs

Quote

GPTZero’s plagiarism detection tool compares your inputted text to billions of webpages and academic papers found online. When you use GPTZero’s plagiarism check to detect plagiarism, it will give you a report of what % of your content is plagiarized and flags specific sentences and which sources they are from. Don’t worry - all text inputted to GPTZero is confidential and no one will have access to your scans.

examplens

legendary

Activity: 3472

Merit: 3507

Crypto Swap Exchange

Quote from: aliveNFT on October 25, 2024, 08:58:01 AM

also, does it make sense to use my main account for auto-posting and how secure will it be? Or it's easier to create a new one, I still don't fully understand if someone has any advice.

Use a bot account for this. Something like what bitmover created for posting fee rate FeeBuddy

ABCbits

legendary

Activity: 2870

Merit: 7490

Crypto Swap Exchange

Quote from: aliveNFT on October 25, 2024, 08:58:01 AM

also, does it make sense to use my main account for auto-posting and how secure will it be? Or it's easier to create a new one, I still don't fully understand if someone has any advice.

You don't want to see your main account hacked, since you probably need to store cookies or username/password on online server.

Quote from: logfiles on October 25, 2024, 06:00:34 PM

Quote from: aliveNFT on October 25, 2024, 08:58:01 AM

also, does it make sense to use my main account for auto-posting and how secure will it be? Or it's easier to create a new one, I still don't fully understand if someone has any advice.

I think it's better you use a separate account. I have seen some old members do the same in the past for their bots. As for the words. 300,000 per month is little. With people looking to make at least 150 words per post. That will be used up in a couple of days Grin

In addition, https://ninjastic.space/ currently shows there are 3704 posts in last 24 hours and 101991 posts in August 2024. Although looking at @Mitchell post, OP could create manual filter to exclude bounty posts before feeding posts to the AI detector.

Vod

legendary

Activity: 3668

Merit: 3010

Licking my boob since 1970

Quote from: aliveNFT on October 25, 2024, 08:58:01 AM

also, does it make sense to use my main account for auto-posting and how secure will it be? Or it's easier to create a new one, I still don't fully understand if someone has any advice.

I created vodbot to parse so it wouldn't affect my stats, but I never posted with it.

logfiles

copper member

Activity: 2170

Merit: 1822

Top Crypto Casino

Quote from: aliveNFT on October 25, 2024, 08:58:01 AM

also, does it make sense to use my main account for auto-posting and how secure will it be? Or it's easier to create a new one, I still don't fully understand if someone has any advice.

I think it's better you use a separate account. I have seen some old members do the same in the past for their bots. As for the words. 300,000 per month is little. With people looking to make at least 150 words per post. That will be used up in a couple of days Grin

Findingnemo

hero member

Activity: 2366

Merit: 793

Bitcoin = Financial freedom

Quote from: aliveNFT on October 25, 2024, 08:58:01 AM

also, does it make sense to use my main account for auto-posting and how secure will it be? Or it's easier to create a new one, I still don't fully understand if someone has any advice.

I don't have much expertise about the data management from these tools that are used for auto posting but I guess Creating a separate account that will be used for only the automated posting is better so you do not need to expose your main account credentials to the third-party tool.

aliveNFT

full member

Activity: 147

Merit: 83

aliveNFT.github.io | Track your love.

At the beginning of the next week, then I will have some free time, I will run a test version, see how everything will work and compare which API works more accurately.
Open-source of course
I will not add a neutral feedback feature yet

also, does it make sense to use my main account for auto-posting and how secure will it be? Or it's easier to create a new one, I still don't fully understand if someone has any advice.

examplens

legendary

Activity: 3472

Merit: 3507

Crypto Swap Exchange

Using AI to fight against AI posters. smart... Wink

Quote from: Mitchell on October 24, 2024, 05:39:12 AM

Quote from: aliveNFT on October 24, 2024, 03:43:42 AM

And everything else like that throughout the forum, including autoposting.
Contests, Signatures, Bounties and others...

Bounty posts can already be filtered: https://bitcointalk.org/index.php?action=recent;patrol;nobounty

It may be easier to add only the sections that are most critical regarding the use of AI. For example, this thing would be most useful in the technical part of the forum, and there it is quite tiring to meet with AI posts of users who try to give themselves importance in this way.

What if the results of detecting AI posts were delivered through a browser addon (like BPIP), where there would be a notification next to the post, for example, "AI detected" or something like that?
Maybe you can talk about it with PowerGlove, has already done several patches for the forum

btw. I don't think leaving feedback automatically is a good idea.

Mitchell

copper member

Activity: 3948

Merit: 2201

Verified awesomeness ✔

Quote from: aliveNFT on October 24, 2024, 03:43:42 AM

I just thought that I could also add a parameter so that it doesn't check posts like

Quote from: username

Username: username
BTC SegWit Address: addy

And everything else like that throughout the forum, including autoposting.
Contests, Signatures, Bounties and others...

Bounty posts can already be filtered: https://bitcointalk.org/index.php?action=recent;patrol;nobounty

ABCbits

legendary

Activity: 2870

Merit: 7490

Crypto Swap Exchange

Quote from: aliveNFT on October 24, 2024, 03:12:04 AM

For example, GPTZero offers up to 300,000 words per month for $24, which isn’t much, but is it worth it? Overall, it works quite accurately.

I think you need to subscribe or pay multiple AI-detection service. Otherwise, you may report post that actually written by human which makes thread "AI Spam Report Reference Thread" clogged. In addition, some AI/chatbot generated post usually is very long which could drain your wallet quickly.

aliveNFT

full member

Activity: 147

Merit: 83

aliveNFT.github.io | Track your love.

I just thought that I could also add a parameter so that it doesn't check posts like

Quote from: username

Username: username
BTC SegWit Address: addy

And everything else like that throughout the forum, including autoposting.
Contests, Signatures, Bounties and others...

aliveNFT

full member

Activity: 147

Merit: 83

aliveNFT.github.io | Track your love.

Hello everyone, recently I spent some time creating a simple website using MongoDB, Flask, and HTML, with a script that pulls the latest posts from =recent and checks them for AI-generated content. This happens automatically via an API or script (unfortunately, there are no good and accurate opensource solutions at the moment, and there likely won’t be any).

Naturally, it only checks original messages without the _[/quote] tag, and I have a question: is it worth the effort and money?
Also, how many words are typically posted on average in forum messages? If anyone could provide an approximate number, that would be helpful.

For example, GPTZero offers up to 300,000 words per month for $24, which isn’t much, but is it worth it? Overall, it works quite accurately.

There’s also a feature that automatically assigns a neutral tag to a user, indicating they post AI-generated content, along with a reference link to the post, if the content matches by 95% or more. And report it here of course AI Spam Report Reference Thread

In general, I’m conflicted about this whole idea, so I’d appreciate your opinions. Maybe I’ll drop it and focus on something more useful.

Topic: Bitcointalk AI diarrhea auto-report (Read 239 times)