The J.A.R.V.I.S AutoReply Protocol Initiative (JARPI) - Powered by ChatGPT

Igebotz

staff

Activity: 1316

Merit: 1610

The Naija & BSFL Sherrif 📛

Quote from: joker_josue on April 25, 2023, 09:09:35 AM

Quote from: NotATether on April 25, 2023, 08:00:55 AM

That's basically correct, but see my reply to borovichok. I'm eventually going to have this run through all newly created topics and run the detection automatically (using an API to some websites or Selenium).

Sounds good to me.

Just suggested that in this case, the autotext be a little different, just saying it's suspicious content and not making a direct recommendation for it to be reported. That is, to be a more alert text for readers to be aware of, and not a condemnatory text.

Is it possible to configure your proposed JARPI to behave similarly to the auto ChatGT bots on Twitter? It triggers whenever someone tags it in a post and provides extremely good analyses. If I see a suspected AI post that I'm not sure about, I'll just @JARPI on the thread to determine if it's AI generated or not. I think it'd be better this way than you doing all of the work. Create JARPI and make it available to anyone who needs it.

joker_josue

legendary

Activity: 1862

Merit: 5154

**In BTC since 2013**

Quote from: NotATether on April 25, 2023, 08:00:55 AM

That's basically correct, but see my reply to borovichok. I'm eventually going to have this run through all newly created topics and run the detection automatically (using an API to some websites or Selenium).

Sounds good to me.

Just suggested that in this case, the autotext be a little different, just saying it's suspicious content and not making a direct recommendation for it to be reported. That is, to be a more alert text for readers to be aware of, and not a condemnatory text.

NotATether

legendary

Activity: 1568

Merit: 6660

bitcoincleanup.com / bitmixlist.org

Quote from: borovichok on April 25, 2023, 06:14:24 AM

Are you going to manually look for suspected AI posts, do manual scanning, and check whether or not it's AI created before triggering JARPI to act? I thought JARPI was meant to be a bot conducting this errand for you rather than you running the errand? What's the point if there's still a bunch of shitwork for you to do?

This is a new idea, so the automation has not been coded in yet. I would definitely like to automate the AI detection part sooner or later.

Quote from: joker_josue on April 25, 2023, 06:34:49 AM

Quote from: NotATether on April 25, 2023, 04:02:48 AM

Third option is correct. I ran OP's text through many AI detectors and all of them gave AI-generated indicators, so I fired up JARPI to send the warning.

So basically you browse the forum normally, and when you think an OP was made by AI, you submit the topic for JARPI analysis.
Then it will analyze and if it concludes that the text was generated by AI, it will post the ad. This is it?

Thus, you stop worrying about analyzing these cases in detail, and you can continue to browse the forum normally looking for other cases. Seems to be an interesting tool, to free your time from those boring analyses.

Honestly, I find this option much more useful and interesting than answering repetitive topics.

That's basically correct, but see my reply to borovichok. I'm eventually going to have this run through all newly created topics and run the detection automatically (using an API to some websites or Selenium).

Quote from: Josefjix on April 25, 2023, 05:39:35 AM

This is somewhat preferable to the previous plan, but my concern is what would happen if everyone disregards the JAPI response and proceeds with their discussion regardless of whether it was generated by AI or not, similar to this post https://bitcointalksearch.org/topic/essentials-to-starting-a-business-5448187. Since there are currently no explicit forum rules against AI and no 99% proved AI detector tools, nobody appears to care and the moderator won't lock a thread just because JAPI says so. there is a light at the end of the tunnel but the challenges looks bigger.

I guess I'll have the bot file a report every 24 hoursw for a period of 2 weeks, or until the thread is locked. Whichever comes first.

joker_josue

legendary

Activity: 1862

Merit: 5154

**In BTC since 2013**

Quote from: NotATether on April 25, 2023, 04:02:48 AM

Third option is correct. I ran OP's text through many AI detectors and all of them gave AI-generated indicators, so I fired up JARPI to send the warning.

So basically you browse the forum normally, and when you think an OP was made by AI, you submit the topic for JARPI analysis.
Then it will analyze and if it concludes that the text was generated by AI, it will post the ad. This is it?

Thus, you stop worrying about analyzing these cases in detail, and you can continue to browse the forum normally looking for other cases. Seems to be an interesting tool, to free your time from those boring analyses.

Honestly, I find this option much more useful and interesting than answering repetitive topics.

borovichok

hero member

Activity: 1148

Merit: 518

Quote from: NotATether on April 25, 2023, 04:02:48 AM

Quote from: joker_josue on April 25, 2023, 03:47:52 AM

Quote from: NotATether on April 25, 2023, 03:38:13 AM

I guess now I'm going to be using JARPI as an AI-buster:

Look, I think this idea is better than answering questions!

But, based on the automatic text, I had a doubt:
Do you first manually validate the AI's suspicion and then post, or does it post automatically?
Or are you the one who suspects a topic, and puts the JARPI to analyze which then makes the post in case of a positive result?

Third option is correct. I ran OP's text through many AI detectors and all of them gave AI-generated indicators, so I fired up JARPI to send the warning.

Are you going to manually look for suspected AI posts, do manual scanning, and check whether or not it's AI created before triggering JARPI to act? I thought JARPI was meant to be a bot conducting this errand for you rather than you running the errand? What's the point if there's still a bunch of shitwork for you to do?

Josefjix

sr. member

Activity: 1288

Merit: 305

yes

Quote from: Learn Bitcoin on April 24, 2023, 02:27:19 PM

Quote from: Josefjix on April 24, 2023, 07:30:25 AM

Aren't you the same person whitelisting them back to the forum now? how many of those users became a big part of the forum^{well I don't expect you to answer}

Sorry Op for going off-topic

I don't know why but I feel like you are trying to trigger him (good try). None of them are paid jobs. He is doing it voluntarily and he is not responsible for what they do after. Most of His whitelisted users doing good. He often posts the Ban %. If you are trying to trigger him, I am sorry but he is unlikely to pay attention to you as he proved himself as a thick-skinned gang leader.

We were just having a chat and he felt it wasn't the proper place because it was a little off-topic from the OP, I simply made my perspective plain to him, it's nothing personal, and BTW, we're still on the same track. Leave it to the moderator to decide and quit pressing.

Quote from: NotATether on April 25, 2023, 04:02:48 AM

Quote from: nutildah on April 25, 2023, 03:50:06 AM

I have to ask out of curiosity, and due to my own thread on the issue, which AI detectors are you going to use, and how are you weighing the results? Are there different levels of likeliness, or will it only post replies in threads that come back as "highly likely" to be AI-generated?

I use at least three different AI checking websites. Or more than three if even one of them doubts that the text is not completely AI-generated. I do not use the GPTZero website - I had too many false positives with it.

This is somewhat preferable to the previous plan, but my concern is what would happen if everyone disregards the JAPI response and proceeds with their discussion regardless of whether it was generated by AI or not, similar to this post https://bitcointalksearch.org/topic/essentials-to-starting-a-business-5448187. Since there are currently no explicit forum rules against AI and no 99% proved AI detector tools, nobody appears to care and the moderator won't lock a thread just because JAPI says so. there is a light at the end of the tunnel but the challenges looks bigger.

NotATether

legendary

Activity: 1568

Merit: 6660

bitcoincleanup.com / bitmixlist.org

Quote from: cafter on April 25, 2023, 05:06:15 AM

@notatether

don't you think, when many people will report that thread or post by watching JARVIS's reply, then maybe the moderator will be frustrated by too many reports. (i don't know how it seems from mod side, but maybe it will become confusing when too many people will report.)

How many of them will actually report though, out of the ones that are reading it

Unfortunately, even after making the announcement, and showing the AI detection results in big fat pictures, there are still idiots who are mindlessly replying to that thread.

Moderators probably have a huge backlog of reports to deal with anyway, and the worst that will happen is that the report is simply unhandled.

cafter

full member

Activity: 448

Merit: 225

@notatether

don't you think, when many people will report that thread or post by watching JARVIS's reply, then maybe the moderator will be frustrated by too many reports. (i don't know how it seems from mod side, but maybe it will become confusing when too many people will report.)

NotATether

legendary

Activity: 1568

Merit: 6660

bitcoincleanup.com / bitmixlist.org

Quote from: joker_josue on April 25, 2023, 03:47:52 AM

Quote from: NotATether on April 25, 2023, 03:38:13 AM

I guess now I'm going to be using JARPI as an AI-buster:

Look, I think this idea is better than answering questions!

But, based on the automatic text, I had a doubt:
Do you first manually validate the AI's suspicion and then post, or does it post automatically?
Or are you the one who suspects a topic, and puts the JARPI to analyze which then makes the post in case of a positive result?

Third option is correct. I ran OP's text through many AI detectors and all of them gave AI-generated indicators, so I fired up JARPI to send the warning.

Quote from: nutildah on April 25, 2023, 03:50:06 AM

I have to ask out of curiosity, and due to my own thread on the issue, which AI detectors are you going to use, and how are you weighing the results? Are there different levels of likeliness, or will it only post replies in threads that come back as "highly likely" to be AI-generated?

I use at least three different AI checking websites. Or more than three if even one of them doubts that the text is not completely AI-generated. I do not use the GPTZero website - I had too many false positives with it.

nutildah

legendary

Activity: 3010

Merit: 8114

Quote from: NotATether on April 25, 2023, 03:38:13 AM

I guess now I'm going to be using JARPI as an AI-buster:

Quote from: ?? on ??

AI-generated topic
It is highly likely that the OP of this topic has created it entirely using AI such as ChatGPT. This has been verified by @NotATether who has ran the content through multiple AI detector websites. Therefore, it is recommended that this topic is locked, as no further constructive discussion can be added to this topic.

Users who see this message should report this topic to a moderator, requesting it to be locked, instead of facilitating the spread of spam.

This reply was auto-generated by JARPI, an AI protocol created by NotATether for automatically replying to frequently asked questions, and other common types of posts.

Alright, now I think you're on to something.

I have to ask out of curiosity, and due to my own thread on the issue, which AI detectors are you going to use, and how are you weighing the results? Are there different levels of likeliness, or will it only post replies in threads that come back as "highly likely" to be AI-generated?

joker_josue

legendary

Activity: 1862

Merit: 5154

**In BTC since 2013**

Quote from: NotATether on April 25, 2023, 03:38:13 AM

I guess now I'm going to be using JARPI as an AI-buster:

Look, I think this idea is better than answering questions!

But, based on the automatic text, I had a doubt:
Do you first manually validate the AI's suspicion and then post, or does it post automatically?
Or are you the one who suspects a topic, and puts the JARPI to analyze which then makes the post in case of a positive result?

NotATether

legendary

Activity: 1568

Merit: 6660

bitcoincleanup.com / bitmixlist.org

I guess now I'm going to be using JARPI as an AI-buster:

Quote from: ?? on ??

AI-generated topic
It is highly likely that the OP of this topic has created it entirely using AI such as ChatGPT. This has been verified by @NotATether who has ran the content through multiple AI detector websites. Therefore, it is recommended that this topic is locked, as no further constructive discussion can be added to this topic.

Users who see this message should report this topic to a moderator, requesting it to be locked, instead of facilitating the spread of spam.

This reply was auto-generated by JARPI, an AI protocol created by NotATether for automatically replying to frequently asked questions, and other common types of posts.

Learn Bitcoin

hero member

Activity: 462

Merit: 767

Instant cryptocurrency exchange with own reserves!

Quote from: Josefjix on April 24, 2023, 07:30:25 AM

Aren't you the same person whitelisting them back to the forum now? how many of those users became a big part of the forum^{well I don't expect you to answer}

Sorry Op for going off-topic

I don't know why but I feel like you are trying to trigger him (good try). None of them are paid jobs. He is doing it voluntarily and he is not responsible for what they do after. Most of His whitelisted users doing good. He often posts the Ban %. If you are trying to trigger him, I am sorry but he is unlikely to pay attention to you as he proved himself as a thick-skinned gang leader.

About the six posts you mentioned, look at the percentage, please. If a user made ten thousand posts and has 100 deleted posts, that's nothing. It's only 1% of his posts. At the same time, if anyone writes 60 posts and 6 of them got deleted. That's ten percent of his posts. No one forcing moderators to ban them. Users reported posts and the rest of them were on Moderators. They will decide what to do.

What's the problem here?

NotATether

legendary

Activity: 1568

Merit: 6660

bitcoincleanup.com / bitmixlist.org

Quote from: Josefjix on April 24, 2023, 07:30:25 AM

Quote from: NotATether on April 24, 2023, 06:09:17 AM

Quote from: Josefjix on April 24, 2023, 04:50:08 AM

For the JARPI bot, no matter how hard you spin those texts manually, they won't be original, and have you considered the amount of posts this bot will be posting per day since it triggers every time someone asks a question, or are there limitations I'm not aware of, this AI thing is getting out of hand, these tools would render some of our education sector useless, resulting in the loss of jobs if nothing is done now.

I haven't ran JARPI ever since the manual experiment earlier last week.

Okay, what were the response speed and success rate? I appreciate the time and effort you have put into this, but I'm concerned that this bot will hijack every thread and post thousands of times per day, this is one part you are yet to address.

Well firstly, report handling times are exactly the same as manual reporting, so that's one.

I don't intend to unleash the bot on autopilot and find threads to post in on its own. Generative AI is not sophisticated enough for that yet. I would need to train a homegrown neural network on all recent Bitcointalk posts, but that is very difficult because I don't have all the posts locally (yet).

Also the bot is bottlenecked by how fast I can solve Cloudflare captchas. Maybe 10 at once before I rip my brains out.

I'm not quite sure of the resource requirements such a neural network would require, but if it's large or even possibly average, I believe it's not worth it considering the negative perception of this protocol.

cafter

full member

Activity: 448

Merit: 225

I think all these is a waste of time and gain of experience for @notatether
because when a low quality thread is posted, after some minutes some users might already replied with very good answers, and your bot will post another reply which maybe a little spammy and off topic(when OP is weak in english maybe AI misunderstand it)) and answers to that thread is already humanly answered by some user with very good explanation, it may be a waste of time for you but definitely you will learn something new in this project.

My idea:
many low quality posts are posted by newbie, so we can try to tell them(by making search button bold, writing about using search button in top of the forum homepage, etc) to use the search tool for their answer. I many times used search bar and got my answer quickly without wasting others and mine time to waiting for replies.

(this was pretty good discussion to read, many other replies are all same in most of the threads)

Josefjix

sr. member

Activity: 1288

Merit: 305

yes

Quote from: NotATether on April 24, 2023, 06:09:17 AM

Quote from: Josefjix on April 24, 2023, 04:50:08 AM

For the JARPI bot, no matter how hard you spin those texts manually, they won't be original, and have you considered the amount of posts this bot will be posting per day since it triggers every time someone asks a question, or are there limitations I'm not aware of, this AI thing is getting out of hand, these tools would render some of our education sector useless, resulting in the loss of jobs if nothing is done now.

I haven't ran JARPI ever since the manual experiment earlier last week.

Okay, what were the response speed and success rate? I appreciate the time and effort you have put into this, but I'm concerned that this bot will hijack every thread and post thousands of times per day, this is one part you are yet to address.

Quote from: LoyceV on April 24, 2023, 06:57:58 AM

Quote from: Josefjix on April 24, 2023, 06:04:31 AM

I can't check every individual on the list, but I'm sure most became active forum users

Cute. But wrong. I checked the first 10, and 9 of them have 0 posts (AKA Nuked). One of them has 580 Activity now, and 0 Merit because bounty spam isn't worth Merit.

How about you check all this links before ticking the wrong box for me? I bet many survived and are doing pretty good now.

Quote from: LoyceV on April 24, 2023, 06:57:58 AM

Quote

and those who were nuked moved on to open new accounts. Nothing was solved

Hundreds of farmed account were Nuked and their IP-addresses got evil point for it. Of course there will always be spammers, but every account that gets banned is one spammer less. The fact that there's much less spam now than we had 4 years ago confirms it's working.

I'll end this discussion here. If you want to continue this, feel free to open another topic for it.

Aren't you the same person whitelisting them back to the forum now? how many of those users actually became big part of the forum^{well i don't expect you to answer}

Sorry Op for going off-topic

LoyceV

legendary

Activity: 3290

Merit: 16489

Thick-Skinned Gang Leader and Golden Feather 2021

Quote from: Josefjix on April 24, 2023, 06:04:31 AM

I can't check every individual on the list, but I'm sure most became active forum users

Cute. But wrong. I checked the first 10, and 9 of them have 0 posts (AKA Nuked). One of them has 580 Activity now, and 0 Merit because bounty spam isn't worth Merit.

Quote

and those who were nuked moved on to open new accounts. Nothing was solved

Hundreds of farmed account were Nuked and their IP-addresses got evil point for it. Of course there will always be spammers, but every account that gets banned is one spammer less. The fact that there's much less spam now than we had 4 years ago confirms it's working.

I'll end this discussion here. If you want to continue this, feel free to open another topic for it.

NotATether

legendary

Activity: 1568

Merit: 6660

bitcoincleanup.com / bitmixlist.org

Quote from: Josefjix on April 24, 2023, 04:50:08 AM

For the JARPI bot, no matter how hard you spin those texts manually, they won't be original, and have you considered the amount of posts this bot will be posting per day since it triggers every time someone asks a question, or are there limitations I'm not aware of, this AI thing is getting out of hand, these tools would render some of our education sector useless, resulting in the loss of jobs if nothing is done now.

I haven't ran JARPI ever since the manual experiment earlier last week.

Josefjix

sr. member

Activity: 1288

Merit: 305

yes

Quote from: LoyceV on April 24, 2023, 04:55:39 AM

Quote from: Josefjix on April 24, 2023, 04:50:08 AM

You don't expect the admin to ban someone for 6 nasty posts, do you?

The forum would be better without such users.

That's why we have mods; report them and let them do their thing; they follow the rules; 6 ugly shit shouldn't get anyone banned.

Quote

Remember when you asked the Admin to ban DdmrDdmr for spamming when he was a newbie?

That's a different case: DdmrDdmr's username came up in an automated search, which wasn't meant to be exhaustive. Luckily, the Mods did their checks. Around the same time hundreds of users got banned based on the list I provided.

It's the same thing; you pushed hard to have everyone on that list nuked. Fortunately, the moderators did their jobs; I can't check every individual on the list, but I'm sure most became active forum users, and those who were nuked moved on to open new accounts. Nothing was solved

LoyceV

legendary

Activity: 3290

Merit: 16489

Thick-Skinned Gang Leader and Golden Feather 2021

Quote from: Josefjix on April 24, 2023, 04:50:08 AM

You don't expect the admin to ban someone for 6 nasty posts, do you?

The forum would be better without such users.

Quote

Remember when you asked the Admin to ban DdmrDdmr for spamming when he was a newbie?

That's a different case: DdmrDdmr's username came up in an automated search, which wasn't meant to be exhaustive. Luckily, the Mods did their checks. Around the same time hundreds of users got banned based on the list I provided.

Topic: The J.A.R.V.I.S AutoReply Protocol Initiative (JARPI) - Powered by ChatGPT (Read 918 times)