Topic: The J.A.R.V.I.S AutoReply Protocol Initiative (JARPI) - Powered by ChatGPT - page 5. (Read 918 times)

My bot cannot send PMs and I have no plans to add that feature. It is strictly limited to making replies (It cannot even start its own topics).


The target audience that is supposed to be reading these kind of posts are not so much Legendaries like you and I as it is for new members who've just arrived here <3 months ago.

That would be against one of these forum rules:
Depending on how you're planning to do it, those rules cover it. It's better to report those posts.

I've whitelisted it, because you (NotATether) don't need those limitations. But that doesn't mean I like it

J.A.R.V.I.S is on my ignore list already. I can't get myself to read this generic AI crap. It uses far too many words to tell nothing I didn't know yet.

It isn't (yet):




Maybe. I'll think about it. The reason I put it here is because the bot is primarily focused on the forum itself.


Yeah, imagine instead of having to fill in a Report to moderator form a thousand times with very similar reasons, you could just collect a list of post IDs and set it on autopilot while you sip on your morning coffee like, I don't know, @Ratimov or another large reporter.

Because of such things, we have a "report to moderator" button, and then such topics will end up in the trash and will not become havens for signature spammers. However, I have nothing against you playing with AI in that way, because it seems that there is no area in which AI has not already intervened.

Isn't it forbidden to post AI-Generated content? I thought it was not allowed. But as long as it's already mentioned in each post that the Post is AI generated with a source, It should be fine. But, You should talk with one of the Admins/Global Mods before you start. I read the purposes, and it seems reasonable to me. It will reply to low-quality topics only and report them if it's spam or a scam attempts. That's great.

Now I am afraid of AI's. It's going to takeover us. Once Elon Musk said AI is far more Dangerous than you think. No one hears me. People will regret doing this.


I understand his comment. He asking why this topic is in Meta? Shouldn't it fit better in Project development? IDK. Maybe Project development is for Bitcoin related Projects only.

I don't understand your comment?

---

Anyway, time for a little experiment, related to one of the non-AI use cases of this robot.

In Development and Technical discussion right now, we have two spam topics, which I will call topic number 1 and 2:

Topic 1: https://bitcointalksearch.org/topic/--5449252
Topic 2: https://bitcointalksearch.org/topic/--5449249

In Topic 1, I have reported the thread for spam advertising. and in topic 2, I have used JARPI to request users to do the same. Both actions were taken around 11:30 UTC time.

The goal of this experiment is to see which spam thread is deleted first. If topic 2 is deleted first, then that means this use case is effective, otherwise, that means it is not effective to use auto-generated replies to ask people to report posts, and I will halt that particular usecase.

So now it is time to wait. No more using the bot during this time period.  

EDIT: They were deleted in about 70 minutes at the same time. So I don't see any particularly amazing use case for spamvertising alerts. However, malware and phishing link alerts are still important. Let's try that next.

It's not a forum related topic.

Pure AI generated posts are likely to be lousy and factually incorrect. However, there should not be any problem with AI posts that are manually fact-checked by humans:


(emphasis mine).

Imagine if somebody asks a question such as the following:


Normally such questions are havens for signature spammers who otherwise do not have anything useful to contribute to the discussion. However, AI (with a few redactions or additions in the content where necessary) can make a much superior answer:



The grey color combined with the italics enables it to stand out from the other shitposts. Although maybe, I will make it a darker shade of grey. It is also possible to fine-tune the answers from ChatGPT to precisely answer specific questions such as: "Is XYZ a scam ", "How to safely use metamask", "should I use a second phone number when buying bitcoin with an exchange", "How to avoid being a victim of SIM swap attacks" and so forth.

I should note that JARPI is private software and as such, only I have the ability to make the posts.

---

P.S. I haven't asked anyone about this yet, because I don't think anyone sees the potential for this yet (and partially because this is not bitcoin-dev mailing list where most proposals get instantly shut down or ghosted  ). Instead of wading through endless pages of posts looking for the answer to a question, you can immediately stop looking when you find a JARPI-created answer, and maybe even lock the thread at that point. In other words, JARPI reduces spam by preventing other people from paying attention to subsequent low-quality answers.

Would we really appreciate if someone was using AI to respond on the forum? In my opinion, no. JARPI is going to do the exact thing some other people used here. The difference is JARPI is doing it as an experiment.
It's a place for discussion, interact one with another. I don't think it's going to serve the purpose of a forum. Also, have you asked admin whether you can use it or not? I haven't seen any exact rule on AI written post but I think it will be forbidden when the unofficial rules of official bitcointalk forum will be re-written.

JARPI (J.A.R.V.I.S AutoReply Protocol Initiative) is the name of a system I'm designing to automatically reply to low-quality, repetitive Bitcoin questions that the community is tired of regurgitating. This project is meant to demonstrate the positive impact that AI can have on Bitcointalk when it is used correctly.

Posts are directly pasted into OpenAI's ChatGPT (manually), and the answer is screened manually, and then automatically copied and used to reply to a bitcointalk thread.

It is powered by my shiny new toy, J.A.R.V.I.S. I found a way to defeat Cloudflare's captchas using Selenium and solving them manually, so this opens the door to a lot of interesting possibilities.

Jarvis text will always be italicized (sorry, not underlined - that was a mistake) and in grey, and have the following disclaimer at the bottom:


Some other useful things that JARPI can do (theoretically), including things that don't even utilize AI:

- It can make a reply on newly created spamvertising topics alerting users that the content is spam and telling them to use the "Report to moderator" feature to remove the entire topic (because lets face it, we humans are being collectively outgunned in this area).
- Similarly, for malware posts.
- Ditto, for obvious scam attempts such as Bitcoin doubler, bitcoin mining exploit, and so on.
- It can tell a user when a thread belong in the wrong board and prompt them to move it to the correct board, with a short instruction.
- It can alert users when they are multi-posting (most likely 3+) and tell them to place all replies in one post, linking to the relevant Bitcointalk rule.
- Depending on the plagiarism tools available, it can alert a user when a post risks being classified as plagiarism and tell them to provide a link to the source material - reducing the number of ban appeal threads of people who did not know this was a rule.

Limitations:

- JARPI cannot create new topics.
- JARPI cannot quote-reply (yet).
- J.A.R.V.I.S is a newbie account and thus currently has the same restrictions as other newbie accounts.
- JARPI will not post on local boards, or on English boards where it makes no sense to ask a Bitcoin question there (e.g. News, Gambling, Marketplace > Goods).

Roadmap:

- Currently there is a single sample reply that is hardcoded in the script. Work is underway to allow any arbitrary reply to be passed. Done
- Apply aforementioned BBcode and footer. Done
- Prompt the operator for a thread ID to make the reply on.
- Allow the operator to insert any quote using a quote ID that is in the BBcode.

---

Examples of how JARPI can place high-quality answers on common questions:


ChatGPT's reply:


JARPI reply (after manually cleaning up the AI-generated reply):


Keep in mind that the basic idea is for the reply to be as correct as possible, so in this case, replies from other parts of the thread could also be put inside the blender, and irrelevant parts of ChatGPT answers are shaved off.

---

Tips: Send them to bc1q4djl6pxt90nfs8fufdul26ufxukxxrczsfjj0h