It would be a fun experiment. If nothing else I hope to be impressed by the content that I generate: I've already seen some incredible things coming out of GPT-2 and 3.

And hopefully, we're able to see just how sneaky an account farmer could be.

What is understanding, anyway? Epistemically speaking, human experiences are the foundation of our knowledge.

https://www.lesswrong.com/posts/L5JSMZQvkBAx9MD5A/to-what-extent-is-gpt-3-capable-of-reasoning

Some of the example scenarios here are pretty cool. Still a long ways to go before we can officially label it "Artificial Intelligence". I would call these prototypes "recycler networks" that simply recycle information in a new way.

---

Here's an example of a single prompt and continuous writing, no redos.

Cross posting into a more appropriate thread
I don't think this person stands to gain anything by making two posts with a model someone else has created. My theory is whoever is behind Rohani6360 is using various forum accounts to make these kinds of posts, and is using the number of accounts that are banned and/or the number of posts that get removed as a measure of performance.

Look at his other post:
This looks like a plagiarized article, but multiple google searches of sentences in his post come back with only his post. lovesmayfamilis was able to find the original article. I don't think he found the correct link though because the link lovesmayfamilis provided does not have the "related read" links under the paragraphs.

In this post, he is probably using some kind of text spinner or perhaps a GAN of some sort intended for the use of plagiarism.

You are contradicting yourself here. Someone with the skill set to create a model that can write semi-coherent posts, and actually post using a bitcointalk account is not going to be wasting their time trying to farm accounts. Their time will simply be too valuable for someone like this to earn coin like this.


I don't think this is the case of the "user doesn't want to do much work" I think this is the case of the user wants to have their posts be 100% automated, without intervention without humans. The latter would allow someone to do much more than earn a few hundred dollars per week farming accounts. It would allow someone to effectively create propaganda using a troll farm. After the Wall Street Journal published an opinion piece with the headline "China Is the Real Sick Man of Asia" that was critical of the Chinese government, the Chinese government denounced this as racist, and flooded the Editorial Board's email inbox with "complaints about the headline, all containing remarkably similar language and demanding an apology". It is common for propagandists to send form letters to elected officials purporting to be from their constituents advocating for a particular cause.

If someone could write 5000 letters, all conveying the same message, but are written differently, it would be possible to make it appear there is public outrage about something that has public support. The problem is that if the recipient of these letters can detect they are the same, the letters will lose all credibility. Hence an attacker would want to try to use their working model where there are robust measures to detect plagiarism.

I think it will become a new set of skill.
As we are (supposed to be) now quite proficient at understanding if a piece of information found on the internet is reliable or less, I guess in the future we will be able to detect an human produced text, or a deepfake video or other AI-generated media.

I am not aware of any publically available GANs (specifically generator models) that are pretrained and ready to generate content. (there are a few websites that will use a pretrained generator model to generate an image).

Most of the GAN architectures I am aware of involve images, and a few involve generating sounds. I have built generative models involving text before, but they were not GANs, and they could not come close to generating the two posts we are discussing (they would look at an image and create a caption of the image).

Even if one were to implement a GAN architecture that already exists on a text dataset to create their own GAN to create posts, they would need to have the skills that far outweigh the possible income generated from farming accounts. You can know how much you can earn from farming accounts by reviewing pay rates of signature campaigns, and estimating how many accounts could be farmed at once. I would question if someone could earn anything at all from farming accounts using AI because I am not sure if they would earn enough merit to rank up to even a Junior Member. 


The type of content we allow here is very broad, we do not allow plagiarism, and are under consistent spam attacks because we allow our content to be indexed by search engines (which is not the case for many social media platforms). We also require posts to be coherent and replies to be related to the OP of a thread, neither of which apply to other social media platforms. Someone tweeting incoherent nonsense here would probably get banned, while someone tweeting the same thing would simply be ignored by anyone who reads the tweets on Twitter.

This is actually how GANs work. A generator model will create fake content and a discriminator model will try to differentiate between the real and fake content (aka try to detect the fake content).

---

Here is one example as to why someone may want to use the forum to test a model that is ultimately deployed elsewhere:

The reddit sub 'WallStreetBets' has been credited for causing GameStop (and other stocks) to go up tenfold in a number of days. Some of this is likely due to a short squeeze and a flurry of call options buying. I am curious if there are any bots interacting with redditors that are creating the appearance of more hype around the stock than there really is. Bots posting content and interactions on 'WallStreetBets' would not receive the same scrutiny as they would here. I don't have any actual knowledge one way or another that this is actually happening.

If the above were to be employed, someone could potentially know when many investorsspeculators are going to be bidding up certain stocks in advance and could trade accordingly.

I am not so sure about this. Large institutional investors want to keep their positions private when they are in the process of buying or selling a stock because when they start to buy or sell, it generally means they will buy or sell a lot of said stock. A retail investor on the other hand will not typically trade large enough amounts to move the market. There are also other ways to monitor retail trading activity, such as monitoring odd lot trades.

What I was referring to was someone actually creating the illusion that there is interest in a stock when said interest does not exist, but the illusion of the interest creates actual interest.

Well, some of them managed to make it to signature campaigns or have signatures that might belong to one which kinda surprised me in the first place