Jump to content

I'm tired of AI accusations and nonsense


russflex

Recommended Posts

Life was so much easier as a writer on Fiverr before AI became a thing. Now I get clients who think they are AI experts and falsely accuse me of using AI because they think my writing "sounds like AI," as if they know what it sounds like. It's tiresome, annoying, and insulting because I am actually doing the work myself. I never use AI. 

  • Like 14
  • Sad 2
  • Up 1
  • Insightful 1
Link to comment
Share on other sites

You could show them proof by showing the history of your document if there's that option (eg. in Word) and maybe show sources/research done/writing process.

You could also show them your content checked in a few AI checkers that show it's human (while also pointing out that AI checkers aren't accurate).

Maybe point out you've been doing this and have many reviews since a long time, well before the AI large language models like the ones used by ChatGPT (such as GPT 3/3.5? Gpt 4) started being used (though maybe don't use the word reviews in case it flags something).

Edited by uk1000
  • Like 12
Link to comment
Share on other sites

1 hour ago, russflex said:

It's tiresome, annoying, and insulting because I am actually doing the work myself. I never use AI. 

Welcome to my world. 

I agree, people trust AI tools searching for AI rather than a real writer's knowledge and expertise. And the worst thing is that even if there are many tools to search for AI, people either trust just one, or 2 of them, and disregard the others. It's disappointing, especially when you see that even after a minor change, the "AI" percentage is very very different. It's clear these tools are unreliable. That's why I am at a point right now where I vet buyers before placing the order and learn what tool they want me to check with, if they actually care about AI at all.

1 hour ago, uk1000 said:

You could also show them your content checked in a few AI checkers that show it's human (while also pointing out that AI checkers aren't accurate).

 

Most buyers will usually disregard that if they have an AI checker they prefer...But obviously they won't say anything about what tool they want to use until you deliver, then all of a sudden they accuse you of using AI, which is I assume is the exact thing that the OP dealt with as well.

1 hour ago, uk1000 said:

Maybe point out you've been doing this and have many reviews since a long time, well before the AI large language models like the ones used by ChatGPT (such as GPT 3/3.5? Gpt 4) started being used (though maybe don't use the word reviews in case it flags something).

I had this happen myself, and buyers don't really care. I just finished talking with a person right now and it's exactly what OP said. They didn't even share the exact requirements, they just wanted me to confirm the content will pass their AI tool, without even telling me what that tool is. "If it's human content, it will pass". How can a writer guarantee anything without using the tool himself? Seems like a severe lack of trust, and who would start a business relationship based on that? I always skip these "potential orders" because they smel like a scam to me.

  • Like 13
  • Up 1
  • Insightful 1
Link to comment
Share on other sites

Non personalised writing, as a field, is dead. The only writers that will still have business very soon are the ones who have a great network of contacts, that want to hire them specifically. Anyone looking to just have some writing done, and have no preference for an individual, will be very well served by AI, if not now in the very near future. Writers that survive will be selling their personality, their brand, more than the writing itself - because most writing (the vast majority) just needs to be functional, to work for a certain purpose, and AI will nail that sooner or later (it will be sooner). But people will always like to help each other, and friends will always want to hire friends. A lot of people will prefer hiring a human of their trust instead of writing prompts themselves, so the most successful freelance writers will be the ones who are more appealing on a human level, customer relations, and that can they have an AI do all the work, and they simply act as the bridge, the human connection, that a lot of clients want and demand. They'll be essentially prompt writers and, above all else, customer relationship managers. And they will be able to do it on a massive scale, since all their work will be talking to clients, the writing itself, which used to be the most time consuming part, will be automated away.

It is what it is.

Edited by visualstudios
  • Like 12
Link to comment
Share on other sites

23 minutes ago, visualstudios said:

Non personalised writing, as a field, is dead. The only writers that will still have business very soon are the ones who have a great network of contacts, that want to hire them specifically. Anyone looking to just have some writing done, and have no preference for an individual, will be very well served by AI, if not now in the very near future. Writers that survive will be selling their personality, their brand, more than the writing itself - because most writing (the vast majority) just needs to be functional, to work for a certain purpose, and AI will nail that sooner or later (it will be sooner). But people will always like to help each other, and friends will always want to hire friends. A lot of people will prefer hiring a human of their trust instead of writing prompts themselves, so the most successful freelance writers will be the ones who are more appealing on a human level, customer relations, and that can they have an AI do all the work, and they simply act as the bridge, the human connection, that a lot of clients want and demand. They'll be essentially prompt writers and, above all else, customer relationship managers. And they will be able to do it on a massive scale, since all their work will be talking to clients, the writing itself, which used to be the most time consuming part, will be automated away.

It is what it is.

Hm...
I definitely think that my job (I started as a short story writing) has evolved a lot as a freelancer - even before AI. AI for now still isn't quite 'there' - but it's readable material for the most part. 

That being said, I don't think all is lost (yet.)

For me, this year has been better than expected (not great, but a comeback after the second half of last year. 😄 AI can do writing, but it can't do puzzle design/etc. as well. It can't mesh well with a team of game designers (yet). So... basically exactly what you said (just in a more niche field.)

I do realise not everyone is as lucky as I am, though (and that right now everyone wants to make a game to get rich, ha.)

I think in general, AI still has a long way to come, but it's definitely improving much faster than I'd ever thought.

For now though, most of the issue I have with it is people asking if I use it (prior to ordering), asking me to use it (to cut down costs), or sending me AI material that leaves a lot to be desired. (the last one is an issue because they tell me to follow it, which I do...and then we both hate it, which creates more work. I think for now, the majority of people still don't know how to write efficient prompts, either. It's not something I messed around with too much, but for now I don't feel too much at risk (then again, I would say that I do a lot more than just writing now, since 90% of what I do is game design at this point. 

3 hours ago, russflex said:

Life was so much easier as a writer on Fiverr before AI became a thing. Now I get clients who think they are AI experts and falsely accuse me of using AI because they think my writing "sounds like AI," as if they know what it sounds like. It's tiresome, annoying, and insulting because I am actually doing the work myself. I never use AI. 

Honestly, a lot of people can't tell the difference! Especially if you have a more professional tone or you are writing something very technical, it CAN sound AI-generated. I'm curious as to how this will be filtered later - both on here and just in general. 

Edited by katakatica
  • Like 11
  • Up 1
  • Support 1
Link to comment
Share on other sites

@katakatica The more "creative" and "human" your work is, the better - for now. Novelists won't be out of a job for now. Poets won't be out of a job for now (eh, they were already out of a job for decades, what am I saying, writing poetry ain't paying any bills).

But run of the mill writing? Website copy, product descriptions, mailing lists, etc.? Which is, like, 90% of freelance writting? (citation needed) That's dead with AI. Dead. That is, the act of writing itself. People will still hire people to be in charge of that, since they don't know how to prompt the models properly, or they don't want to have to deal with it, to polish it, edit it, etc. 

Edited by visualstudios
  • Like 11
Link to comment
Share on other sites

AI content detectors aren't reliable at all. I myself have tested these "major" tools and they have consistently identified my own writing to be AI-generated lol.

Don't take it from me. Open AI themselves have done testing on these AI content detectors, and apparently they have found that these detectors have flagged even Shakespeare's writing to be AI-generated lmao. So yeah.. there's that.

Here's the link: 

Quote

In my opinion, there's no way to definitively tell that a piece of writing is generated using AI (unless it's really badly generated). Also, there's not much of a difference between human and AI written content at face value from what I've seen what AI can pull off. What makes the difference (in my opinion) are the nuances and tiny errors that humans make. To be honest, even tiny errors are out of the question since most writers use Grammarly nowadays.

To sum things up in my opinion there's no real difference between AI-generated content and human-written content (and that's totally okay. It'll take some time for more people to realize this). I know it's insulting to hear as a writer, but it is what it is.

What AI cannot replicate though is the style of a writer (for example, I've noticed that I'm a bit sarcastic when it comes to writing, and that's how I talk too.. I say things like "so yeah" / "and there's that" which would be really difficult for an AI to replicate).

Of course, this doesn't apply to complex technical subjects that are known as "Your Money, Your Life (YMYL)" content.

What I'm trying to point out here is that this is such a gray topic and no one would really be able to prove that a piece of content generated using AI by someone who knows what they're doing to be AI-generated.

So yeah.. I think Fiverr needs to do something about this (with disclaimers or something) to ease the tension between sellers and buyers as it would be ridiculous to hear someone say that the content piece that took you over 3 days to write is AI-generated. -_-

  • Like 11
  • Up 1
Link to comment
Share on other sites

Further to what I've said above, and reading what @katakatica and @visualstudios said, it's safe to say that in the future the role of a writer will be to edit content pieces generated by AI instead of writing things from scratch. We might need to be open-minded to embrace that to avoid being left out like what happened to that Camera manufacturer (Kodak if I remember correctly) who refused to make the switch to digital.

Edited by creative_howl
  • Like 10
Link to comment
Share on other sites

None of the pro-AI people realise that AI can generate duplicate content for multiple people at any given point of time and its quite possible that multiple businesses would end up with same content/copy for their marketing assets. In due time, everything and everyone would start sounding pretty much the same (if not identical). 🫠 Generative AI feeders are already using AI-generated content from the web to generate more AI content. 

Wish humans could recycle plastics and chemicals instead of words!! 😂

And not to forget the fact that AI also has a certain writing style and it loves using certain words (to de*th) - so much so that it is easy to identify AI content within a few seconds. AI is capable of providing a variety of tonalities and styles too but it cannot replace human rational, emotions and critical thinking capabilities. 

PS It is easy for people who are not professional writers to spell DOOM and even write obituaries for the writing fraternity. And same way, content writers can look at Dall-E or Sora to assume that other professionals have no future left!!

Edited by priyank_mod
  • Like 11
  • Up 1
Link to comment
Share on other sites

2 hours ago, priyank_mod said:

PS It is easy for people who are not professional writers to spell DOOM and even write obituaries for the writing fraternity. And same way, content writers can look at Dall-E or Sora to assume that other professionals have no future left!!

There will always be people that hire a professional writer for good content. Despite the doom and gloom that others mentioned, I don't think there's any major issue. There's still quite a lot of demand in the writing space, and in my experience, a lot of people took the AI route, but now they are coming back because AI ruined their website. And it's not an isolated incident, either. I see a lot of people doing this.

AI will always be an alternative for someone that wants writing, video creation/editing, even programming. But as mentioned, it copies other places, it uses those as inspiration, so it's not something unique.

The problem here is not that. Instead, OPs point is clear, AI checkers are a problem, because some buyers specifically rely on AI tools rather than the experience of a writer. It's not that often, at least for me, but it does happen. You will have someone that couldn't care less what you write, as long as it's not seen as AI by their tool. Last night I had a particularly interesting discussion with a person that said they want me to rewrite all their website content, which was fine, but it needs to pass their AI checker, without obviously telling me what tool is that. Who would agree to such a task that has a very high probability of a refund?  

  • Like 10
  • Up 1
Link to comment
Share on other sites

3 hours ago, priyank_mod said:

None of the pro-AI people

Not Pro or Anti, I just recognize what it's coming, and what is unavoidable.

3 hours ago, priyank_mod said:

In due time, everything and everyone would start sounding pretty much the same

Welcome to the corporate world... they already mostly want to sound the same anyway. Even their "creativity" is formulaic.

3 hours ago, priyank_mod said:

And not to forget the fact that AI also has a certain writing style and it loves using certain words (to de*th) - so much so that it is easy to identify AI content within a few seconds.

AI is not chatgpt. Also, give it time.

3 hours ago, priyank_mod said:

PS It is easy for people who are not professional writers to spell DOOM and even write obituaries for the writing fraternity. And same way, content writers can look at Dall-E or Sora to assume that other professionals have no future left!!

Yes, they can - it's a matter of time. Illustrators are basically out already, to a large degree. Voice actors very soon. Eventually video editors, musicians, actors. I'll give it 5-10 years. Gotta get the bag in the meanwhile.

Edited by visualstudios
  • Like 9
  • Up 1
Link to comment
Share on other sites

8 hours ago, creative_howl said:

To sum things up in my opinion there's no real difference between AI-generated content and human-written content (and that's totally okay. It'll take some time for more people to realize this).

One difference will be in the rights. Currently AI generated content can't be copyrighted but human-written content can. If Google detects AI written website content it was said (I think on this forum) that it wouldn't rank as well. So if buyers want to own all the rights then they'd want human writers to create it.

3 hours ago, priyank_mod said:

None of the pro-AI people realise that AI can generate duplicate content for multiple people at any given point of time and its quite possible that multiple businesses would end up with same content/copy for their marketing assets. In due time, everything and everyone would start sounding pretty much the same (if not identical).

They can I agree, but it depends on the model and the prompt & context and probably other parameters (like temperature etc.). So with the values set for more randomness it should be less likely to generate the same content. Probably having a bigger AI large language model should also help it generate more unique content.

9 hours ago, creative_howl said:

What AI cannot replicate though is the style of a writer (for example, I've noticed that I'm a bit sarcastic when it comes to writing, and that's how I talk too.. I say things like "so yeah" / "and there's that" which would be really difficult for an AI to replicate).

It probably can to some extent. But some AI models aren't very good at it. You could ask an AI to write an article in the style of a famous character or like Shakespeare and it might. If it trained (probably a fine tuned model as it might be too difficult to train one from scratch) on your content it might be able to do it (or if it's part of the context/prompt).

  • Like 9
Link to comment
Share on other sites

4 minutes ago, uk1000 said:

One difference will be in the rights. Currently AI generated content can't be copyrighted but human-written content can. If Google detects AI written website content it was said (I think on this forum) that it wouldn't rank as well. So if buyers want to own all the rights then they'd want human writers to create it.

The point is there will be no difference. Specially in writing (but eventually in everything, an image is just pixels, a video is just a series of images), a sufficiently advanced AI is indistinguishable from a human, that's the whole point. We're not quite there yet, but there will come a time where an AI generated image and a photo will be literally indistinguishable, at a pixel per pixel level. From there, video will follow. Sound as well, both music and voice, byte by byte, the same as "real" captured audio. Then what? 

Sure, some models can insert digital signatures on the files, on purpose, to say they are AI, but anyone can run a model locally (eventually), and remove those. In the future, nothing will be human and everything will be human, the distinction will be pointless.

Edited by visualstudios
  • Like 8
  • Up 1
  • Thanks 1
Link to comment
Share on other sites

34 minutes ago, visualstudios said:

Sure, some models can insert digital signatures on the files, on purpose, to say they are AI, but anyone can run a model locally (eventually), and remove those. In the future, nothing will be human and everything will be human, the distinction will be pointless.

34 minutes ago, visualstudios said:

The point is there will be no difference. Specially in writing (but eventually in everything, an image is just pixels, a video is just a series of images), a sufficiently advanced AI is indistinguishable

The point will be (for now, at least until the laws are made clearer), to make sure you have the rights though and be able to prove it, as well as probably not fail AI detectors (if they're accurate, which they're not). eg. they could create video using AI video generators and sell those if they're good enough, but if the AI they used was trained on copyrighted content without permission it could get them and the AI creators into legal trouble. The same with chatgpt content. Though millions of people have used it and millions of people are unlikely to get into legal problems because of that (but many will be using it not really for profit).

If AI content was created for a book, the human person who owned the book (got it published) wouldn't own the copyright of it as just AI content can't be copyrighted. Which could lead to anyone copying the book. They couldn't sell the rights to it (eg. for use in TV/film) as they wouldn't own them. The sora demos had images that I think an interviewer thought were too like existing content (because it might have been trained on youtube content etc.), which could also lead to legal action with current laws. But if someone paid someone to create the text content, they'd own the rights after it (depending on the gig).

Edited by uk1000
  • Like 9
  • Thanks 1
Link to comment
Share on other sites

25 minutes ago, uk1000 said:

The point will be (for now, at least until the laws are made clearer), to make sure you have the rights though and be able to prove it, as well as probably not fail AI detectors

But the point is that can't be done if it's indistinguishable. With text it's particularly trivial, but it can and will eventually be applicable to any other digital medium. Like, if I have a model that can create video of an elephant, how can anyone prove I didn't film that elephant myself, if the file I get out of it is exactly the same, byte by byte, as an actual shot of an elephant taken by a videographer (including metadata)? You can't. 

That's the problem here. If you compose something, write something, film something, record something, you don't get a receipt and title of ownership on a centralized ledger, like if you bought a car or a house. Nothing can be proven. "Oh, i get the project files" - AI can do those as well, it's all bytes. Any digital information can be created out of thin air, and it will be impossible to tell, that's the point. Any byte is the same as any other byte.

"Prove I have the rights"? How? I just wrote a poem on my computer. How can I "prove" it's mine? I can't. AI may have written it, I may have copied it from some obscure source, a random mystery person may have used my computer to write it.

I just composed a track on logic. How can I "prove" I did it? I can't. I may have lifted all the melodies from somewhere else, I may have used AI to do it, etc.

I just have the digital files. Even if I record myself doing it, how can I "prove" that video is real, and not AI made? I can't. If there are "authenticity receipts", those are digital as well, so they can also be replicated and made up.

Nothing can be "proven".

Edited by visualstudios
  • Like 7
  • Up 1
  • Thanks 1
Link to comment
Share on other sites

51 minutes ago, visualstudios said:

In the future, nothing will be human and everything will be human, the distinction will be pointless.

Let's not focus on the doom and gloom future. Who knows what will happen. What we do know is OP's situation, there are a lot of writers that are accused of AI due to random, unreliable AI checkers. And for me, the best solution is to just vet buyers and understand the tools they use, if I am ok with those, etc. If there's any red flag, I just don't accept the project or, if the order was made without my consent, it's better to jus cancel it entirely. 

As for creating stuff, you can just copyright the stuff you created. There are ways to protect your stuff. But I do agree that AI is killing a lot of niches, and that will end up removing a lot of meksells from the platform, especially when it comes to writing. 

  • Like 7
  • Thanks 1
  • Support 1
Link to comment
Share on other sites

14 hours ago, uk1000 said:

It probably can to some extent. But some AI models aren't very good at it. You could ask an AI to write an article in the style of a famous character or like Shakespeare and it might. If it trained (probably a fine tuned model as it might be too difficult to train one from scratch) on your content it might be able to do it (or if it's part of the context/prompt).

100%

I've been really impressed with what AI can achieve nowadays, and it's remarkable to see how technology has developed in a short amount of time like this. 

Most people are scared that AI is gonna replace them. And those are the ones who either don't understand it or, expect it to do everything for them.

I'd say the smart ones will capitalize on that by using it to HELP them instead of replacing them. It's like hiring an assistant who is ready at all times. 

I mean, I've seen plenty of services out there that offer to "humanize" AI content so these detectors won't show it as AI written. Now what?

I was intrigued by the AI model thingy that one of you guys posted on that group about an AI that can scan messages. I think it was custom built by someone right? That thing was amazing.

 

  • Like 6
  • Up 2
  • Thanks 1
Link to comment
Share on other sites

14 hours ago, uk1000 said:

One difference will be in the rights. Currently AI generated content can't be copyrighted but human-written content can. If Google detects AI written website content it was said (I think on this forum) that it wouldn't rank as well. So if buyers want to own all the rights then they'd want human writers to create it.

 

I don't think this would apply to written content as it's just words arranged in a certain order that makes sense in a nutshell right? 

Google has stated (as far as I know) that they don't "de-rank (honestly I don't even know if that's an actual term lol)" content that is AI generated as long as it's quality and useful. This says a lot honestly.

At the end of the day, it's all about how you use these things. Many people say that Fiverr is against VPN's but this is not the case as they have clearly stated they're not against that. And yet I see so many times that people spread misinformation stating that VPN's can get you banned. Any thing can get you "banned" if you use it to deceive and manipulate.

https://developers.google.com/search/blog/2023/02/google-search-and-ai-content

Edited by creative_howl
  • Like 6
  • Up 1
  • Thanks 1
Link to comment
Share on other sites

6 minutes ago, creative_howl said:

I don't think this would apply to written content as it's just words arranged in a certain order that makes sense in a nutshell right? 

If you want to sell a book on amazon you'd need to declare to them if it was written with AI (I haven't tried but that's what I read). They might do something with that setting in future (eg. if they think there's low quality AI written books being sold that are supposed to be factual but aren't).

If you wanted to sell some rights to it for film etc. you'd need to own the rights first, and you wouldn't if it was all chatgpt content.

If you wanted to register it with the US copyright office, as far as I know, they wouldn't currently register it if it's 100% AI content. There might be more legal risk if it's all AI content too (in case of plagiarism). Though it could be checked with plagiarism checkers, but the AI might have learnt it from sources that the plagiarism checkers don't know about.

Edited by uk1000
  • Like 8
  • Thanks 1
Link to comment
Share on other sites

2 minutes ago, uk1000 said:

If you want to sell a book on amazon you'd need to declare to them if it was written with AI (I haven't tried but that's what I read). They might do something with that setting in future (eg. if they think there's low quality AI written books being sold that are supposed to be factual but aren't).

If you wanted to sell some rights to it for film etc. you'd need to own the rights first, and you wouldn't if it was all chatgpt content.

If you wanted to register it with the US copyright office, as far as I know, they wouldn't currently register it if it's 100% AI content

But you would lie, and there's no way to know, if the AI is good enough. And the AI will be good enough.

  • Like 8
  • Up 1
  • Thanks 1
Link to comment
Share on other sites

Just now, uk1000 said:

Though it could be checked with plagiarism checkers, but the AI might have learnt it from sources that the plagiarism checkers don't know about.

Very true. I honestly have no idea about copyright laws. But regarding the Google thing, I've actually taken it upon myself to start a blog from scratch using content responsibly written by AI (I will explain what I meant by "responsibly" if and when I might share the results for the sake of it xD) just to see what the fuss is about. And it's doing pretty well so far for a brand new domain blog. Apparently so many people have been hit by the August algorithm update, but that's not the case for my little test subject. It's been smooth sailing for me since the day I started it. Granted it's new (about 2 months old), but I'm seeing constant growth. So there's that.

Regarding copyrights, there could be coincidences where a form of plagiarism (an exact duplicate of a sentence or more) is written by an actual human being with no intention to plagiarize too right (hypothetically)? This whole subject is big ol gray area in my opinion as there could be so many false positives.

  • Like 7
  • Thanks 1
Link to comment
Share on other sites

32 minutes ago, visualstudios said:

But you would lie, and there's no way to know, if the AI is good enough. And the AI will be good enough.

If you want to take the risk, yes. There might be big fines or other legal issues if people get found out though.

24 minutes ago, creative_howl said:

Regarding copyrights, there could be coincidences where a form of plagiarism (an exact duplicate of a sentence or more) is written by an actual human being with no intention to plagiarize too right (hypothetically)?

Yes that's possible. It's unlikely to be a problem if it's one that could be done by chance by a human. For AI content though, the AIs could have been trained on  many copyrighted books that the plagiarism checkers might not have access to to check. So in theory an AI might generate whole sentences or storylines from/based on one or more of those books. Though it might be quite unlikely.

Edited by uk1000
  • Like 7
  • Thanks 1
Link to comment
Share on other sites

1 hour ago, uk1000 said:

There might be big fines or other legal issues if people get found out though.

But how can people be found, that is the question. It's all bytes (I'm mainly talking about text here, as that's one of the few areas where we are pretty much already there - but it will apply to others sooner or later).

  • Like 6
  • Thanks 1
Link to comment
Share on other sites

8 hours ago, visualstudios said:

But how can people be found, that is the question. It's all bytes (I'm mainly talking about text here, as that's one of the few areas where we are pretty much already there - but it will apply to others sooner or later).

For other types they're already talking about adding identifiers for AI content (though it might be a problem for open source AIs where people could remove it more easily), like having all cameras add something to the file I think (maybe it was a certain place in the USA making that a law).

For text they already have detectors but they're inaccurate like has been said, but they could train AIs to recognise the difference.

Or they could add some sort of text marker/watermark within the text. Though once people find out there might be other people writing software to remove them/change them (though those could get made illegal and removed when found). Some sort of text marker would probably work best in longer output, where it could place text/characters in a way that would be very unlikely for a human to do. Like encoding the name of the software and version number somewhere in the text not as is but in a hard to detect way. Maybe hidden unicode characters could be used.

This is what perplexity.ai said about it:

Quote

OpenAI has developed a sophisticated method for watermarking ChatGPT-generated text, though they have not implemented it yet. Here's how the watermarking system could work:
Cryptographic Watermarking
The proposed watermarking technique uses a cryptographic approach to embed an imperceptible pattern into the text generated by ChatGPT. This method involves:

    Modifying the AI's word choice probabilities
    Creating a pseudorandom distribution of words
    Embedding a statistical pattern or code into the text

The watermark would be unnoticeable to human readers but detectable by specialized software.
How It Works

 * Tokenization: The AI breaks down text into semantic units (tokens) like words and punctuation
* Probability Distribution: As ChatGPT generates text, it calculates the probability of each subsequent token
* Pseudorandom Selection: Instead of randomly selecting the next token, the system would use a cryptographic pseudorandom function to make selections
* Pattern Creation: This process creates a subtle pattern in word and punctuation choices that serves as the watermark

Effectiveness and Challenges

    OpenAI claims this method is 99.9% effective in detecting AI-generated content

The watermark doesn't affect the quality or readability of the generated text. It requires a sufficient amount of text to be reliable

 However, there are concerns about potential workarounds:

    Using another AI to paraphrase the output
    Translating the text to another language and back
    Inserting and then removing unique characters or emojis

Implementation Considerations
Despite having this technology ready, OpenAI has not deployed it due to several factors:

    Concerns about false accusations of AI cheating
    Potential bias against non-native English writers
    User surveys indicating up to 30% of users might switch to rival AI tools if watermarking is implemented

Edited by uk1000
  • Like 5
  • Thanks 1
Link to comment
Share on other sites

You will be able to run your own models locally. Other companies, based in other countries where copyright laws don't matter will have other models that don't watermark anything. 

It can't be done. It's like say "we'll end piracy online". No, you won't. Can't be done.

  • Like 5
  • Up 1
  • Thanks 1
Link to comment
Share on other sites

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
×
×
  • Create New...