Using AI in writing

[trio]

I have been catching some flak on the issue of me using AI writing my stories. And I think this warrants a larger discussion.

I am an engineer and I have been involved with self learning system almost my entire professional career, from the first professional use of neural networks, over the time we called it machine learning, until now, when the world fell in love with its marketing name.

It is not intelligent, not by far. Most systems in use are just systems trying to “guess” the next word. They are very complex systems and when the model is trained with good data, it can output some cool stuff. But don’t forget, it all comes from data that was fed through its learning algorithms.

I have been testing models for a long time now and I am running more and more into their limitations. Just last week I uploaded an epub book that is still under copywrite protection (I own the book and the model does not store it, just uses the text within the confines of the session). The model was not initially trained on it. Then I started to ask questions about the characters in the book. I did not get any decent answer and I used the professional ChatGPT model to do it.

These systems give you bullshit answers with the confidence of a flat-earther wielding a high school diploma and a YouTube playlist. (Yep I used AI for this sentence as I could not find a good reference so quickly … and if you are a flat-earther, keep going strong my man)

There are now code generating AI platforms that have been trained with the data from GitHub and Stack Overflow. More and more of the code on these platforms is now AI generated, so future models are being trained on code that was not written by humans. And believe me there is some crap floating around there, first written by humans and now by AI.

Like with all complex systems, it is bullshit in, bullshit out. But people seem to trust these systems and its output more and more.

Another one of my pet peeves is context. AI is really bad at context. It takes everything at face value. You know the sayings about assumptions don’t you? And AI is proving that those are not wrong. Just have fun with letting it translate (something I know some of you are doing with whole texts, and that works as there is more context available from the start). But just type in a sentence in your native language, one that can have different meanings and let it translate. Then start to feed it more context, one step at the time. It is so much fun seeing it struggle and in the case of ChatGPT apologising over and over again.

Context is one of the most complex problems in human communication. It is know that it is impossible to formulate a thought and convey this to another person, being 100% sure that the whole message you tried to convey is understood. As humans we ask questions, show our emotions, have a conversation until we are confident the other has understood what we meant to say. A machine tries to do this in one go, and is really bad at it.

Also these systems are using more and more of our resources, they are already in the 1 or 2% of our global energy usage and it is climbing fast. And still, they are outputting a lot of crap. There is a limit on what they can do and we are starting to find out what those limits are.

Back to me. I do use AI and its ability to finish my sentences, or to rewrite some of my texts. I also use it to summarise. But I never use its first version because I have never trusted or even liked its initial output.

But to be clear, AI does not write my stories. I write them. I just use it to organise my thoughts, to rewrite my poorly formulated sentences, and yes sometimes I let it predict what will come next, certainly when I am stuck in a plot line.

I never liked my writing if I am honest, I don’t mean the stuff I am writing about, but more how it flowed on the paper. I could use a professional editor all the time, cleaning up after I formulated all of my ideas. Well AI helps me with that now, it makes my writing look good to me. I now can read them without feeling embarrassed, absolutely not the same level of some of you fine people here. But I am happy with it. Uh, I almost forgot to mention, the image I use as my avatar is AI generated too.

I am just throwing my thoughts out there, just let me know what you think.

p.s. I just made ChatGPT output a rape story without the filters trip, I added it to an other discussion on this board, you can find it here.

[Lucius]

Trio, you've written and posted fourteen new stories in little more than a month. Which makes you the most prolific RavishA author by far.

Writing that many stories should obviate the need for AI assistance in no time at all. My advice is to do the work.

[trio]

No, no, this is work of more than 20 years. I never published before. But I understand your concern. But this is one of these things about context, and assumptions. I have never posted my stories anywhere until I decided to post them here. Am I so different from other authors that repost their old work they posted earlier on other messaging boards?

I think I have done the work. So please don't assume that you think you know what is going on. I still have like ten stories left, and I had time now as I am between projects. I am just having a bit of fun.

Trio, you've written and posted fourteen new stories in little more than a month. Which makes you the most prolific RavishA author by far.

Writing that many stories should obviate the need for AI assistance in no time at all. My advice is to do the work.

[Shocker]

I’m slightly confused how you owning the book, allows you to use it for the training of an AI. Do you own a copy of that book, or co you own the copyright if said book because you wrote it?

[trio]

I am not training the model. This is not how AI works. When the system, like ChatGPT is in place it is fully trained and you as the user gets to ask question and let it do stuff. Ask the model you are using what its cut of date was. This the date that the developers stopped feeding it data. It is often more than a year in the past. So if you ask it about something recently, you see it searching the web for information, not using its own model.

By uploading the text I did not train it. It says it clearly in the user agreement documentation. It will use the input within the session, but it is not used to train the model. Just ask it clear details from like say Lord of the rings. It will not be able to give you that info.

The same when you upload your personal writing to let it summarise or do whatever with it, the training algorithms will not touch it. Of course when you start using models that don't limit themselves to these practices, the gloves come off but OpenAI needs to be careful in the eyes of the law.

And talking about what I did, the law is currently not clear (I check it for a client just last year). Nobody but me could do anything with the data I uploaded and after the session was deleted, all data was cleared (or should have been). And to be clear, I own a copy of the book.

I’m slightly confused how you owning the book, allows you to use it for the training of an AI. Do you own a copy of that book, or co you own the copyright if said book because you wrote it?

[trio]

@Shocker, I just checked and the current model of ChatGPT I use professionally. Its training data goes up until June 2024.

[Shocker]

…

By uploading the text I did not train it. It says it clearly in the user agreement documentation. It will use the input within the session, but it is not used to train the model. Just ask it clear details from like say Lord of the rings. It will not be able to give you that info.

The same when you upload your personal writing to let it summarise or do whatever with it, the training algorithms will not touch it. Of course when you start using models that don't limit themselves to these practices, the gloves come off but OpenAI needs to be careful in the eyes of the law.

And talking about what I did, the law is currently not clear (I check it for a client just last year). Nobody but me could do anything with the data I uploaded and after the session was deleted, all data was cleared (or should have been). And to be clear, I own a copy of the book

I do disagree, you do train your model, or for what other purpose do you need to upload the text into it. And the fact that nobody but you could use that data, didn’t change the fact that you used a piece of copyrighted material to create have something different created. I’m no lawyer, but as far as I recall that is not the extent of rights you gave gained by buying the book.

But you do what you want, I simply don’t like reading AI generated stories, and am somewhat prickly when it comes to copyright.

[Lucius]

No, no, this is work of more than 20 years. I never published before. But I understand your concern. But this is one of these things about context, and assumptions. I have never posted my stories anywhere until I decided to post them here. Am I so different from other authors that repost their old work they posted earlier on other messaging boards?

The first publication of a story makes it a new one. Say, a hypothetic publication of someone's Nachlass is a new book, and it doesn't really matter how long ago the texts were penned.

I think I have done the work. So please don't assume that you think you know what is going on.

I just use [AI] to organise my thoughts, to rewrite my poorly formulated sentences, and yes sometimes I let it predict what will come next, certainly when I am stuck in a plot line.

Isn't that what's going on?

[trio]

Sorry but that is not how it works. I am using the model to read the text and ask it to answer questions. I am not training it. Training an AI and using it are two completely different things. Some companies will use the data of their users to retrain the models, not the model I used in my experiment.

I don't want to assume anything here, but some people have no clue on how these models really work. And you can be picky about copyright, but what are you going to do when every OS has AI build into its core, like most of them have already? It will read every piece of information you have on your devices to "help" you. It is exactly what I did manually, I have done the legal research and I know exactly what is coming over the horizon.

And I hope to say this for the last time. My stories are not AI generated, I use AI as a tool, it was me who generated to story. People like to talk about AI generated stuff all the time, but most have no clue that almost all the text in media have been run through an AI system in one form of another. They have better people than me at the controls, but there is almost no industry on this planet that is not using AI today.

But I give up, I have talked about machine learning and adaptive models for a long time. I tried to be open and honest on here, but the more I do, the more frustrated I get honestly.

I do disagree, you do train your model, or for what other purpose do you need to upload the text into it. And the fact that nobody but you could use that data, didn’t change the fact that you used a piece of copyrighted material to create have something different created. I’m no lawyer, but as far as I recall that is not the extent of rights you gave gained by buying the book.

But you do what you want, I simply don’t like reading AI generated stories, and am somewhat prickly when it comes to copyright.

[trio]

The first publication of a story makes it a new one. Say, a hypothetic publication of someone's Nachlass is a new book, and it doesn't really matter how long ago the texts were penned.

Yes that is the definition of being new. I just wanted to say that more than 90% of the stories had been written a long time ago.

But like I said, I give up. I think I have posted my last story here. I'm done defending myself against people who come at me with their own preconceptions.