They also state on their website that they “never use your text to train our algorithms”, if you care about these things.
2 Likes
This has changed considerably since you wrote this post. Chat bots are becoming more and more sophisticated and I now use them every day for copy editing and research chores. In fact, I now find them so invaluable that I just made a suggestion to your support staff that they find a way to incorporate them into Scrivener. I think people need to stop fearing and hating AI. It’s not going away. And for the writer—as long as you maintain creative sovereignty (meaning don’t let the bot actually write for you)—AI is a godsend. I had AI read a chapter of my latest book and it caught a logic issue that had completely escaped me. Something I would have had to pay a copy editor a couple thousand to tell me.
1 Like
At the same time, you need to be careful about taking an LLM at its word. I asked ChatGPT-5 (which is quite powerful) to help me figure out how to adapt a CRM system to support my writing business. It gave me detailed plans and ideas, and completely ignored major features of the app that provided a simpler, easier solution. When its error was pointed out, it simply said “oh yes, that’s much better.”
Good idea to double check whatever it’s telling you. If nothing else, ask another AI the same question and compare answers.
2 Likes
Yes. Always double check. Do the same even when you’re doing old-fashioned non-AI research. Don’t just take one website’s word for it. Humans make mistakes, too.
4 Likes
I just happened to come across this thread. I’ve been using ChatGPT for various tasks. I don’t use it for content generation.
More importantly I’ve been developing my own integration using OpenAI. Sort of a tool set. Things like pacing, character sketches. Although I’m working with the Windows version of Scrivener. This has only been for personal use at this point.
My basic belief is the writing is still mine, but a tool that can help me look at the bigger picture is what I’m trying to achieve.
If you have specific items you think are important I’d be in listening.
And that’s fine for you to believe, but unless you’ve set up the model you’re using with 100% your own writing, then it’s generally argued that the generated output is not “yours” but is determined by following rules that weren’t made ethically.
The creators of those training works are the ones you’re truly engaged with. ChatGPT and its kin are in essence the word-suggestion tool I have here on my phone, but this tool has been trained by me, in a sense, from its neutral state. If I begin to type my email address, it offers up a suggestion for the whole one, because I do type that a lot. ChatGPT and other models offer more choices, with greater speed and length. It’s not clear which writing(s) were distilled to make it do so.
If you were to develop such a tool, my number one request would be an OFF button, so I control its involvement. My second request would be the same as years of teachers: cite your influences. If a model can’t show me it’s been trained ethically, then what are the ethics of me using it and passing off work as my own. This way, I feel, lies plagiarism.
7 Likes
“Belief” isn’t a reliable metric when it comes to things like copyright law.
7 Likes
This is substantively inaccurate.
U.S. District Judge William Alsup of the Northern District of California ruled Monday in Bartz v. Anthropic that AI company Anthropic’s training of its Claude LLMs on authors’ works was “exceedingly transformative,” and therefore protected under the fair use doctrine as specified in Section 107 of the Copyright Act.
Which has since been updated a bit:
I’m not here to (re)litigate the issue. Using a model that had to be taken to court, and which is settling so it’s not taken to court again (and again, and again) does not speak, to me, as being steeped in ethics.
I’ll keep my works away from LLMs, and I’ll ask that they do the same, with copyright.
5 Likes
That’s a district court, though, not an appeals court. A different district court judge, in Kadrey v. Meta, made exactly the opposite ruling. Fair Use and AI Training: Two Recent Decisions Highlight the Complexity of This Issue | Insights | Skadden, Arps, Slate, Meagher & Flom LLP
(For readers outside the US, the district court is the first level of the federal system. The appeals courts – the Ninth Circuit in this case – and the Supreme Court can both overturn the district courts. Only the Supreme Court can establish binding precedents for the whole country.)
Also, no court has yet addressed whether the AI output itself is copyrightable. The US Copyright Office has so far held that it is not. Judge Alsup ruled in the Anthropic case that training is fair use, but explicitly did not rule on copyright questions surrounding the output.
4 Likes
Quite correct. Among other courts, I’m admitted to practice before the U.S. District and Ninth Circuit Court of Appeals — though I’ve retired from practicing law and am on inactive status these days. Moreover, I used to teach communications law, which of course includes copyright. I find it fascinating to watch the courts grapple with the intellectual property aspects of AI. I am enjoying this running discussion from the point of view of we writers.
As an undergrad in the late seventies, I wrote a paper postulating the use of quantum bits as a basis for a computing device that could leverage superposition to solve super complex problems. (My professor thought it was an idea bordering on fiction, though only a few years later this idea emerged as a genuine field of inquiry. Ha!)
Later, I was fascinated, purely as an amateur because I was now a lawyer, to follow the research into machines that can mimic human intelligence. I remember idly making notes (now in the eighties) about combining quantum mechanics with layered nodes to mimic neurons (an idea I had read about), but with the capacity to consider all neural pathways simultaneously (an idea that I had). This thought so frightened me that I have had a hard time enjoying the idea of AI ever since. Imagine a thinking machine that can consider all possibilities at once, while we humans are limited to linear neural pathways. If we think AI is too disruptive now, well, we haven’t seen anything yet. When the bastards finally figure out how to combine quantum computing with classical AI (it’s funny calling AI classical, but that’s what it is), we are really screwed as living, thinking beings. And I’m an optimist by nature.
1 Like
It’s hard to say what will happen. Quantum computing may certainly help us invent faster mimics, which is what the LLMs are. They are still fairly garbage at reasoning, and don’t emulate physical reality. Generalized AI is really the robot companion/holodeck/evil planetary intelligence we imagine.
The models we have now are parrots that do math quickly (“this word follows that word after those words.”) A quantum model may answer more quickly, but the tech we have now doesn’t indicate that it will be any more thoughtful.
Cryptography better watch out, though. That depends on slow computation. But our squishy electric fat lumps we call brains are computationally slow and not quantum, and yet are pretty amazing. Speed isn’t what nature selected for, and yet random chances got us these wonderful portable devices in our heads. And we’re learning to let go of our sapient egos and appreciate intelligence in our animal kingdom kin. Maybe faster-and-in parallel isn’t the answer to making a thinking thing.
Good points.
I wasn’t thinking of LLMs for the reason you mention, but rather the neuromorphic approaches.
On the positive side, combining the relatively primitive LLM approach with quantum computing could reduce the massive amounts of electrical power required to a comparatively trivial level. So there’s that.
I think someone has already said somewhere that 1984 was a warning not a game plan but here we are.
4 Likes
Here are my predictions of the legal state of AI in five years:
- AI will continue to train on copyrighted materials;
- Owners of those materials will not be significantly compensated, with some notable exceptions like Disney;
- Infringing output (like making Deadpool artwork) will be ruled to be copyright infringement, but…
- The major AI providers will successfully claim “common carrier” status like Facebook has, so the infringement is the liability of the user, not the service provider;
- AI will become “good enough” at writing, illustrating, photography, music production, audio and video production that all industries based on those media will undergo a dramatic loss of jobs.
- A smaller number of artists who use AI as a tool in those industries will flourish;
- All of those industries will experience an avalanche of automatically-produced, meme-inspired low-cost, “junk” commercial products that will make it much more difficult for human artists to stand out and make a living in the domain.
Anyone think I’m wrong on al counts? What are your predictions?
1 Like
I don’t think the AI industry has five years before it collapses under the combined weight of huge investments, relatively paltry revenue, and environmental impacts. In particular, the paradigm of “make the model bigger for better results” is already hitting diminishing returns.
If AI-generated content is not copyrightable and/or is infringing, then there is no commercial market for AI products.
3 Likes
I don’t think the AI industry has five years before it collapses under the combined weight of huge investments, relatively paltry revenue, and environmental impacts. In particular, the paradigm of “make the model bigger for better results” is already hitting diminishing returns.
I will take that bet! I think in five years, AI will be thriving. For two reasons: (1) Nvidia’s $100 Billion investment in OpenAI will be in full swing, which will prompt China and other competing interests to continue vying for dominance in AI over OpenAI, and (2) the fact that open-source developers are becomeing more and more successful at optimizing AI models to run on less and less compute.
If AI-generated content is not copyrightable and/or is infringing, then there is no commercial market for AI products.
That’s absolutely not true. Having content that has no copyright prevents no-one from selling or buying the content. Companies sell public domain content all the time. The only thing a lack of copyright prevents is exclusivity. However, so much of the future automated, meme-driven AI content will be designed to have a lifespan of weeks, that having knockoffs hit the maket won’t impact the originators. They will have moved onto the next one.
Also, I fully expect that a great deal of this automated content (commonly referred to as the “tsunami of crap”) will be itself infringing, and will be distributed from servers in copyright-unfriendly countries like China and Russia. You’ll be able to see a feature film of Deadpool fighting the Care Bears, without the permission of any of the rightsholders. The crapmeisters won’t care a whit about the copyright status of AI.
It’s only a matter of time before Big Tech starts spending serious lobbying money to get the USCO to change its mind, so that studios can use AI and keep the copyright.
I hate it, but suspect you are right on all of that. I hate that AI has become such a large part of my day job over the last 18 months, too, and I don’t see that ever changing now.
1 Like
The open source models don’t have access to the giant training sets, though. I question whether they’ll ever be “better” than the big models for the kinds of tasks we’re talking about.
As for Nvidia’s investment, they’re effectively subsidizing the purchase of their own chips. I don’t see how that’s a sustainable model for either company.
Edit to add: Thinking more about this, it’s not unusual for a customer to invest in a vendor. That’s how Boeing’s biggest customers get priority access to new airplanes and how ASML’s semiconductor lithography tools got built. But here, you have the single largest AI compute hardware vendor investing in the single largest AI compute hardware customer. So OpenAI is getting chips – a tangible physical object with value on the open market – in exchange for stock in a non-public company with a weird governance structure, minute revenue, and serious obstacles to ever going public. (And that has no choice but to use those chips because there is no real competition.) And Nvidia’s board signed off on this?
3 Likes
It doesn’t have to be sustainable - it just has to last for five years for your prediction to be wrong.
You’re looking at it wrong – they’re not just investing in their customer, they’re investing in the AI economy. They are guaranteeing that either OpenAI will completely dominate the AI market with Nvidia as a significant shareholder, or that other companies will strive to compete with OpenAI, and buy even more GPUs from Nvidia. Either way, Nvidia wins.
1 Like