I completely agree. Indeed, I almost wrote of my own possible confirmation bias (should have kept it in). Nevertheless, irrespective of the source (I, too, am sceptical), I arrived at many of the same conclusions before seeing the video, though I strongly suspect the timeline suggested is too aggressive.
Just to be clear, a video such as this is not evidence that my thinking is right; it’s just that I found it to be presenting the argument in harmony with my own thinking.
I know this thread has gone quiet, but the question on whether L&L were considering adding AI to Scrivener is something I have been wondering about too, recently.
I am currently using the huihui-qwen3.5-27b-claude-4.6-opus-abliterated LLM with LM Studio, and as I write this post, the AI LLM is writing the synopses for the 3 parts, 23 chapters, along with their sections and subsections of my non-fiction book. It is pretty impressive, I have to say.
I think it is worth pointing out that AI has moved forwards in leaps and bounds since the first post back in March 2023.
For what it is worth. Choosing the right LLM, learning how the specific LLM works (thinks), and devising the right strategies for working with the LLM effectively has taken some time (partly because I was an AI novice when I started this).
However, I think that there are some valid concerns raised with respect to the quality of the output and the term it uses sometimes, hence why I am only using it for planning. But when it comes to building and pulling ideas together it is and I say it again - impressive.
Thinks is doing a lot of work here.
It’s not just about the quality. It’s also about surrendering a part of the creative process in a way that no longer makes it yours. Maybe that’s just me. Writing is the process, in its entirity, not the outcome or finished product. It’s a lot of work.
I don’t think our position has changed:
The LLM I use lets you see its ‘thinking’ and the final output does not always reflect the LLM’s thinking as it directs the output towards the User request.
Your second point is valid if you over depend on AI, which is I why I am only using AI for planning. However, it does greatly reduce the planning workload and helps keep track of the large number of interrelated topic threads that I am writing about. I disagree that it is ‘surrendering a part of the creative process that no longer makes it yours’ though, in the same way that you are not surrendering the creative process to an editor when you work with them.
Surely, if AI becomes an available option, people who strongly object can simply not use the AI feature. However, it is a little bit on the authoritarian side of the fence to suggest that those who object to AI should ultimately decide for those who want to use AI. Don’t you think?
Modern computers are pretty good at multitasking. As you yourself have discovered, LLM-based tools are readily available for anyone who wants them.
Given that LLMs are built on stolen data, trained by some of the poorest people in the world, destroy the environment as a matter of course, and overwhelmingly benefit billionaires who openly despise the humans they seek to replace, it’s pretty funny to describe opposition to LLMs as “authoritarian.”
In any case, L&L is not and has never been a democracy. The idea that people who don’t like it can “simply not use” a feature ignores the development and support resources that even unpopular features require.
AI is best used as an auxiliary tool, such as telling me about conflicts and duplicate content between different index cards, prompting me to revise or merge them, and replacing my previous review work, rather than writing for me.
In other words, integrating AI, but only allowing AI to see titles and synopses, is the extent to which I can accept.
I didn’t describe those opposed to LLMs as “authoritarian.” I described your reasoning as a authoritarian. If you are a developer and don’t want to do it, or can’t do it, then say so, but don’t use other people’s objections as an excuse.
As for your Socialist rant:
”Given that LLMs are built on stolen data, trained by some of the poorest people in the world, destroy the environment as a matter of course, and overwhelmingly benefit billionaires who openly despise the humans they seek to replace”
This is untrue as LLM’s have to respect copyright laws and training equipment is expensive, but it makes me ask, why are you even using a computer or any other piece of technology for that matter? Yet here you are posting on the internet using computers and other technologies that ‘benefit billionaires who openly despise the humans they seek to replace’.
I agree. In other words using AI as an assistant and editor, but not to use it as a ghostwriter.
Anthropic, the parent company of Claude, recently settled a lawsuit over their failure to respect copyright. Similar lawsuits against other companies are pending.
So many logical fallacies it was hard to pick just one!
Ah. Got it.
These are literally proven and observable facts and currently standard practice within the industry. They don’t even try to hide it. It’s part of the brochure.
Your statement is inaccurate.
The Court has ruled in the Anthropic case that training an LLM using copyrighted works is legal under Fair Use.
Anthropic is settling the case of it downloading copies of copyrighted works without paying for them. The infringement was committed by downloading the books, not using them to train an LLM.
Other companies use licensed and public domain materials to train their LLMs, which is entirely legal.
QED, LLMs are not built on stolen data.
It would be nice if L&L would either build or facilitate the building of an MCP for Scrivener. This would allow AI Agents to read, manipulate and write to Scrivener projects. There is an MCP for Obsidian, and there are some very useful Claude/Obsidian integrations.
Having nothing to do with AI-written prose, it can be very helpful in allowing the AI to search, summarize and cross-reference data in a Scrivener project.
That’s not what the relevant copyright owners are saying.
Lawsuit vs. OpenAI: OpenAI Copyright Lawsuit Explained: Why Authors and Media Companies Are Suing AI Firms
Lawsuit vs. Meta: https://allaboutlawyer.com/meta-llama-publisher-copyright-class-action-ai-training/
Lawsuit vs. Alphabet: Publishers Move to Join Copyright Lawsuit Over Google’s Gemini AI Product - Publishing Perspectives
Anybody can sue anyone over anything. The interests in favor of LLMs will make it all go away in a few years. They will either be settled for cash, ruled Fair Use, or someone will pass a law giving them legal cover. At the end of the day, AI will keep growing and they will continue to train their models.
“AI companies will buy a change in the law” is rather a different statement from “LLMs are not built on stolen data.”
Not to mention if an LLM is trained on your work & someone else uses it to do a copy-paste replica of your style… These class actions are worthless. The VAST sums of money involved, the author potentially gets f-all. It’s not like these machines can ever be “un-trained”.
If only we lived in a world where these AI companies were required to destroy LLMs trained on stolen content, eh? Anywhere else, if you get caught with someone else’s stuff, it gets taken off you.
Just wanted to pop in and express my happiness that L&L has refrained from jamming so-called AI into Scrivener. Applications have been becoming unusable right and left (never mind the ethical considerations about power and environment and labor and plagiarism) and it’s so nice to be able to use something that just works.
You misunderstood what I wrote.
LLMs are not built on stolen data. The Court ruled that using copyrighted works to train an LLM is legal. The settlement Anthropic is negotiating is for infringing on the copyright of books it downloaded from the internet without a license, instead of buying them. It has nothing to do with AI at all. It’s simple copyright infringement.
Also, because anyone can file suit over anything, the beginning of litigation doesn’t prove anything except that there’s a difference of opinon.
I’m saying that the Court will probably rule against these cases because there is already a Fair Use precedent from the Anthropic case. If it looks like the ruling will go against the AI companies, then they will settle, but not give up any training practices. If neither course is available, the interested parties will simply pass a law or an Executive Order that will permit them to continue as they have been.
The US Government has publicly stated that the unfettered development of AI in the USA is an isue of National Security. I don’t think the power behind Washington is going to allow any slowdown to the AI juggernaut.