I guess this kind of discussion will probably be an ongoing discussion (or more likely: on- and “off-going”) for a long time. I am more on the pro-AI side. I see new online writing tools with AI features every couple of days and at some time there might one emerge that is really good. A lot of younger people might prefer this kind of writing and that could be a long-term problem for Scrivener then (I hope not.)
There a a lot of features in Scrivener I don’t use or need. Some AI integration won’t disturb anyone who prefers to write without using such tools.
Then an API access to different AI-APIs would not be much more complicated than adding just one. The task you can do with ChatGPT would be very much the same you could do with other language models.
A basic set of features I’d like to see:
a set of predefined prompts for any selected text, accessible through context menu
prompt should be customized: the prompts itself and adding to these prompt and the context actions
a prompt library (with kind of a revision mechanism for each prompt or at least a chronological, searchable archive), with option to just paste a prompt or to directly execute
a dedicated page in the Inspector
a special chat window (on the same logical level as an editor window) to chat with whatever AI you like (which by the way would be kind of an HTML page or an embedded browser), where you can select text and add it to a prompt library or just move the text as from any editor window to any other place
setting the model one would like to use per AI-API (like 3.5 or 4 for ChatGPT, as this would affect the individual costs)
An excellent Chrome/Firefox addon is Superpower ChatGPT, which shows the potential.
Probably that would do the trick for most things I’d like to do.
They would be flexible, they would be optional, and they would not be product-specific.
And no-one should underestimate the power of such language models. The difference between the free GPT-3.5 and the (at the moment) subscription-only GPT-4 is immense.
You can harness their power and can get much more than “just stereotypes”. The creativity lies in your own means to utilize the tools by prompting, exploring, and of course: working with any input the machine (which it still is) gives you .
I don’t speak for Keith, but I would say that this level of integration is unlikely as out-of-scope. Scrivener is a tool for writing, not an AI front-end.
I would also say that someone who thinks this sort of integration would be easy has probably never tried to implement it.
I suppose it could be faster. I don’t know how Scrivener performs under the hood.
If all this can be done protecting rights, etc., then I think this bot could be a great tool. Other tech (such as brain tech) has become downright Orwellian, so my enthusiasm is dampened.
I obviously speak for myself, but I would be very surprised if Keith were not working on an integration between Scrivener and AI. I am currently experimenting with a couple of AI tools. I have a paid subscription to POE, which offers access to several AI platforms (Sage, GPT4, Claude+, Claude-instant, ChatGPT, Dragonfly, NeevaAI) each with its own capabilities. This has enabled me, first, to overcome the idiotic block imposed on ChatGPT by an Italian authority (then why against ChatGPT alone?) and, second, to experiment with different platforms. I also have a paid subscription to Sudowrite, which is one of the most popular writing applications using AI. I find Sudowrite’s writing support still rather basic, but I have grown to appreciate some of its features, e.g. summarizing a plot or brainstorming. it still does a pretty inconsistent job with languages other than English, most of the time giving a mixed output. While I have found ChatGPT of very limited use and reliability, so far, when conducting searches, I find that it can become a useful tool when you can feed a lot of text and information into it, and then ask some very specific questions.
I also believe that the inevitable integration of Scrivener with AI would require to move away from a one-time purchase to a subscription based service.
Well. At least now you’re already prepared to be very surprised. (Does this still count as a surprise if one is expecting it? )
I think it’s more likely that we’ll witness several other events before that happens, like: The launch of Scrivener for Android. Or a Linux version. Scrivener for watchOS. Ditching RTF in favor of Markdown. Free beer for everyone. World peace. And of course:
While you’re at it: Please try asking them to use paragraphs.
I am afraid I miss your point. Have you made any experience with Sudowrite by any chance? In spite of the mentioned shortcomings, some of its features are awesome (see Rephrasing, for example); other ones are just glitter, but all are worth experimenting with
As I indicated above, I would expect to see, at most, the same sort of integration that we offer for bibliography tools. I would very much not expect to see either close integration with a specific machine learning platform, or integration that increases the base cost of Scrivener.
If you sell something, depending on where you live, you may have to disclose that it is partially written by AI if you get found out, so I really wouldn’t. If you wrote the whole book, just do a summary. You got this far already.
)