It requires some sense of humour.
5 Likes
Your requests are getting tougher, but I’ll try my best.
How about:
OH MY WORD! Have you heard?!! No-one is allowed to talk about the new app yet!!! That’s sooooo craayzee!!!
1 Like
This is a worthy goal. “Eschew the Bun!”
1 Like
@pseingalt you will have to forgive @pigfender as they are our resident pedant.
You may find these links of use: [one], [two], [three], and [four].
No need to thank me, just helping humanity is enough.
They are, however, not the only pedantic person here however, so they generally tend to stimulate tangents outside the bounds of the original intent.
1 Like
I’m a bit mesmerized here.
I don’t know if you meant your post as some humour too, @Dain, but if it wasn’t intended so, can someone please explain to me how
isn’t in itself pedant.
1 Like
il n’y a pas de hors-texte
Yes.
A pedant is a wielder of pedantry and is overly pedantic.[1]
In context with the links from the source post, the humor should appear as intended.
Just like this explanation. ↩︎
1 Like
Je suis d’accord. La blague réside dans l’absence de données.[1]
Polyglot I am not, Google Translate for the gag. ↩︎
Topic split since it is no longer even tangentially about the new app.
5 Likes
Wait, there was a topic? This was almost like the olden days, when Scrivener was still a bouncing baby Beta and we were all pedants on this bus.
2 Likes
Curious 
Did this article also mention that Electron apps are notorious for poor performance and hogging resources?
Electron apps are for devs who want to hit many platforms for as little outlay as possible. They’re not really for devs who put the user experience over cost.
Quite a bit to unpack here.
The problem with Markdown is that it’s pretty lightweight for a markup language. It was never designed to handle much beyond basic formatting (bold, underline, lists, and basic tables) which is why there are so many non-compatible versions floating around.
RTF while being somewhat long in the tooth, does have two advantages over MD.
- It has a standard
- It is much richer than Markdown
And it’s text-based (though many people think it’s a binary format).
If L&L switched to MD for the backend text format then they would probably need to write a hell of lot of code to bring it up the same level that Scrivener enjoys with RTF.
Not sure it would be worth the effort just to natively support AI. I’d rather keep Scrivener’s ability to generate near-publishable quality output. If wanted to use AI then I’ll just copy the text into a Chatbot and say “Rewrite this so it’ll win the Booker Prize.”
1 Like
The article was linked to on Ycombinator and shouldn’t be that hard to find. The author, a developer, was familiar with the negative publicity around Electron apps and so decided to write a native MacOs app using Swift. According to him, he almost immediately ran into problems and so ended up turning to Electron/Rust. The new L&L app was threatened to appear a few months ago but it didn’t. I remember when L&L was working on an iOS app there were delays as well. My memory of those is that they were personnel-related but it was a while ago. I looked at Keith’s post and saw that L&L is planning on a multi-platform rollout. If native programs are planned, I wondered if the Ycombinator article detailing difficulties with Swift (or was it Rust?) might be relevant. Also, as a longtime Scrivener user, I’m simply interested in the project.
As to the Booker Prize, maybe not, but for another kind of writing, a kind that mixes strict formulae and some, albeit not a lot, of creativity, AI does shockingly well. I wrote a one paragraph factual statement involving an arrest where the police simply failed to give American Miranda warnings. I fed the paragraph into AI which then wrote the brief in minutes. Yes, AI forgot to add a Table of Authorities but everything else was there. AI also identified the correct appellate court, even though this was not identified in the factual statement. There were no hallucinated cases. The reason why these still occur in legal writing is because West/Lexis/Juris have not granted access to their database. I do not know if Google Scholar’s collection is complete. There is a defined list, a limited universe of cases and the citation is either included in that list–no hallucination–or not–hallucination. It’s a little more nuanced than that since an accurately cited case may nevertheless be inaccurately quoted. Point being, it may be a time before prize-winning novels can be generated, but as to legal writing, we are already there.
Regardless of the objective merits of such a solution, the horse has long since left the barn.
3 Likes
There seems to be a new case every week where a lawyer is being disciplined for hallucinated citations.
3 Likes
Did look for it, but couldn’t find it (weak Google Fu is the problem I think).
In any case there are a whole lot of reasons that he was having problems with using Swift from bugs in the Swift framework to inexperience with using the language. If he was already proficient in JavaScript then perhaps that was the better alternative than wrestling with Swift (Electron being a massive memory hog notwithstanding).
Yes, if I remember rightly, the problem was caused by the appearance of the Liquid Glass which broke stuff.
Yup, that’s how I remember it. I also remember that L&L went from a standing start to a beta release in a pretty short time frame.
Still no joy with that article. The closest match I found was an app that went from Electron to Rust/Tauri.
And yet …
I think part of the problem is use of the term “hallucination.” It tries to label the old “garbage in/garbage out” moniker as a human condition, which then makes people inclined to forget what is causing the problem is the fact that these LLMs aren’t really “artificial intelligence” at all. If they were really intelligent then I wouldn’t have to keep telling it that dogs don’t come with five legs.
1 Like
- Native all the way, until you need text | Artem Loenko
- I am aware of lawyers getting in trouble for not checking cites. First, let’s keep in mind that there is a defined universe of citations, starting with 1 Cranch. As far as I know, no one has simply put togetber a list of these. If you had access to West/Lexis, or even Juris/Flite, it wouldn’t be that hard to do. For now, Ask Google what it has and what it’s missing. If you wanted to run tests, pick Alaska or Hawaii because you’re starting in 1959. A case cited by AI is either on the list, or it isn’t. There will be problems of disambiguation, but these are readily solved. For example, there may well be a 19th century case styled Miranda v. Arizona–Miranda is a common name–but if the case isn’t from the second half of the 20th century, it’s not what you’re looking for. More subtle are errors where the case is correctly cited but doesn’t stand for the proposition claimed: “Since the Roaring Twenties, an individual arrested must be advised on his rights before any custodial interrogation. See, Miranda v. Arizona, * U.S.*.” Even where the cite is correct, the Supreme Court case that imposed that rule wasn’t decided until the 1960’s. The run of the mill hallucination involves citations to cases that simply don’t exist. This issue is easily solvable. Court clerks will reject an appellate brief that fails to meet formal requirements. It’s easy enough to check “the list” in addition to page and word count, typography, etc.
- The above relates to the USA only. Cases in the UK go back for millenia. The Magna Carta had to be consulted in the trial of Roger Casement, so this is not a merely theoretical problem.
- Scrivener is an excellent tool for writing appellate briefs. Templates are needed.
Some general (and some random) observations and comments:
- I can’t comment much on that other developer’s experiences, although I imagine that TextKit is not the best choice for a messaging app.
- For native iOS and macOS apps, Swift is absolutely the best choice, and is a wonderful language. I personally wouldn’t want to spend hours in an Electron app.
- Swift is not the same thing as SwiftUI, in the same way that Objective-C is not the same thing as AppKit. Swift is a language, whereas SwiftUI is an API for building apps and interfaces built on that language. Swift works just as well with AppKit and UIKit.
- As I understand it, SwiftUI is definitely lacking, especially when it comes to building Mac apps. It has a bunch of limitations and there’s a lot you cannot do with it, so that developers using it frequently have to fall back on AppKit or UIKit. Our Scapple developer uses a lot of SwiftUI, but even for that, an app with a relatively simple UI, he had to revert to using AppKit and UIKit for a whole host of things, and at one point considered abandoning SwiftUI altogether.
- Scrivener is mostly written in Objective-C (with some new code in Swift), using AppKit on macOS and UIKit on iOS.
- Our new app is mostly written in Swift, again using AppKit on macOS and UIKit on iOS (all non-UI code is shared between platforms).
- TextKit has its problems but is still overall a great framework. Many great native Markdown apps use it (e.g. Ulysses).
- TextKit 2, despite being introduced five years ago, is incomplete, limited, and not suitable for anything other than the most basic of text editors. It can’t yet handle tables or multiple pages, for instance. It’s as though Apple started on it and then abandoned it for other things (which is often TextKit’s fate - it’s not sexy enough for WWDC announcements, I guess).
- TextKit 1 is still great, but Apple has let some bugs creep in. It’s also best used with Objective-C, so all of our TextKit code remains in Objective-C, even in the new app.
- As much as I love Apple hardware and software, by the way, and respect many Apple engineers I have interacted with, and believe in native software, I’m no Apple apologist. The way they treat developers is frequently insulting, but I will reserve my rants to offline interactions accompanied by alcohol.
- As for the new app, I’ve been using it exclusively for my own writing for the past two years. The macOS and iOS versions were ready for release last year, but we were waiting on the Windows version. Then, last summer, Apple introduced Liquid Glass, which threw a spanner in the works. The macOS version of the new app was using quite a custom UI, meaning that to support macOS 26 it needed a serious rewrite. The update took months.
- Right now, the new app is again pretty much ready to go (on all three platforms), aside from translations and getting the site ready. However, before it’s released I really need to update Scrivener for iOS, which has been neglected for too long. So the plan is to release the new app later this year, with no current set release date.
- The new app is not a Markdown app.
All the best,
Keith
11 Likes
Now that you mention it, I have to… Anything we (the volunteers) can do to prepare for a certain workflow?
1 Like
And yet.
I guess my point is that (1) a tool that is sold to enable laziness will trip up lazy people, and (2) if you have to check all the cites manually anyway, that chews up a substantial part of the time you “saved” by using AI.
3 Likes
Good grief.
It’s worse than I thought.
Still with the sub-par support for tables. 
I take they used some other secret framework to build Pages then.