There is absolutely nothing that this man might have to say that is worth 90 minutes of my life. What’s the TL;DR summary?
Use AI as a writing assistant, not writer
Does it mean I’ve to have Claude pro? I already have ChatGPT pro.
You don’t have to have Claude Pro.
The API is separate. You can set rate limits, auto refill, etc.
With Claude Pro Max, the API will use the Max subscription.
But the API can be used without any other Claude subscription.
The tool currently uses Claude 3 Haiku, 3.5 Haiku, and 3.5 Sonnet.
All are very cost effective and very good models for writing, especially creative writing.
I tried it. It seems to be a standalone app (which is good), but it also needs to integrate with Scrivener.
Thank you for the feedback!
It integrates with it in a sense that it can retrieve text that is selected inside of Scrivener and can paste anywhere the cursor is placed or anywhere text is selected.
I’m testing out where it’s a menu selection, but being Scrivener is closed source, it is a lot of trial are error to get a more direct integration.
But I am trying to make it like a hovering option as well.
But currently, it does access any selected text or where you cursor is at.
One of the big stars of it, is it’s really good at keeping your story, edits, expansions, on track without wandering off without having to copy and paste it in a chat window, and then paste it back into Scrivener.
I will post some updates with my progress on it as well with the deeper integration.
Thanks!
I’ve had my first brush with AI; to be specific, my friend and collaborator sent me a couple of transcriptions by ChatGPT Iinto both graphemes and IPA for a couple of phrases in Minnanhua (Hokkienese). Absolute gibberish. Amongst a multitude of errors, it transcribed the same morpheme in two completely different and unrelated ways (/e/ vs /to/), and made significant errors in the IPA versions.
Almost laughable!
Mark
It depends on what model you are using, what prompt did you give it, etc. I have learned that the more details you give it, the better output you get. Also, it’s important to remember that it can sometimes hallucinate.
I experimented too.
I wanted to know what I was up against as regard to “writing” a novel.
I tried Bing-Copilot.
Gave it for prompt a cop + a very vague situation and ending made up on the fly.
I repeated twice the day after : each time a cop + a different situation and ending.
Each time Copilot spitted out an outline, and each time I told it to just skip that and write.
No surprise (in French): all three times it sucked unaccountably.
But what did strike me the most is that, despite the thing unable to remember anything from a session to the next, it gave all three of my cops the same name. (Morel, each time.)
Originality = 0
. . . . . . . .
For kicks I just tried again. This time my cop’s name came out as “Moreau”.
(About a month since the first three trials.)
Wow! We’re getting somewhere.
If you are replyting to me @aithal, I didn’t give nit any prompt. My collaborator had given it two strings to transcribe in Minnanhua. She sent them to me to ask if I thought they were right. In the transcription into graphemic (standard transcription) and IPA there were many glaring errors and inconsistencies.
I’m not interested personally in using it… I have no reason to need it… but this encounter doesn’t inspire me in any way.
Mark
In more rigorous disciplines, they call this a leading question for a reason.
That’s a very interesting take, thanks @popcornflix.
Is Paul any good? Given that he’s reached this conclusion and I REALLY want AI to fail, I’m kind of hoping he’s coming to this conclusion because he simply doesn’t have it anymore. (Not a great motive, I know, and I’m not proud of it but hey… honesty).
I’ve seen he gets made, which is in his favour of course, but the last thing he made that I’ve actually heard of was in 1988 (and the last one I enjoyed was 1986).
Okay, I’ve done some reading up on Paul. Sounds like I disagree with him on several things and he might not be someone to deputise into an argument.
Something else to be aware of is that the AI companies have a lot of money to throw around. When you’re spending billions on data centers, “sponsoring” people who might endorse your products is pretty cheap. Not casting any aspersions on Paul Schrader specifically, but I have seen a lot of commentary that ultimately traces back to an interview with Sam Altman or something like that.
Paul Schrader wrote Taxi Driver and Raging Bull for Martin Scorcese. So yeah, he knows a thing or two about writing screenplays.
I disagree with him on a lot of things, too. The important thing here is that he’s one of the acknowledged masters of screenwriting, and he considers that the ideas coming from AI for the kinds of movies he makes are as good or better than what he comes up with himself.
I don’t hold that out as definitive proof, but as a trend.
I’ve found that the kind of movies that I write are more plotty than Schrader’s, and LLM’s haven’t been very good at generating those kinds of ideas, especially where irony is involved.
My position on IA is that it’s here to stay, and it’s getting more powerful every day. People who lack the discipline and craft to write will be using LLMs to compete with experienced writers like us. I think it’s important to learn how to use these tools to our own advantage.
Saying that they will never be good enough, lacks soul, aren’t human enough, blah,blah,blah – it’s just wishing that it will all go away. As writers, we need to deal with the world as it is, not as we wish it were.
Am I using LLMs to write my pages? Nope. Am I experimenting with using it as a writing assistant and brianstorming tool? You bet. You’d be a fool not to. it’s like insisting on writing longhand with a quill on vellum instead of using Scrivener.
Schrader is famous for putting his principles ahead of his own self-interest. It is laughable to think that he was paid to make this statement, Also, there’s video of him talking about it online. His deflation and despair of being replaced by an LLM is sincere and palpable.
Couldn’t agree more. Sticking one’s head in the sand, or saying it’s cheating to use AI, is naive.
It’s here to stay, it’s becoming more powerful, and indeed, it will replace bad authors, in the same way email replaced carrier pigeons.
Or, he did in 1976 and 1980 respectively. So over 45+ years ago. As I said, I’ve not heard of any of the things he’s done since 1988 (and I pay attention to the movies).
I guess what I’m trying to gauge is whether that’s because AI is getting great, or if it’s because Paul hasn’t had a good idea since 1986. (That’s not to say he hasn’t compensated for that with world class execution)
Allegedly sexually assaulting his assistant (50 years his junior), agreeing a settlement and then reneging on that settlement. Those principles?