I've also compared Claude 3.5 Sonnet with all the other LLMs and find it works best for developing screenplays. I've been especially impressed by how well it can improve dialog using a couple prompt tricks. First, I created documents summarizing the most important points from Robert McKee's book, Dialog, and added this to the knowledge base. Second, I create a system prompt that specifies, "You are a god-tier screenplay writer with special skill at creating emotional and musical dialog at the level of Paddy Chayefsky, Robert Towne, Billy Wilder, Christopher Nolan, and Aaron Sorkin." By adding these names to the prompt, the LLM zooms in on the associated concept space within the model, raising the probability of certain word choices and associations. When I have Claude rewrite existing scenes with alternative dialog, it frequently produces higher-quality dialog that can be cherry-picked and added to the script. It is important to also develop detailed character descriptions and backstories for the characters and add these to the project knowledge base to further narrow the space of word choices made by the LLM.
hi sir, can you please share the summarized document and what are the command prompts you use, so that i can convert them to my language, i do not write in english
I’ve always preferred Claude for writing; in fact, I’m currently drafting a novel with the platform. However, I recently started using the paid version and was disappointed that it still throttles your output at some point.
when it starts to warn that you're reaching limit, I ask it to generate a comprehensive prompt summarising the entire conversation and use it to start a new thread. I usually put the content it has generated so far into the project knowledge base. Works okay for me that way.
I would love to see what they both do with your feedback. Claude is really good at iterating and improving on feedback (from my experience on other areas)
Thanks for another great video. Why not just use the paid version of Perplexity in writing mode? You get access to both GPT4o and Claude3.5 sonnet plus a bunch of other models. And then, as you mentioned, you have the internet capabilities that are really great, with good source referencing.
Thank you for great videos. I am a beginner at book writing and want to understand why use software such as novelcrafter when it seems Claude 3 does the writing quite well enough. Do you have any videos that answer this very simplistic question? Thank you
I do prefer Claude. However, I don't like that they not only limit the free version, but also the paid version has a limit. I believe it is around 20 prompts per 6 hours, but it doesn't seem to be consistent.
It’s a minimum of 20 prompts. I get only 20 if I’m using a beefy model and a beefy prompt, like transcribing an image or including a whole books worth of text in the context window. Most of the time I get WAY more.
@@TheNerdyNovelist Ah, that's probably it! I made a "project" with a draft of my book as project knowledge. That's when I encountered that limit. Thanks for clarifying!
Is there a way to increase my limit of my pro claude account? I've been told I might have to use the API for unlimited pay-as-you go... i'm tired of having to wait 5 hours. I have a lot of context with my prompts, so it eats up a lot... any advice on how to use the API for writing assistants?
Hi Jason, I decided to re-write the previous comment from this morning. I started reading your new book I got in Amazon, I had to pause reading it for days after your introduction on chapter about your journey. That mental block you had and overcame was just nerves. Glad to know you moved past it. I'll resume reading it after I finish my current Kindle book, Alejo Carpentier's The Age of Enlightenment (error in here edited out) and your book will be next!
@@alvarobaltaxar9128 OMG, you are right! I think I got that confusion off my Amazon recommendations when looking for Carpentier's books it added in other Latin-American writers like Garcia Marquez' and Puig's 'Boquitas Pintadas' which I'm getting too. Thanks @alvarobaltaxar9128, I edited out the author error above. I read Carpentier's El Reino de Este Mundo when I was around 11 years old in 1972 and Puig's Boquitas Pintadas when I was like 18 or 19 years old and saw the play in NYC with Chita Rivera as lead. When I found these old books of my youth in Amazon a month ago, I jumped off my seat.
Oh wow, nice. I'm from Argentina, so that's my advantage! We are drilled in school with these books. Btw, every single book from Puig is an absolute gem. Enjoy!
Just found your channel yesterday and I'm really excited! I've been trying to figure out some tools to make my writing more efficient (and honestly more fun) and at the same time mix in some of the intermedial level AI work I've been doing. I do have a question though. Which do you think is a better fit, a fine tuned OpenAI model or Claude 3.5? I have a project brewing that I want to test the AI out on.
Can you do a video where you compare the quick fine tuned versions. You did a video where you gave Claude context to do a quick fine tune. Can you do that for both GPT 4.0 & new Claude 3.5. Thanks
Started to look at Claude a bit more, I prefer the Claude prose, Chatgpt can be a little to « emphatic » or talkative or repeating itself more. But Claude has a lot of limitations in the number of requests per day I think…
im always telling chatgpt to quit yapping and just give me exactly the info im asking for. instead it wanted to re-iterate half the damn project. you have to say "keep it brief" in every single prompt if you want it to quit yapping.
@@chrishayes5755 Yes, I also ask Claude to be "concise", it seems to do the trick more easily than with ChatGPT, I will try the "keep it brief" approach, thanks !
How do you keep claude from like foreshadowing at the end of every prompt? Or like "gather round kids, we're summarizing all the last prompt!" It's so annoying.
OPenAI is dead to the consumer market. They only care about government contracts now. They do a ton of hype and everyone else is speeding past them in actual released product.
What a pointless comment. What claude does is a first draft. I've seen human made first drafts that are in far worse condition than a claude written novel. It's not the end point, it's the start.
your audio is noticably more 'muddy' than most other channels I watch. sounds sorta like your voice is being projected through your chest rather than your mouth. not sure if it's your mic or something with post-processing, but there's no upper range or it's too bassy
How do you keep claude from like foreshadowing at the end of every prompt? Or like "gather round kids, we're summarizing all the last prompt!" It's so annoying.
Usually if I have a complex prompt I will just ask it not to and that helps. If it still is there I will just chop it off and move on to the next section.
I've also compared Claude 3.5 Sonnet with all the other LLMs and find it works best for developing screenplays. I've been especially impressed by how well it can improve dialog using a couple prompt tricks. First, I created documents summarizing the most important points from Robert McKee's book, Dialog, and added this to the knowledge base. Second, I create a system prompt that specifies, "You are a god-tier screenplay writer with special skill at creating emotional and musical dialog at the level of Paddy Chayefsky, Robert Towne, Billy Wilder, Christopher Nolan, and Aaron Sorkin." By adding these names to the prompt, the LLM zooms in on the associated concept space within the model, raising the probability of certain word choices and associations. When I have Claude rewrite existing scenes with alternative dialog, it frequently produces higher-quality dialog that can be cherry-picked and added to the script. It is important to also develop detailed character descriptions and backstories for the characters and add these to the project knowledge base to further narrow the space of word choices made by the LLM.
Great tip. Thank you
hi sir, can you please share the summarized document and what are the command prompts you use, so that i can convert them to my language, i do not write in english
Precious, thanks!
I’ve always preferred Claude for writing; in fact, I’m currently drafting a novel with the platform. However, I recently started using the paid version and was disappointed that it still throttles your output at some point.
Yh the limitations on Claude Pro are pretty ridiculous. Very frustrating
when it starts to warn that you're reaching limit, I ask it to generate a comprehensive prompt summarising the entire conversation and use it to start a new thread. I usually put the content it has generated so far into the project knowledge base. Works okay for me that way.
If writing is subjective; how do you conclude what is correct and what is just a writing style. What benchmark do you follow?
I would love to see what they both do with your feedback. Claude is really good at iterating and improving on feedback (from my experience on other areas)
Same
Thanks for another great video. Why not just use the paid version of Perplexity in writing mode? You get access to both GPT4o and Claude3.5 sonnet plus a bunch of other models. And then, as you mentioned, you have the internet capabilities that are really great, with good source referencing.
Which one is better at academic writing?
Thank you for great videos. I am a beginner at book writing and want to understand why use software such as novelcrafter when it seems Claude 3 does the writing quite well enough. Do you have any videos that answer this very simplistic question? Thank you
To avoid repetitive scenes and organize character
I do prefer Claude. However, I don't like that they not only limit the free version, but also the paid version has a limit. I believe it is around 20 prompts per 6 hours, but it doesn't seem to be consistent.
It’s a minimum of 20 prompts. I get only 20 if I’m using a beefy model and a beefy prompt, like transcribing an image or including a whole books worth of text in the context window. Most of the time I get WAY more.
@@TheNerdyNovelist Ah, that's probably it! I made a "project" with a draft of my book as project knowledge. That's when I encountered that limit. Thanks for clarifying!
Agreed same here I have the paid version and I hate the limitations
You're right. I am a pro user but I always run out of limits. Although I love it, I am thinking about switching back to GPT Plus.
Really interesting videos, thanks for all your work! And not so nerdy I would say !
For your support
Is there a way to increase my limit of my pro claude account? I've been told I might have to use the API for unlimited pay-as-you go... i'm tired of having to wait 5 hours. I have a lot of context with my prompts, so it eats up a lot... any advice on how to use the API for writing assistants?
The API or OpenRouter are your best bet.
Excellent quality video. Exactly what I wanted to know at the perfect level of depth. Love your channel. Thank You!
Glad it was helpful!
Hi Jason, I decided to re-write the previous comment from this morning. I started reading your new book I got in Amazon, I had to pause reading it for days after your introduction on chapter about your journey. That mental block you had and overcame was just nerves. Glad to know you moved past it.
I'll resume reading it after I finish my current Kindle book, Alejo Carpentier's The Age of Enlightenment (error in here edited out) and your book will be next!
The Kiss of the Spiderwoman, that's Manuel Puig's book!
@@alvarobaltaxar9128 OMG, you are right! I think I got that confusion off my Amazon recommendations when looking for Carpentier's books it added in other Latin-American writers like Garcia Marquez' and Puig's 'Boquitas Pintadas' which I'm getting too. Thanks @alvarobaltaxar9128, I edited out the author error above. I read Carpentier's El Reino de Este Mundo when I was around 11 years old in 1972 and Puig's Boquitas Pintadas when I was like 18 or 19 years old and saw the play in NYC with Chita Rivera as lead. When I found these old books of my youth in Amazon a month ago, I jumped off my seat.
Oh wow, nice. I'm from Argentina, so that's my advantage! We are drilled in school with these books. Btw, every single book from Puig is an absolute gem. Enjoy!
Just found your channel yesterday and I'm really excited! I've been trying to figure out some tools to make my writing more efficient (and honestly more fun) and at the same time mix in some of the intermedial level AI work I've been doing.
I do have a question though. Which do you think is a better fit, a fine tuned OpenAI model or Claude 3.5? I have a project brewing that I want to test the AI out on.
For writing, Claude.
Claude. It mimics style pretty well and follows directions way more than a fine tune. At least for now.
Great video! thank you
Could you do more videos on nonfiction? And non fiction tools?
Can you do a video where you compare the quick fine tuned versions. You did a video where you gave Claude context to do a quick fine tune. Can you do that for both GPT 4.0 & new Claude 3.5. Thanks
Can you do a perplexity vs Gemini on all things that have to do with the Internet? 🙏🏼
Started to look at Claude a bit more, I prefer the Claude prose, Chatgpt can be a little to « emphatic » or talkative or repeating itself more. But Claude has a lot of limitations in the number of requests per day I think…
If you use it via openrouter you pay per use, no limitations at all, and it's probably cheaper anyways.
im always telling chatgpt to quit yapping and just give me exactly the info im asking for. instead it wanted to re-iterate half the damn project. you have to say "keep it brief" in every single prompt if you want it to quit yapping.
@@chrishayes5755 Yes, I also ask Claude to be "concise", it seems to do the trick more easily than with ChatGPT, I will try the "keep it brief" approach, thanks !
@@Pyriold good tip, thanks
How do you keep claude from like foreshadowing at the end of every prompt? Or like "gather round kids, we're summarizing all the last prompt!" It's so annoying.
Have you tried Mistral Large 2 and Llama-3.1 405B?
Yes. My last live stream was about them.
@@TheNerdyNovelist I've been really enjoying running Mistral Large locally for writing.
I like claude opus for writing
Just on a side note. I was shocked how much gpt 4o CRUSHES Sonnet 3.5 in math
Based on the findings out there I’m not surprised.
@@TheNerdyNovelist I did some tests with really tough exercises from math olympiades. Sonnet had no clue. 4o just solved it.
Is claude 2.0 and 2.1 still available? Thanks
why would anyone need these
for writing prose, AI, in general, still has a long way to go.
Not as much as you’d think. As long as your prompt is good.
I prefer Claude opus for writing
yeah but claude doesnt like my stories, I am writing about warfare and chatGPT can be bypassed whereas claude gets mad at a papercut
Best paid for puzzles
I love u
First!
OPenAI is dead to the consumer market. They only care about government contracts now. They do a ton of hype and everyone else is speeding past them in actual released product.
Claude's...virginity...is a problem.
chatgtp is always better ...
No it is not.
Both suck. read more, write more.
What a pointless comment. What claude does is a first draft. I've seen human made first drafts that are in far worse condition than a claude written novel. It's not the end point, it's the start.
No they don't. AI is good and it's here to stay and it's only getting better and you hate that hahaha. Get ready for Opus 3.5, a new level begins.
if you're right you'll outshine the competition. objective reality will let you know.
your audio is noticably more 'muddy' than most other channels I watch. sounds sorta like your voice is being projected through your chest rather than your mouth. not sure if it's your mic or something with post-processing, but there's no upper range or it's too bassy
How do you keep claude from like foreshadowing at the end of every prompt? Or like "gather round kids, we're summarizing all the last prompt!" It's so annoying.
Usually if I have a complex prompt I will just ask it not to and that helps. If it still is there I will just chop it off and move on to the next section.