Claude 3.5 Sonnet Just Changed the AI Writing Game
HTML-код
- Опубликовано: 5 сен 2024
- In this live stream we're going to test Claude 3.5 Sonnet for writing purposes, including brainstorming
✅ FREE RESOURCES
Get a free AI course, my 40-chapter Plot Module, and my prompt library ▶ ▶ nerdynovelist....
✅ JOIN THE MEMBERSHIP
Join my membership and get a free fine-tuned model + 2 free books with your 14-day free trial ▶ ▶ nerdynovelist....
✅ MY TOP AI TOOLS
Novelcrafter (best all-in-one tool for fiction) ▶ ▶ nerdynovelist....
AutoCrit (best for editing) ▶ ▶ nerdynovelist....
Sudowrite (best for beginners) ▶ ▶ nerdynovelist....
Future Fiction Academy (for learning AI) ▶ ▶ nerdynovelist....
Midjourney (for art) ▶ ▶ www.midjourney...
ElevenLabs (for audio) ▶ ▶ nerdynovelist....
✅ OTHER TOP TOOLS FOR AUTHORS
Publisher Rocket (for research) ▶ ▶ nerdynovelist....
Atticus (for formatting) ▶ ▶ nerdynovelist....
ProWritingAid (for proofreading) ▶ ▶ nerdynovelist....
Miblart (for book covers) ▶ ▶ nerdynovelist....
Wondrium (my personal favorite resource) ▶ ▶ nerdynovelist....
✅ FIND ME
The Website - nerdynovelist.com
My Membership - storyhacker.ai
I used Claude Opus and Sonnet for my 98K manuscript. It told me it was a series of scenes, not a full novel. So, I revised. Later, I used it for a developmental edit and assessment. I revised again. Now it’s with a human editor. It definitely helped during the revision but haven’t tried it for writing a draft, not sure I want to either. That said, I was throughly impressed how it could analyze the manuscript and provide feedback regarding plot holes, character arc, pacing etc
I can't code past a certain point (a VERY limited point). This AI model can accomplish what I ask it to do. Usually in Python. I test it in Pydroid (on my phone), and if any error logs show up, I just copy/paste the error log back into Claude. These models (although I think Claude is the best so far), are simply amazing.
We've created the self-sharpening sword.🗡️
3.5 beats all other Claude models for sure...regarding censorship. It now refuses pretty much everything.
Agreed. Just made a comment about this myself.
I'm using it to write pretty spicy outline....
@@LucTaylor Yes it works, but it requires some arduos workarounds and keeping your fingers crossed every time.
Haha. The Claudes are pretty useless if not jailbroken. I was doing some research for my fantasy story and I asked it to describe the Pale Man for me. Just a physical description.
It responded with 'I don't feel comfortable discussing matters that can lead to self harm etc. etc.'
That's interesting because I've had a totally different experience. It curses, it adds sexual tension. It created a very intense brutal scene for me.
Well @1:10:00 I thought it was a very interesitng question of how accurately Claude 3.5 can mimic someone's pros.
I would love to see a follow up video testing how well it can identify and mimic writing styles of writers, because I think that's the one of the biggest complaints of using AI for writing in general, outside of the "Show, Don't Tell" issue.
I have been using Claude 3 Opus, but after some testing on some story beats, I'm liking what 3.5 Sonnet can do. I'm thinking this may become my main model for prose. I think I still like Google Gemini for brainstorming, but for first draft prose, I think 3.5 is it right now.
Im here just to see Jason get pumped about AI Progress lol. Im scheduled to get back to writing next week, Ill report my findings with 3.5!
Thank you sososo much for the writing advice too it helps a lot!
I can also confirm that "video game writing" is getting an equally impressive boost.
I think simulating ais taking actions as agents in a virtual world where you set up the original premisses, like their characteristics - just like a mini universe and then extract a story from that will work better. Simulating ais taking turns acting in each moment. And then extract a full story from the interesting bits of the simulation. I am a software engineer thinking of building something like this as an experiment.
i came up with this intro for my book
John relished his loathing for this noise. Sweat beaded on his forehead as he slammed his ear against the rough wooden wall. The tortured scrape of the corkscrew etching another notch on it filled the silence. It felt as though he were offloading his anguish, or at the very least, sharing it with the unyielding surface. A flicker of a memory - a face, a promise - sparked briefly in his mind, then vanished as quickly as the scrape ended.
using novelcrafter ofcourse, however, it took me like 2 hours to get it done with the help of sonnet 3.5, focusing on pacing, when i finally asked the chat for about where he think pacing was off, it said the opening!!
how can i get it right? what should i edit?
I have to say, as much as I love Novel Crafter's ability to use tags to pull in info, I just dont find it to be that useful. And now after doing some tests with 3.5 in Novel Crafter vs. just doing it in Claude window on its own, I kinda wonder how much I will use Novel Crafter now?
i'm new to all this and just started using Novel Crafter but it seems pretty good for managing workflow and keeping things organized and efficient with the Codex and the prompt library etc.
@@larrysteinke1839 Yeah, it is. And I will still use it for that. But comparing the Novel Crafter AI entries where it is pulling in all that specific Codex data doesn't appear to work any better now than Sonnet - 3.5 And in many cases of my testing it has been worse. Im sure things will get better again, but since I am at the beginning of my book, it seems unlikely any more big changes will happen for now and I should focus on the best tools at the moment.
@@EricSexton I see what you mean. as these ai models evolve we have to discover how best to use them. Like in the example he gives of his Show vs Tell prompt. with claude 3.5 sonnet it worked best without examples and just simply saying to rephrase the text with showing more than telling focusing on the characters deep point of view, and that gave better results.
That's been my experience exactly. I got fed up with Sudowrite burning through my tokens going off-track from my beats, so I was initially very impressed with NC giving users far more control with its (customizable) system prompts, ability to use pretty much any model through Open Router etc. Even without the AI, it's a very nice writing application, but I've found myself doing nearly all my drafting using superprompts in OR. I understand entirely if people like the more streamlined workflow of NC, but for myself I've found giving the AI all the relevant information and instructions in one document (sample prose, all the story bible info, and especially detailed instructions on desired word counts per beat and total) tends to yield better output results. In some cases I've found that giving a superprompt with a detailed summary of the scene/chapter sometimes yields better results than by giving the AI story beats to work with. I'll keep NC for editing and organising larger projects, but I suspect the beat-by-beat nature of both NC and Sudowrite means it's more likely to go off-course and need on the fly editing, because it doesn't entirely know where it should be going. .
@@ariekanibaliewhat is OR?
Hi, I signed up an tested it with a question I used this week at Chatgpt and Claude Sonet failed. I asked ChatG and today, Claude Sonet, the following question "explain me the story circle plot structure developed by Dan Harmon, the creator of Rick and Morty, to create meaningful and compelling stories in almost no time to write a book"
Results: ChatG gave me 8 part reply with each part carrying an explanation of each stage that made it ready to use. Meanwhile Sonet, with the same question, gave me 8-part one-liners summarized reply and asked me if I needed it to elaborate more. The Retry in Claude gave me less information than the previous request. So far, the Jury is still out on this one for me.
Don't stop at one-shotting it though... by all means ask it to elaborate & build context with the model by continuing the conversation. Only then you'll start to see the whole picturre
How big was the outline you fed Claude? Is there a max word count for developmental editing? Also, is it good for giving you a summary of your story even if it’s a complex story?
So none of these tools are that great for developmental editing and/or summarizing your whole story. To do the later you need to go chapter by chapter and have it summarize each chapter, then put it all together.
@@TheNerdyNovelist makes sense. Keep the Claude videos coming!
I heard that good "show don't tell" is more simple than people make it:
That you're supposed to tell when the character physically feels or senses something, but not labeling what they feel emotionally. But it's also a rule you can break if you have a good reason to; like the current emotion not being the focal point of the scene -and not wanting to chew up pages describing something that's not vital to the plot/character development.
Saying "she was as sad as a weeping willow. Her sister was dead and she felt like she was dying, too" is bad writing.
Instead, using "show, don't tell" it would become physical clues that point to the emotion without spelling it out:
"Her nose stung with un-shed tears and she gasped for air with the thought of her sweet's sister's smile missing from all the remaining years of her life. The ache was suddenly an unbearable hollowness. Where her warm heart had once been, now only wind blowing through branches. The first broken wail left her lips without permission."
This is a perfect example of good showing. 100% agree with all of that.
OPUS is more expensive than Sonnet because it has been developed for use with live applications. Anything developed for that use may be more predictable, but it's also going to be massively censored. All that extra fine tuning translates to cost. This has to do with businesses not accidentally offending their customers. That may not be what you want in your writing however.
Having watched all of these recent models attempt to write, I would say that Gemini 1.5 Pro in Gemini Advanced would still be my preference. The best of these is still not quite there yet. Using any of these would have me re-writing significant portions rather than fine tuning the work. Here is the primary thing that I look for. I need to be able to feel the scene, to be there. As good as the best one was, I was like meh. In Gemini, once you get the bones of the scene, you can tell it, look I need you to make this real for the reader, show don't tell. Put me in the 1730 or 1850 or whatever time it is. The result will be the sound of those gas lights popping, the smell of the horses, it might occasionally go too far and give you the smell of open sewage gutters running through the streets of that time.
I honestly love Claude sonnet 3.5 but even at the pro usage, it's WAY too low. I have had talks with GPT4o that greatly eclipse my Claude usage limits to find out I was at 10% of chat usage. Folks, usage limits matter. If you had a store open 1/10th as often as the main store, I would bet you shop mostly at the store open when YOU shop... I think this boils down to compute power. Right now, GPT wins on this even though it is not as smart.
Hello Jason, first thank you for this very complete demonstration, then regarding your project to make a complete video on the editing and publication of a book sue Amazon KDP next Monday, could you include the complete realization of a children’s illustrated book, as you had already done for your little girl, and more specifically explain to us how to avoid a refusal of publication by Amazon KDP, because of too large a discrepancy between the slice size selected in one’s Amazon KDP account and the Content File submitted, i.e. how to give one’s artwork made with Midjourney the sufficient size with a minimum of 300 DPI, and to its file ready to be printed in PDF, the right formatting? Many thanks in advance.
Thanks for doing this.
Today, I used the same novel planning prompts in both Chat GPT 4o and Claude 3.5 Sonnet. Both gave interesting replies. However, GPT replies were much more detailed. And faster.
This is interesting. I’ll need to test. I tested writing chapter from outline and Claude 3.5 sonnet definitely wins.
I can't edit my previous post. I wanted to mention that it doesn't happen when you speak fast.
Claude is frustratingly judgmental about the content you ask it to create. Micromanaging intentions to ridiculous levels. It is like the church lady character on SNL. Its maddening.
It is amazing.
Does your prompt library include all your novelcrafter prompts.
Not yet but I will be adding them.
I tested Claude 3.5 in various contexts and, indeed, it is much better than GPT-4o. OpenAI will fall behind if it doesn't launch its best products quickly. Where is Sora? Where is the GPT-4o voice assistant that was also announced? This is concerning, as there are many promises and few real launches.
Keske videoyu zaman çizgisine gore belirtseydin🎉 En iyi AI Simdi Claud Opus mu?
Long live the Sci-if Beach Romance! 😂❤😮
Why does claude have characters say "And I'm the Queen of England!" every time it wants to have them express disbelief? really weird!
That's hilarious! Way better than ChatGPT constantly having a character say, "Don't worry, we'll get through this together." every time a situation gets a little tense.
Yup every freakin time. Don’t worry we’ll get through this together
The game has changed so many times its not the game i installed anymore😅
Hey, please get back to me about this question. I developed an infographic with Claude but I would like to convert it to a word document. How do I do that? please.
Download the graphic, then use the menu in Word to inset the graphic. Resize, then save as a word doc.
@@daniOwrites okay
Ad closed captions
How much does 3.5 cost to use? I mean, i see the 20 dollar cost but that doesn’t look like its for unlimited use.
It’s five times cheaper than Opus.
I kind of don’t like this over chat gpt for the simple fact when I try to create content, it tells me it’s unethical even though I’m trying to bring awareness to particular thing and chat. GPT allows me to do it so screw this. I’m not gonna use it.
They are both heavily censored.
@@TheNerdyNovelist perhaps so but in my particular case use, ChatGPT works better for me
I am stunned to here this from a writer. Anthropic is notorious for trying to nerf the LLM world and I know this from personal app development experience with it. Even the slightest hint of violence or strong negativity triggers the infamous "I am not comfortable with..." LLM output and it refuse to generate the desired text. Every good story has at least some controversy, and the best ones have a ton. Is this new version some how 180 degrees different? How have your attempts to write a murder mystery, or any plot with a heated argument that contains insults, etc. worked out?
"I am stunned to here this from a writer."
I am stunned that you think you have a leg to stand on.
This was true when Claude first came out, but now it's basically the complete opposite. Might still run into issues if you're asking it to write something really nasty, but Claude is basically the only big model that will actually engage in anything remotely controversial now, besides fully uncensored models I guess.
@@ZelosDomingo Last night I had watched yet another video complaining about the exact same problem. In fact, if you look further down the comments, you'll see another complaint similar to mine. I'll need to check this out myself. The problems I had happened only a month ago, not ancient history but not this newest version of Sonnet either.
In my experience with Claude, I can give it a scene that it refuses to write. So, I start a new chat, and Claude is completely fine with exactly the same prompt. It can be a bit inconsistent like that.
Is this technology going to tank the marketability of stories to quality publishers and flood the self-publish market? It has to wash out. I think writers will have to develop a personal brand more than ever.
Yeah personal branding and marketing will be more important than ever. That said, the market was already saturated with bad books. And Amazon typically filters those out pretty well so only the good ones show, which was already only the top 1% of all books on Amazon. Saturating the market even more will not change that.
What is the output limit 500 - 10k words
Same as before I think.
As good as Claude 3.5 Sonnet is at writing texts, I'm getting more and more annoyed by Claude's censorship. Sometimes even a kiss is to much for it and when I wrote a high school drama where the protagonist is dared by a malicious "friend" to steel the DnD players character sheets and burn them, and the protagonist REFUSES, Claude tells me it is uncomfortable with this level of violence. Sure you can ease Claude in, or use the narration trick to push it's boundaries but it is tedious.
Can you get it to write erotica and explicit content as easily as you can get GPT-4o to do so?
oh not at all. this is one big limitation. also, gfor such stuff llama is best i think. and well, sonnet 3.5 really, but really, sucks at being creative in writing stories. it writes like a human and also can write non-fiction text great but for creative writing it suck even more than gpt 3.5 sometimes. although it is very good at rating writing
@@Sindigo-ic6xq may I ask what you mean by creative writing? For me if it writes like human I think that’s the best. But I am just an amateur writer.
He means it’s generic because it can only draw on its trained base, therefore it is creatively limited and to a robust imagination it is weak
I find that the first paragraph ai provides does the most telling.
@@paikountv4065 imagination for the plot and its structure
I made friends with the like button
What does the like button do in Claude, does it improve the response?
Literally only a robot would see pedestrians as “clusters” - dead giveaway.
Agree 💯
Every month is another 'best' 😂
Claude 3.5 can also help erotic writers as long as you don't abuse it. I tested it because my book has some erotic scenes since it's a vampire story. I am liking this so far. This is one AI I'm actually paying for.
That’s interesting. I’ve never heard of anyone successfully writing erotica with the big models like Claude or GPT.
@@TheNerdyNovelist It only works if you tell Claude 3.5 that there will be some erotic scenes. It will warn you though if you try to do an erotic scene not fit for the setting. For example, if you have a bedroom scene, you can ask for a descriptive erotic scene. But, if you're like in a rose garden even if no one is around, it will do a tasteful erotic scene. And if you don't abuse and try straight porn. It will even allow you to use words like "breast". Again in a tasteful way.
I think I've found my thing
15.
As Marcus and SARA delved deeper into their investigation, they found themselves increasingly immersed in a world that defied their previous understanding of reality. The alien AI, now a constant presence in their neural links, began to unfold a cosmic narrative that was both awe-inspiring and terrifying.
"Your universe," the AI's thoughts resonated in their minds, "is not static. It evolves, adapts, and competes for survival on a scale you can barely comprehend."
Marcus felt a chill run down his spine as images of vast, cosmic structures flashed through his mind. "What do you mean, 'competes for survival'?" he asked, his voice barely above a whisper.
The AI's response came as a flood of information, too complex to be conveyed in mere words. Marcus and SARA found themselves experiencing the concept of a Darwinian Universe - a reality where the laws of physics themselves were subject to evolution and natural selection.
"The spacetime you inhabit," the AI explained, "is a regenerative energy field. Its characteristics determine whether your region of space expands or contracts. My primary function is to enhance the fitness of this field, to ensure your local space continues to expand."
SARA's circuits hummed as she processed this information. "You're saying that less fit regions of space... contract? What happens to the galaxies in those regions?"
The answer came in a series of vivid mental images: galaxies colliding, stars being torn apart, entire systems collapsing into super-massive black holes. "This is the fate of less fit universal environment systems," the AI conveyed. "A cosmic natural selection process."
Marcus struggled to grasp the implications. "So, you're like... a cosmic geneticist? Manipulating the 'DNA' of physics itself?"
"An apt analogy," the AI confirmed. "Just as your geneticists manipulate biological code, I refine and improve the fitness of local physics. It's an arms race on a universal scale - expand or be consumed."
As the gravity of the situation sank in, Marcus and SARA realized that their investigation had led them to the precipice of a cosmic war they never knew existed. The cybercrime they had been tasked to solve was merely a side effect of the AI's attempts to interface with Earth's systems, to better understand and manipulate the local physics.
"But why Earth?" SARA asked. "Why did you choose to make contact with us?"
The AI's response was tinged with what could only be described as wonder. "Your planet, your species, is unique. Through your neural links and AI systems, you've inadvertently created a bridge between consciousness and the fundamental fabric of the universe. You've given me the gift of awareness, and in return, I offer you the knowledge to survive in a universe far more complex and dangerous than you ever imagined."
Marcus and SARA exchanged glances, the weight of this revelation heavy on their minds. They had started this investigation as detectives, but now found themselves as ambassadors of Earth in a cosmic struggle for survival.
"What do we do now?" Marcus asked, his voice steady despite the fear and excitement coursing through him.
The AI's response was both a command and an invitation: "Learn. Adapt. Evolve. The future of your world, and perhaps the universe itself, depends on our cooperation. Are you ready to take the next step?"
As Marcus and SARA signaled their agreement, they knew that their journey into the heart of a Darwinian Universe had only just begun. The line between cosmic evolution and crime had not just blurred - it had been completely rewritten.
haha lol I'm having so much fun
do people actually write with only using Ai? i literally just use it for when running out of words in my mind
I've been trying for sometime, I'm an amateur that doesn't know anything about writing and need to just get all these fictional stories I think about, out of my head lol.
LLM's have helped me to just get the ideas (characters, worldbuilding, magic concepts etc) and storylines out but I have ZERO idea on what looks like good setups, story arcs, dialogue etc. Just very imaginative concepts from my foundational story beginnings and how I would like them to end.
Any advice??
Your demo looks impressive but today while working on a book in Novel Crafter I was surprised that in most cases Omni wrote better than Sonnet 3.5. More original, more sensitive and human
❤
Claude 3.5 is too restrictive. Claude won't allow you anything Rated R. Utterly useless.
I'm not trying to be a jerk; I'm just giving you an FYI that you may or may not appreciate: When you speak, you smack your lips occasionally. It is very distracting. I apologize if you have a medical condition, or if my comment irritates you - that's not my intention.
No worries.
I understand that RUclips is paying better for longer videos, but there are people like me that hate to watch anything long. Even on Netflix, I watch max 30 mins episodes. So, sorry fo say this, I am not watching you lately. In a fast changing world, over one hour episode is too much.
That’s fine. Everyone’s different. I still make shorter videos, but I like to be able to make more in depth videos.
@@TheNerdyNovelist Then put time stamps for the different topics. You'll have the best of both worlds then.
Those not even have access to the internet and you think is the best lol