Thanks for doing this research! I suspect they created ChatGPT 4.5 with AI agents in mind rather than pure creative writing. Though their release announcement seemed to indicate otherwise. It will be interesting to see how it shakes down against other, less expensive models as an AI agent. We may be seeing an end to those old scaling laws otherwise.
I like claude 3.7 for its verbosity, but then I take all that dump and give it to 4.5 for ideation. Its not necessarily any better at writing itself, but the ideation is off the charts. I think you are doing a disservice to the model by sticking with very simple prompts. It's nice to want a model to be something it's not, but that's not very useful.
Interesting. You should test 4.5 vs 4o for nsfw content- I haven’t pushed 4o to the limit but it’s been way better for edgier nsfw content lately- yesterday ChatGPT had no problem assisting with some thing that DeepSeek even rejected as something that they couldn’t have a conversation about which is interesting considering how open source DeepSeek supposed to be more friendly for nsfw writers
GPT is 80 prompts every 3 hours. Since they moved the limit to 80 prompts a year ago, I haven't hit a limit once. But back when it was 40 prompts every 3 hours, I hit the wall at least 75% of the time. Claude is 35 prompts at MOST, and scales down drastically with prompt size, every 5 hours. It also scales down based on the length of your chapters while GPT does not. I use 140,000-character chapters, which tanks Claude entirely, resulting in a maximum of 9-10 prompts per 5 hours when near the end of a chapter. This practice is an integrity violation, and they know it. When I am writing my books, the prompt sizes that I usually have resulted in 12-18 prompts every 5 hours with Claude. They can take that worthless, cheapskate business model and cram it.
To say nothing of their 1950s era, draconian censorship. I thought GPT's was the worst in the industry, until I saw Claude's refusals for basic descriptions. Not NSFW, not egregious violence, just based on its broken, warped, Victorian era morals.
I'm new to using ChatGPT as a writing aide, but I came to it as a marketing research tool for my business first. At a seminar on AI for business it was recommended spending a minimum of 8 hours teaching the AI all about my business so I could receive answers tailored to me. When I learned it could be used as a writing tool I uploaded all of my old writing and began another long hours long conversation letting the AI understand my thoughts on writing and using my past work as a style guide. When I started writing with it I didn't find that it was as vanilla as you reference to, it was writing like me, in my style. When my prompts were not explicit "….in the style of story X, and Y references uploaded" I got very spare writing that was well below the prompted word count. As it stands, I'm finding my ChatGPT 4o is actually doing a good job at writing like me, so my role is more as film director with prompts at the beginning, and editor at the end.
Why do people keep making videos on polls? lol it's like we learned nothing. Polls dont matter. Opinions don't matter. None of that is going to get him out of the white house. These videos are just pacifiers so you people can vent angrily about Trump and dumb supporters and nod to yourselves about how you were right. You're acting like Trump supporters after 2020. And I bet the polls are barely above 50 percent… but yea the title says DROWNING lol seriously learned nothing. Still posting exaggeratted headlines for views- because after all you wanna get PAID and these suckers in the comments are making sure you do. Welcome to indy FOX news
Just had to go check… 52 percent in the oh so factual "Polls" lol That's another day at the golf course for Trump. Theyve always ALWAYS been low. But yeah… he's DROWNING IN DISAPPROVAL!!! Did you really have to put three exclamation points? Does the algorithm that pays give you more recommendations per exclamation? lol put your soul up sale, it's sellers market.
I’ve tried both Claude and ChatGPT, but neither stays on track, even with precise prompts. I know where my writing needs to go but don’t have the time to do it all myself. I’m open to any suggestions moving forward.
My experience is limited but i found the free versions of both chatgpt & gemini were impressive with modern day fiction stuff. Im thinking the sci-fy & fantasy categories being in the more extreme area of creative writing. I wonder if there might be just more creative prompts needed or other developmental steps to take to get more useable results. If sudowrite is nearest to being king of creative writing, how do the building components of it vs latest chatgpt compare in types of prompts & steps involved to get useable results? If all competitors in ai creative writing tools are moving towards "most simple" prompts to effect results, are they all really moving in same direction with same standards? Im guessing the leaders are trying to find more ways to differentiate themselves from the pack, that qualitative results might be difficult to achieve using across the board same methods used to arrive at similar points.
can i use claude 3.7 in the chat window to write a book? or do i have to get the api? the online version has a limit that can be used up quite fast i think?
I use paid chat 4-0 for brainstorming sometimes and in the last week it’s gone from being a real help to epitomising a lazy teenager. Not going to be worth the effort unless something changes.
How would using an AI to write with model prompts be different from conversing with the AI as a writing assistant that you bounce ideas off of or area those different steps in the same process
AI writing tools have come a long way, but I’ve found that the key isn’t generating content, it’s refining what you’ve already written. I used to struggle with polishing my drafts and making my ideas flow smoothly without sounding robotic, until at the end of 2024 when I discovered BeLikeNative. This advanced AI-powered tool doesn’t write for you; it enhances clarity, fixes grammar, and helps maintain your unique voice effortlessly. If you’re serious about improving your writing without breaking the bank, BeLikeNative is the only tool you need!
I've had amazing results from GPT 4.5, it's taken the main slot from Sonnet 3.7 for me. It likes a lot of prompt engineering, it's very context sensitive.
I am using ChatGPT pro for that. The API pricing is crazy.
GPT 4.5 is not available to users yet.
I have a question: should I use the extended thinking mode for Prose writing and Outline generating or just the normal 3.7 sonnet?
Thanks for doing this research! I suspect they created ChatGPT 4.5 with AI agents in mind rather than pure creative writing. Though their release announcement seemed to indicate otherwise. It will be interesting to see how it shakes down against other, less expensive models as an AI agent. We may be seeing an end to those old scaling laws otherwise.
I disagree with you on the "compelling" being too much in that sentence.
Still waiting on the Grok 3.0 review.
Pointless reviewing any other ai jason, you will always be a claude fan
I love 4.5. I haven’t had issues with it. It’s all depends how you use it.
Sounds like they're charging a steep premium for having a slightly higher number in the model name, and that's basically it?
Finally! The only YouTube channel I trust on this topic! Especially after so many baseless claims.
I like claude 3.7 for its verbosity, but then I take all that dump and give it to 4.5 for ideation. Its not necessarily any better at writing itself, but the ideation is off the charts. I think you are doing a disservice to the model by sticking with very simple prompts. It's nice to want a model to be something it's not, but that's not very useful.
thank you for thia great video
Interesting. You should test 4.5 vs 4o for nsfw content- I haven’t pushed 4o to the limit but it’s been way better for edgier nsfw content lately- yesterday ChatGPT had no problem assisting with some thing that DeepSeek even rejected as something that they couldn’t have a conversation about which is interesting considering how open source DeepSeek supposed to be more friendly for nsfw writers
GPT is 80 prompts every 3 hours. Since they moved the limit to 80 prompts a year ago, I haven't hit a limit once. But back when it was 40 prompts every 3 hours, I hit the wall at least 75% of the time. Claude is 35 prompts at MOST, and scales down drastically with prompt size, every 5 hours. It also scales down based on the length of your chapters while GPT does not. I use 140,000-character chapters, which tanks Claude entirely, resulting in a maximum of 9-10 prompts per 5 hours when near the end of a chapter. This practice is an integrity violation, and they know it. When I am writing my books, the prompt sizes that I usually have resulted in 12-18 prompts every 5 hours with Claude. They can take that worthless, cheapskate business model and cram it.
To say nothing of their 1950s era, draconian censorship. I thought GPT's was the worst in the industry, until I saw Claude's refusals for basic descriptions. Not NSFW, not egregious violence, just based on its broken, warped, Victorian era morals.
BTW: Thanks for these videos.
Thanks! Very useful ideo!
I'm new to using ChatGPT as a writing aide, but I came to it as a marketing research tool for my business first. At a seminar on AI for business it was recommended spending a minimum of 8 hours teaching the AI all about my business so I could receive answers tailored to me. When I learned it could be used as a writing tool I uploaded all of my old writing and began another long hours long conversation letting the AI understand my thoughts on writing and using my past work as a style guide. When I started writing with it I didn't find that it was as vanilla as you reference to, it was writing like me, in my style. When my prompts were not explicit "….in the style of story X, and Y references uploaded" I got very spare writing that was well below the prompted word count. As it stands, I'm finding my ChatGPT 4o is actually doing a good job at writing like me, so my role is more as film director with prompts at the beginning, and editor at the end.
Can you do a video on how to write a series using Novelcrafter
Why do people keep making videos on polls? lol it's like we learned nothing. Polls dont matter. Opinions don't matter. None of that is going to get him out of the white house. These videos are just pacifiers so you people can vent angrily about Trump and dumb supporters and nod to yourselves about how you were right. You're acting like Trump supporters after 2020. And I bet the polls are barely above 50 percent… but yea the title says DROWNING lol seriously learned nothing. Still posting exaggeratted headlines for views- because after all you wanna get PAID and these suckers in the comments are making sure you do. Welcome to indy FOX news
Just had to go check… 52 percent in the oh so factual "Polls" lol That's another day at the golf course for Trump. Theyve always ALWAYS been low. But yeah… he's DROWNING IN DISAPPROVAL!!!
Did you really have to put three exclamation points? Does the algorithm that pays give you more recommendations per exclamation? lol put your soul up sale, it's sellers market.
I happen to think ChatGPT writes damn good.
Can you maybe share the spreadsheet? I am really interested
Are the prompts you use are same what are available in skool/classroom
GPT gets ‘worse’ as no memory. I was in Grok for days and test its memory- insane
I’ve tried both Claude and ChatGPT, but neither stays on track, even with precise prompts. I know where my writing needs to go but don’t have the time to do it all myself. I’m open to any suggestions moving forward.
My experience is limited but i found the free versions of both chatgpt & gemini were impressive with modern day fiction stuff. Im thinking the sci-fy & fantasy categories being in the more extreme area of creative writing. I wonder if there might be just more creative prompts needed or other developmental steps to take to get more useable results. If sudowrite is nearest to being king of creative writing, how do the building components of it vs latest chatgpt compare in types of prompts & steps involved to get useable results? If all competitors in ai creative writing tools are moving towards "most simple" prompts to effect results, are they all really moving in same direction with same standards? Im guessing the leaders are trying to find more ways to differentiate themselves from the pack, that qualitative results might be difficult to achieve using across the board same methods used to arrive at similar points.
can i use claude 3.7 in the chat window to write a book? or do i have to get the api? the online version has a limit that can be used up quite fast i think?
I use paid chat 4-0 for brainstorming sometimes and in the last week it’s gone from being a real help to epitomising a lazy teenager. Not going to be worth the effort unless something changes.
How would using an AI to write with model prompts be different from conversing with the AI as a writing assistant that you bounce ideas off of or area those different steps in the same process
With the right guidance, Gemini Advanced 2.0 Flash is surprisingly good.
You are the only one out there doing these kind of tests on the models. Great work!
AI writing tools have come a long way, but I’ve found that the key isn’t generating content, it’s refining what you’ve already written. I used to struggle with polishing my drafts and making my ideas flow smoothly without sounding robotic, until at the end of 2024 when I discovered BeLikeNative. This advanced AI-powered tool doesn’t write for you; it enhances clarity, fixes grammar, and helps maintain your unique voice effortlessly. If you’re serious about improving your writing without breaking the bank, BeLikeNative is the only tool you need!
I've had amazing results from GPT 4.5, it's taken the main slot from Sonnet 3.7 for me. It likes a lot of prompt engineering, it's very context sensitive.
I am using ChatGPT pro for that. The API pricing is crazy.
Comments are closed.