I Trained a Fine-tuned Model on My Books (Demonstration)

Fine-tuned models are the future, and in this video I show you how. We cover what fine-tuned models are, how they differ from regular prompting, and especially what you have to do to create the dataset that they need to make your own.

My Top AI Tools:

Find me!

[brandileigh_thepoet]

This is awesome Jason! I usually am the quiet one in the corner that soaks in all the information but no one knows I'm here. I thought I'd comment and say congrats on this launch and I've also signed up!

[starlitpathways]

I love the idea of 'for a certain amount a month, you too can have this', but... Most people need a subscription to an AI platform, plus, a subscription to a writing platform, plus office etc. Then, most people work, leaving maybe one day a week to actually work on their stuff. You have needs, we have needs... Could you consider a smaller package that would include tinkerer's with jobs and families?

[dameanvil]

00:18 🎙 Introduction to Fine-Tune Models:
Fine-tune models are highlighted as a powerful tool, emphasizing their significance in AI writing.
01:27 📈 Fine-Tune Models vs. Open-Source Models:
Fine-tune models, even based on GPT 3.5, are noted to outperform higher-end models like CLAE 2, especially in crafting fictional prose.
03:01 🤔 Methods of Achieving AI Results:
The video compares various methods, including prompting, prompting engineering, and retrieval augmented generation (RAG), before delving into the specifics of fine-tuning.
03:39 🧩 Fine-Tuning Overview:
Fine-tuning is described as a method to change the style of AI output, primarily beneficial for authors aiming to capture specific writing styles, both fiction and non-fiction.
04:12 🚧 Accessibility Challenge:
Fine-tuning is acknowledged as challenging for many authors due to technical barriers, requiring substantial effort in creating datasets and dealing with technical aspects.
07:01 🔄 Fine-Tuning Process:
The step-by-step process of fine-tuning is explained, involving the creation of a dataset with 50 to 100 examples, each comprising a beat, a user prompt, and an assistant prompt.
10:05 ⚙ Example Use Case:
An example scenario for a fine-tune model is illustrated, demonstrating how it can transform overly flowery, GPT-generated text into a more natural and personalized writing style.

[juliefutcher]

Great video. Would these tools work for non-fiction books?

[-Evil-Genius-]

🎯 Key Takeaways for quick navigation:

00:00 🌐 *Introduction to Fine-Tuned Models*
- Overview of fine-tuned models and their significance in AI writing,
- Emphasis on the potential of fine-tuned models in 2024,
- Introduction to the demonstration of a fine-tuned model.
03:01 🧩 *Methods of Achieving AI Results*
- Explanation of different methods to achieve AI results: Prompt engineering, Retrieval Augmented Generation (RAG), and Fine-tuning,
- Distinction between the approaches and their effectiveness,
- Focus on fine-tuning as a powerful method for changing writing style.
05:19 📚 *Creating a Fine-Tuned Model*
- Step-by-step breakdown of the fine-tuning process,
- Importance of a consistent system prompt for creating fine-tune models,
- Example of creating a data set for a fine-tuned model that generates prose from a beat.
08:01 🛠️ *Practical Examples of Fine-Tuned Models*
- Practical examples of fine-tuned models, including one for editing text to match a specific style,
- Demonstration of how to create a data set for an editing fine-tuned model,
- Highlighting the power of fine-tuned models in refining AI-generated text.
12:15 🚫 *Challenges in Fine-Tuning*
- Identification of challenges in the fine-tuning process, particularly in creating the necessary Json file,
- Mention of existing tools that may be costly for fine-tuning,
- Announcement of making a specific fine-tuned model accessible to the audience.
13:24 💡 *Membership Announcement*
- Introduction and announcement of the speaker's membership program,
- Description of the benefits and freebies offered to members,
- Emphasis on the accessibility of a specific fine-tuned model for members.

Made with HARPA AI

[raphaelavondrachek]

This is wonderful! Could you possibly create a step by step video on how to do this on OpenAi? I saw your steps in this video, but I still don’t know where and how to do that and connect it with NovelCrafter.

[evolutiondl]

I've been waiting for a fine-tuning video for a while now. Great video. I believe it's something every author who wants to use AI seriously should know about, but I have a question. Do you think you can fine-tune the AI to be less censored or perhaps fine-tune the AI to produce proses that aren't always positive?

[roybradley]

great as always...thanks for putting this out. AI models have given me the impetus to try and start writing a book myself, really appreciate your content on this as its given me inspiration.

[martymcfly]

Can you do this with regular gpt4? Or do you have to use tbe playground feature? I noticed when i do this on gpt4 it slowly forgets over time

[Paikoun]

Hello Jason

If I subscribe and download your fine-tuned model, will I be able to use it with any AI like GPT-4, Novelcrafter, or Sudowrite? Do I need to have an account on the OpenAI playground? Additionally, is the usage limited only to the duration of my membership, or can I continue using it even after I am no longer a member?

[kaoline]

MANY HAPPY RETURNS OF THE DAY, JASON. WISHING YOU ALL THE VERY BEST ON YOUR SPECIAL DAY!

[unfluster]

Yeah, I spent over a few days (probably 30 minutes each day), writing out the system, user and assistant (not 50 as suggested, maybe 30) and then implemented it on GPT-4. Let me put it this way...if it did work, it was so slight (reduced some of the purple prose) but it really wasn't worth the time. It's possible I did something wrong, but I followed the instructions as best as I could. Did anyone else try this? If so what were your results?

[gypsyproductions]

I have never been a big fan of having to subscribe to get more info. BUT, I took the plunge here and I will say that it is well worth it. It is obvious to me that he is dedicated to putting out a quality product. The ROI here is a great deal. I would like to add that the availability of quality AI writing tutorials is very limited. Not all of us have 2 hours to watch a video of someone writing a novel in AI while trying to pick out points that will apply to our needs. His videos are concise and generally spot on for my needs.

[duesenberger]

At 11:45 You integrated it into Novelcrafter. How do You do this?

[tsentenari]

9:08 Are you sure that giving it a piece of you writing, and then telling it to write this in its own style is a good idea? I could imagine that there are too many clues in the good version.
For example, in my experience ChatGPT has an almost indelible tendency to drift into telling instead of showing. But would it really be as likely to do this, if the prose it got fed already got it right?

[duesenberger]

Fine tuning a model looks like a great idea. But if I use Claude (chat interface) I soon run into the limitation that Claude tells me it cannot continue the conversation. So where can I fine tune a model? Only in the playground?

[drgnmsr]

So, if I was to take a book by an author that I wanted to style my writing on, I could have ChatGPT look at each chapter and identify the beats and segment the chapter based on those beats to use for fine tuning my own model of a authors style?

[ashlynnantrobus]

I'm actually working on a free tool that will create those JSON files for you. Hoping to launch it this week

[johncressmanci]

You've taught me alot so I went and signed up!

[FutureFictionAcademy]

It's a JSONL file, not JSON, and yes... that matters. It's the only way to have multiple lines in your prompt and JSON files are not accepted.