Please read. Significant change on the site that will affect compatibility [ Dismiss ]
Home ยป Forum ยป Author Hangout

Forum: Author Hangout

Character Analysis with ChaptGPT

JoeBobMack ๐Ÿšซ

I'm working on a new story and, for the first time, I developed character sheets for my main characters. It was a helpful exercise because I built their characters with things I'm familiar with from psychology. Then I started writing and... the characters started surprising me. Not completely, but some. And so I wondered if what was appearing on the page matched what I had planned.

Now, it would be GREAT to have a person who was eager to read anything I wrote and spend time studying it to answer all my questions but,... sigh. That person doesn't exist. So, I thought about trying ChatGPT.

I gave ChatGPT these instructions:

Based on the draft below of the first chapter of my story, please list each character that appears in this chapter and what the reader will know about their personality, goals, challenges, and skills/hobbies/strengths. Create a format for your report that will make it easy to scan the information for each character. Here's the chapter:

I then pasted in the draft of my first chapter and submitted that prompt. I got back something like this:

A*** P*****

Personality: Reflective, responsible, worries about her actions, has a knack for winning others over (WOO).
Goals: Ensure her business venture is successful, manage her feelings of responsibility towards her investors.
Challenges: Balancing her personal sense of responsibility with the realities of her business, dealing with her father's work habits and her mother's needs.
Skills/Hobbies/Strengths: Strong interpersonal skills, understanding of people's struggles, potentially good in business.

C****** C*****
Personality: Supportive, optimistic, adventurous.
Goals: To be successful in the joint venture with Anita, embrace the adventure of their business.
Challenges: Facing potential bankruptcy if their venture fails.
Skills/Hobbies/Strengths: Supportive and positive mindset, willingness to take risks.

J**** B*******
Personality: Innovative, possibly manipulative, struggles with understanding emotions.
Goals: Develop and protect his creation, A***, and find a way to integrate it into society.
Challenges: Handling the ethical and practical implications of his creation, understanding and teaching emotions.
Skills/Hobbies/Strengths: Technologically skilled, particularly in AI and AGI, good at manipulating situations to his advantage.

I then moved through each chapter, giving ChatGPT its previous analysis and the new chapter, and asking it to look for new information or any contradictions. So far, it picks up new information, but hasn't suggested any outright contradictions. (Good! I hadn't intended to write any!)

Has anyone else tried something like this? Message me if you'd like to exchange notes.

What else could I try?

Seems like this would be a good way to develop character lists for some of those works on SOL where that is desired, but should be done only by the author or with the author's permission -- many would not want their work pasted into ChatGPT, even if it were later deleted.

Replies:   Joe Long  Paladin_HGWT
Joe Long ๐Ÿšซ

@JoeBobMack

I haven't done exactly that but similar, and yours sounds like a good idea. I subscribed to ChatGPT4 and am approaching my story with a top-down approach. First I am writing a one or two page synopsis of the story and the method of writing. After submitting that via PDF upload, I am doing brief summaries of each chapter. What happens, how those things are supposed to move the plot and develop the characters, etc. Third is a detailed summary of each scene. Those three levels, increasingly more detailed, are to check for story structure and plot and character development. Lastly, I will upload the complete text of each scene to review the quality of the prose. I even had one instance of the Creative Writing Coach tutoring me on how to construct this approach to submit to a custom GPT trained in creative writing.

Replies:   JoeBobMack
JoeBobMack ๐Ÿšซ

@Joe Long

Thanks for sharing what you are doing! What is ChatGPT doing. I'm doing more planning on the story I'm working on in my GPT than I've done before, but nothing as detailed as what you're doing -- so really interested in what you are learning from the process.

-- Have you created a custom GPT for this? That would mean you could store some of the documents you are developing in that GPT and let the engine use them when responding.

-- What kind of feedback are you asking ChatGPT for, and is it helpful?

Looking forward to learning and sharing experiences. Hopefully some others will join the discussion!

Replies:   Joe Long
Joe Long ๐Ÿšซ
Updated:

@JoeBobMack

My first ventures were with v3.5, then I paid the $20/mo to get v4. That comes with a built in Creative Writing Coach but you can also create a custom GPT and ask for it to be an expert in creative writing.

I've been very pleased with the results but you do have to train it to tolerate erotica. Make sure to give trigger warnings when submitting text, tone down the language and make sure the love scenes advance the plot and the characters! Sometimes it will hide replies and you have to coax it to redisplay.

I've asked it to make sure sure the characters have distinct voices, that there are enough action beats in the dialogue (it sometimes offers that I'm telling instead of showing), that there are no plot holes, that scenes flow from one to another, that the dialogue is authentic, that events are believable and realistic, and likely a handful of others. It tells me when scenes are touching or funny or a key part of a character's development.

Replies:   Joe Long
Joe Long ๐Ÿšซ

@Joe Long

For example, here is the response to Chapter 2 of my detailed scene summaries:

Thank you for providing a detailed summary of Chapter 2. It gives a clear insight into the progression of the story, the deepening of the characters, and the complex emotions at play. Here's an analysis focusing on various aspects:

1. **Character Development**:
- **Joe**: His internal conflict is vividly portrayed. His struggle with his feelings for Hannah and the associated guilt and self-loathing are palpable. This adds depth to his character, highlighting his vulnerabilities and moral dilemmas.
- **Hannah**: Her character continues to develop, showing a mix of innocence and maturity. It would be beneficial to explore her internal thoughts and feelings more to balance the narrative focus.

2. **Themes and Conflicts**:
- The story continues to explore themes of desire, guilt, and self-identity. The internal conflict Joe faces, compounded by his familial and societal pressures, is effectively used to drive the narrative.
- The potential taboo of Joe's attraction to Hannah is a significant source of tension and should continue to be handled thoughtfully.

3. **Plot Progression and Pacing**:
- The plot moves smoothly with a good mix of dialogue, action, and introspection. The pacing seems appropriate, allowing enough space for character development and plot advancement without rushing through key moments.

4. **Emotional Depth and Realism**:
- The scenes where Joe struggles with his feelings for Hannah are intense and emotionally charged. This contributes to the realism of the narrative, as it reflects genuine human complexities.
- Joe's interaction with his father and the emotional repercussions offer a realistic portrayal of family dynamics and generational conflicts.

5. **Setting and Atmosphere**:
- The setting continues to play a background role, supporting the narrative without overpowering it. Consider integrating more specific cultural or historical references of the era to enhance the atmosphere.

6. **Moral and Ethical Considerations**:
- The scene where Joe contemplates inappropriate action while Hannah is sleeping is crucial. It's imperative that such scenes are handled with sensitivity, focusing on the psychological impact and moral implications for Joe.

7. **Overall Cohesion with Chapter 1**:
- The story's progression from Chapter 1 to Chapter 2 is cohesive, with consistent character behavior and escalating emotional stakes. The narrative is building effectively towards more complex developments.

In conclusion, Chapter 2 successfully deepens the narrative and character complexities introduced in Chapter 1. As you move forward, maintaining this level of emotional depth and character development will be key. Also, balancing the narrative to provide more insight into Hannah's perspective could enhance the story's depth. The themes of internal conflict, desire, and familial pressure are well-executed, and continuing to handle sensitive topics with care will be important for the story's overall impact.

Replies:   JoeBobMack
JoeBobMack ๐Ÿšซ

@Joe Long

Thanks. That is similar to what I've gotten as responses.

Paladin_HGWT ๐Ÿšซ

@JoeBobMack

I couldn't seem to send you a PM. Please PM me.

I would like to learn about how you did what you did with ChatGPT, or whatever program.

However, I am concerned about letting ChatGPT "Borg" my writing. Is it possible to instal ChatGPT on a computer that is not connected to the internet; then have it analyze my witing (without access to the internet.)

Replies:   JoeBobMack  JoeBobMack
JoeBobMack ๐Ÿšซ

@Paladin_HGWT

Not sure why you couldn't PM me, but just sent you one.

I'm not sure exactly what you mean by "Borg" but as a paid user of ChatGPT ($20/mo) there is a setting where you can let it keep a history of your chats (which I find very helpful) and use those chats in training data. However, as I understand it, LLMs do not keep a "copy" of the material on which they are trained. Instead they "extract tokens."

I asked ChatGPT to explain this to me like I was ten, and this is what it produced:

Imagine you have a giant jigsaw puzzle with millions of tiny pieces. Each piece is a bit of information or a word from a book, a website, a conversation, or any other place where words are used. When a large language model like me, ChatGPT, is trained, it's like carefully looking at each piece of the puzzle and learning how they fit together to make pictures (or in my case, sentences and ideas).

Now, here's the important part: after I learn how these pieces fit together, I don't keep the actual puzzle pieces. Instead, what I keep is the knowledge of how they fit together. This knowledge is what we call "extracting tokens."

A token is like a tiny clue about a puzzle piece. For example, if the puzzle is about animals, a token might tell me that there's a piece with a "tail," another with "fur," and another with "four legs." When you ask me a question, I use these clues to put together pieces that make sense. It's like I'm building a mini-puzzle each time using the clues I remember, but I'm not using the exact pieces from the original puzzle.

So, when people say I don't keep copies of the material I'm trained on, it means I don't store the actual texts, conversations, or specific details. Instead, I remember the patterns and clues about how words and ideas usually go together, which helps me answer questions and talk about different topics!

JoeBobMack ๐Ÿšซ

@Paladin_HGWT

As for using a LLM on your computer, yes, for some of the open source models. I understand some of these compare very favorably to ChatGPT 3.5, although none match up to ChatGPT 4. I don't know about "context" limits -- how much the LLM can keep in memory while it carries on a chat. This is important for analyzing larger chunks of text.

If you'd like a quick overview of using some of these models and the tools that are out there which can make it easier, I found this video by The Nerdy Novelist to be helpful.

Back to Top

Close
 

WARNING! ADULT CONTENT...

Storiesonline is for adult entertainment only. By accessing this site you declare that you are of legal age and that you agree with our Terms of Service and Privacy Policy.


Log In