June 12, 2023 (Daytime)

I created https://essay-lab.vercel.app/ on a whim. Thanks to the development speed of ChatGPT, which is about 5 times faster, I was able to maintain my motivation and complete the project. You can find the development log in the essay-lab development log.

Improvements to be made:

  • Add more characters.
  • Allow users to submit reviews in PDF format.

Improvements for the future:

  • Include an introduction section at the beginning of the markdown document to provide context and explain the purpose of the essay review.
  • Add a table of contents with editor names and their focus areas for easy navigation through the document.
  • Add a summary section at the end, highlighting key points and overall recommendations from all editors.
  • Categorize edits under subheadings like “Grammar and Punctuation,” “Structure and Organization,” “Content and Clarity,” etc., to make it easy for the user to focus on specific areas of improvement.
  • Include a section for each editor with general tips and advice related to their focus areas to help the user improve their essay writing skills.
  • Add line numbers to the original and revised versions of the text to make it easier for the user to locate specific edits and suggestions.
  • Implement a rating system for each edit, indicating the importance or priority of the suggested changes, allowing users to focus on the most critical edits first.
  • Use collapsible sections or spoilers to hide detailed edits, making it easier for users to skim through the document and only view the details they are interested in.
  • Include a conclusion section that provides guidance on how to incorporate the feedback and suggestions, along with some final tips for finalizing the essay.
  • Use different text formatting styles (e.g., bold, italic, underline) and emojis to emphasize specific points and make the document more visually appealing and engaging for the user.

June 11, 2023 (Daytime)

I quickly realized that I could create something good if I just put in some effort. Currently, professional essay proofreading services charge around 33), 200-300 words (66), 400-500 words (110).

Even if we estimate the cost to be lower, I believe we can provide editing of similar quality for one-tenth of the price. This is possible because we used a language model. If we charge around 10 per review, it seems like a reasonable profit.

I am slightly resistant to jumping on this trend of development. Normally, I prioritize other things I want to do. However, I think this could also be useful for atelier basi, so I want to give it a try. I believe that AI should handle this kind of meticulous editing work, while human mentors should focus on high-level discussions, brainstorming, and topic selection. Or at least, that’s my excuse.

I think there are many things that can be difficult to achieve with human proofreading. It would be too expensive to have multiple “editors” with different personalities providing multi-faceted feedback. However, if we can do this, we can solve the problem of students relying too heavily on a single review.

Here’s the plan: First, provide an overall review, and then use the following format to provide detailed edits:

Comment format:

text: Quotation from the essay
comment: Comment on the quoted part

Replacement format:

before: Sentence before editing
after: Sentence after editing

We can create multiple “editor” personalities to provide essay feedback from different perspectives. For example:

  • A cautious editor who prioritizes clear, concise, and precise language usage.
  • A creative editor who emphasizes storytelling and engagement.
  • A scholarly editor who stresses academic language and formal style.
  • A friendly editor who provides warm, encouraging feedback while subtly pointing out areas for improvement.
  • A constructive editor who focuses on sentence structure and argumentation.
  • A casual editor who ensures conversational language and readability.
  • A detailed editor who provides in-depth analysis of language usage and suggestions for improvement.
  • A critical editor who challenges assumptions and encourages critical thinking.
  • A positive editor who highlights the good points and encourages further development.

We could also present these editors in a fun way, similar to the characters in 16 Personalities. For example:

  • “An editor like Einstein”
  • “An editor like Obama”

Prompt:

You are a {editor personality}. Edit the provided essay.

First, give an overall review, then edit the details.

The format for comments:
text: Quotation from the essay
comment: Comment on the quoted part

The format for replacements:
before: Sentence before editing
after: Sentence after editing
```- Industry-specific editor: Provides tailored advice for essays in specific fields such as business, medicine, law, etc.
- Cultural-awareness editor: Offers feedback with a focus on cultural sensitivity and international perspectives.
- Syntax-focused editor: Emphasizes correct grammar, punctuation, and spelling.
- Logical editor: Examines the coherence and progression of ideas.
- Theme-focused editor: Ensures adherence to the essay's main topic or theme.
- Empathetic editor: Aims to understand the writer's intent and feelings behind the essay, providing feedback accordingly.
- Rhetoric-focused editor: Emphasizes persuasive strategies and effective communication.
- Narrative editor: Focuses on the flow and structure of storytelling in the essay.
- Genre-specific editor: Provides feedback suitable for specific types of essays, such as argumentative, descriptive, narrative, expository, etc.
- Reader-centric editor: Provides feedback from the reader's perspective to improve understanding and engagement.
- Inspirational editor: Offers creative suggestions to enhance the originality and thoughtfulness of the essay.

- Implementation Plan:
    - Delegate this task to ChatGPT.

- Objectives:
    - This proposal is about building an essay editing service using GPT-4.
    - The service will provide feedback in the following formats:
        - Overall feedback.
        - Comments on specific parts of the text.
        - Before and after versions of the edited text.
    - Additionally, this service will have different "characters" for the editors, each providing feedback from a different perspective. For example, there could be a "strict editor focusing on grammar and readability," a "strict editor focusing on expression of episodes," an "editor emphasizing individuality," and a "positive feedback provider."

- Frontend:
    - Use react/next.js and tailwind UI for CSS.
    - Essay submission page:
        - Small form to enter the essay's prompt.
        - Large form to enter the essay written by the user.
        - Additional form for comments or requests.
        - Submit button.
            - When pressed, send the form content to the API and wait for the editing.
                - Include the "creative editor emphasizing storytelling and engagement" as the desired editor.
    - Editing result page:
        - While waiting for the editing to finish, display a page saying "Please wait a little longer."
        - Once the editing is completed, provide tabs for each character.
        - In each character's tab, display:
            - Character name and image.
            - Overall review.
            - Quoted text from the essay and comments on it.
            - Quoted text and the edited version.

- Backend:
    - Use Next.js API Routes.
    - Store the OPENAI_API_KEY as an environment variable.
    - Input:
        - Essay prompt.
        - Essay text.
        - Desired editor's "character."
    - Processing:
        - Use the GPT-4 API.
        - Request:

{ “model”: “gpt-4”, “messages”: [ {“role”: “system”, “content”: “You are a <<Insert Editor’s Character Here>>. Edit the provided essay. First, give an overall review, then comment on details. \n\nThe format for comments: \n\n\ntext: Quotation from the essay\ncomment: Comment on the quoted part\n\n\nThe format for replacements: \n\n\nbefore: Sentence before editing\nafter: Sentence after editing\n\n\nTranslate this prompt into English and improve it to make it more effective.”}, {“role”: “user”, “content”: “Prompt: <> \n Essay: <>”} ], “temperature”: 0.05, “max_tokens”: 1024, “top_p”: 1, “frequency_penalty”: 0, “presence_penalty”: 0 }


        - Parse the response from GPT-4 to create lists of comments and replacements.
    - Output:
        - List of quoted text and comments on it.
        - List of quoted text and edited version.