Three user cases with ChatGPT4o

Hey LinkedIn community,

I've spent at least 3 hours daily exploring ChatGPT-4o since its launch to develop and test various use cases. Here are three specific cases that I found particularly inspiring and useful, which I believe can offer valuable insights.

Why

While traveling in Taiwan during the release of ChatGPT-4o, I began connecting pain points with actual solutions. Thanks to recent redundancy, I've had ample time to think, write, execute, and iterate.

Cases

All prompts and output can be found here: github

Case 1: Data Extraction and Visualisation

Pain Point: Information is often not fully digitized or structured for easy consumption, requiring manual collection and input, especially for tables in PDF/JPEG formats.

Requirement: Collect and visualize information from tables(such as income statement).

Expected Output:

• Income statement data stored as a Python dataframe for further analysis.
• A Sankey diagram to illustrate the flow of income.

Process:

Create step-by-step prompts to:

1. Read the annual report in PDF and extract all income statement-related data.
2. Populate a Python dataframe with the extracted data.
3. Draw a Sankey diagram using Plotly.
4. Additional requirements include breaking down income into granular levels, color-coding nodes, and linking node size to data value.
5. Iterate until achieving the final output.

Output:

• A Sankey diagram and Python code created by ChatGPT-4o.
• A dataframe containing the extracted information, verified against the original PDF source.

Case 2: Information Extraction and Summarization

Pain Point: Information is spread across the entire file, making collection and summarization very time-consuming. For example, ESG information in an annual report.

Requirement: Collect and summarise ESG information from a company's 2023 annual report.

Expected Output: A structured summary and some visuals.

Process:

Create step-by-step prompts to do:

1. Read the file and find everything related to ESG.
2. Populate a structured summary, including ESG mission statement, targets, progress, explanations/reasons/comments on progress, and further approaches to meet targets.
3. Additional requirements: add page numbers for reference and fact-checking/verification.
4. Create a subject-predicate-object diagram to depict the ESG information.
5. Several iterations until the final output.

Output:

• Structured Summary: A detailed summary of the company's ESG mission, targets, progress, explanations, and future approaches, with page references for verification.
• Visual: A diagram that clearly depict the ESG information, making it easier to understand and analyze.

Case 3: Rapid Book Understanding: Summarise, Analyse, and Visualise

Pain Point: Limited time to read an entire book and difficulty retaining content even after reading and taking notes.

Requirement: Understand the main argument, supporting evidence, summary of each part/chapter, cases/methodologies and their applications, and the book's novelty and limitations.

Expected Output: A structured summary and possibly a diagram.

Process:

Created step-by-step prompts:

1. Read the book file.
2. Summarise: overview of the book: main argument, background information, supporting evidence/analysis, counterarguments, and logic flows.
3. Summarise:overview of each chapter/part: main ideas, key evidence, key concepts.
4. Cases, methodologies, and applications: summary of several important cases, methodology/framework used, and how these can be applied to certain cases.
5. Review:main idea, supporting analysis, novelty and limitations of the book.
6. Create a subject-predicate-object diagram.
7. Several iterations until the final output.

Output:

• Structured Summary: A detailed summary of the book's main argument, supporting evidence, chapter summaries, cases, methodologies, and a review of its novelty and limitations.
• Visuals: Diagrams that clearly depict the book's content, making it easier to understand and analyze.

Final Thoughts

Thinking Matters:

It's crucial to understand the problem, the expected outcome, and the solutions needed (even at a high level). This allows me to break down complex tasks into smaller subtasks and instruct the LLM to perform them precisely.

Conductor, Not Performer:

My role in testing and building is like a conductor, not a performer. I don't need to have the skill to play instruments, but I need to know what I want to hear and how I want to hear it.

LLM as a Productivity Booster:

LLMs can't solve the 0 to 1 problem, but with proper guidance, they can handle executions better than humans.

Iteration is Key:

It took several rounds of prompting to get the final product.

Many things can go wrong, so multiple iteration is a must:

Background color too dark, making text unreadable;

LLM skipping the PDF and providing random numbers for the income statement; LLM not understanding long commands, requiring breaking down into smaller steps;

LLM sometimes forgetting previous steps, needing reminders to include all results

Next Use Case:

I'm exploring building a portfolio of growth stocks. I've finalized the thinking part and am currently working on prompt creation and iteration.

Your Thoughts:

I would love to hear your thoughts and experiences. Please let me know what you think and share any ideas or feedback you have. Your input is invaluable!