Skip to content

ChatGPT-4 Review

Source: https://www.wte.net/Blog/March-2023/ChatGPT-4-Review
Date: March 2023
Author: Martin W Smith

Opening

Martin shares his ChatGPT journey beginning in December, noting the platform has dramatically improved his Content Marketing Director role. He describes himself as "addicted" to the generative AI tool and paid for ChatGPT Plus ($20/month) without waiting for accounting approval due to mission-critical deadlines.

What is Generative About ChatGPT?

The post explains three phases of AI development:

  1. Supervised Learning: Human "labelers" teach the model what to prioritize
  2. AI Reward Model: Uses reinforcement learning where the AI receives feedback as rewards or penalties
  3. PPO (Proximal Policy Optimization): Speeds up responses for millions of simultaneous users without excessive computing costs

Martin notes this phase gets "lost" in technical complexity but essentially amplifies ChatGPT's ability to serve many users simultaneously.

How He Uses ChatGPT (Priority Order)

Tech Geek Translator: Martin relies on ChatGPT to translate technical jargon from colleagues at his software development company. He finds conversational AI responses five times more effective than Google searches for understanding complex topics.

Content Marketing Assistant: His workflow involves: - Clearscope for keyword research and grading - Grammarly for grammar and voice optimization - UberSuggest for meta descriptions and headlines

He emphasizes ChatGPT's outputs require significant editing before publication.

Idea Generator: Martin created "EcoFresh," a sustainable mouthwash concept using ChatGPT-3, then used DALL-E for images. He stress-tested the idea through Grammarly and Photoshop before any website publication.

ChatGPT-3 vs ChatGPT-4

ChatGPT-4 requires "30% less Grammarly" cleanup and produces "50% more robust" answers with greater depth. Martin hasn't yet explored multimodal features, prioritizing production deadlines over experimental capabilities.

Prompt Engineering

He emphasizes results depend entirely on prompt quality: "It's all about the prompts." Martin tests prompts against Google results to verify accuracy, maintaining healthy skepticism about AI biases and factual reliability.

Business Use Cases for WTE

Internal Search & Interactive FAQ: The company develops "Ask Eric," using ChatGPT to create interactive search experiences—something ChatGPT-3 would've required expensive API usage for, but GPT-4 shows promise.

Content and Social Media Marketing: Marketing roles will evolve to require ChatGPT proficiency. The post predicts algorithms will control content design, ad buying, and automated headline/image adjustments.

Code and Web Design: GitHub's Copilot enables faster coding. Martin emphasizes coders who master AI acceleration tools will become "priceless."

Can ChatGPT Become CEO?

The post questions whether AI can fulfill CEO responsibilities. While ChatGPT excels at strategic vision, Martin doubts its ability to inspire and motivate people effectively.

Guest Contributions

Chris Duke's Technical Use Cases:

  1. Zoom Transcript Summarization: ChatGPT condensed a 31-page meeting transcript to three pages of conversational analysis showing "who said what." The tool handles 4-page limits per query.

  2. Linux Administration: Chris used ChatGPT to debug grep commands and build bash scripts for firewall log analysis. He highlighted ChatGPT's helpfulness interpreting error messages, accelerating problem-solving.

  3. Future Python Project: Chris plans leveraging ChatGPT for Python script development to parse server logs into databases—work that previously required extensive research.

Phil Buckley's Translation Work: Phil successfully translated 1930s Canadian-French typewritten documents from scanned PDFs, noting ChatGPT-4 both translated flawlessly and corrected OCR errors (e.g., "l O 8 0" → "1680").

Conclusion

Martin invites reader feedback via subscription or email (martin at wte.net), emphasizing ChatGPT's transformative impact across marketing, development, and business operations while maintaining realistic expectations about AI limitations.