Introduction
Since my first column about ChatGPT on December 9th last year, more than 100 days have passed. Despite the service not yet being available in Hong Kong, local users continue to show enthusiasm in learning about new technologies. Data reveals that since last December, there have been 33,000 discussions about ChatGPT in Hong Kong, demonstrating a gradual increase in interest. Today, we will explore some misconceptions about ChatGPT and discuss how to make the most of its capabilities.
Figure 1：ChatGPT: 33k discussions in Hong Kong since Dec 2022, less than 5.2M globally, but interest rising.｜ Source：TOCANAN
Misconceptions and Tips
1. Poor Search Capability: Unlike search engines, ChatGPT is a conversation-based AI model designed to generate human-like natural language responses, not provide search results. Users should consider ChatGPT as a tool for simulating human responses.
2. Highly Intelligent, Unbiased, and Error-Free: Although ChatGPT showcases impressive text generation abilities, it is not a highly intelligent system and has limitations. We should understand these limitations and use ChatGPT as an auxiliary tool, relying on diverse information sources for accurate answers in scenarios requiring human judgment and expertise.
3. True Understanding: While ChatGPT can produce coherent responses, it does not truly understand language and may sometimes generate inaccurate or irrelevant answers.
By adjusting our expectations and using prompt engineering techniques, we can obtain better answers from ChatGPT.
Efficient Applications
During my 100+ days of interaction with ChatGPT, I’ve identified several applications for improved efficiency:
1. Email Drafts: ChatGPT can generate email drafts and responses, saving time. Users simply input key information, and it will produce a professional, clear email without extensive editing and revisions.
2. Creative Ideas and Insights: ChatGPT can provide users with novel ideas and insights, stimulating creativity and innovation while offering unique perspectives and suggestions to teams.
3. Content Editing and Optimization: For tasks such as writing reports, articles, or marketing materials, ChatGPT offers text generation and editing support. Users provide topics and keywords, and it will generate a draft for further modification. With prompt engineering, users can issue precise instructions to ChatGPT for optimized results.
4. Learning Support: ChatGPT assists with learning programming languages and foreign languages, enabling users to solve problems quickly and accurately. This not only meets users’ skill development needs but also expands their knowledge domains.
Figure 2：OpenAI President Greg Brockman demonstrates image-guided tasks, completing HTML and JavaScript coding for websites within seconds.｜ Source：OpenAI YouTube
GPT-4: A New Era of AI
On March 14th, OpenAI released its latest AI model, GPT-4. This model accepts both image and text inputs and outputs text. OpenAI President Greg Brockman demonstrated image-guided tasks, like writing HTML and JavaScript code for websites in seconds. GPT-4 exhibited human-level performance on various professional and academic standards, excelling in simulated bar exams, SAT reading, and math tests, placing in the top 10%. However, OpenAI warns that GPT-4 still suffers from “hallucinations,” generating fictional information that appears credible but is unreliable.
Conclusion
By adjusting our expectations and employing prompt engineering techniques, we can effectively utilize ChatGPT as an innovative tool to enhance our work efficiency and learning capabilities. With GPT-4 on the horizon, the future of AI-assisted productivity holds even more promise.
Book a meeting to let us know how we can help.
