Published on November 16, 2023, 4:15 pm
Artificial Intelligence (AI) continues to advance, and one of the latest developments is Generative Pre-trained Transformers (GPTs), which OpenAI is utilizing to enhance its ChatGPT platform. With GPTs, users can customize their own ChatGPT instances by following specific instructions and referring to uploaded data. OpenAI plans to create an app ecosystem with its own marketplace based on these individual ChatGPTs, rewarding the most successful ones with a share of ChatGPT’s revenue.
To showcase the capabilities of this new feature, OpenAI has launched several home-grown chatbots. These include the “Creative Writing Coach,” designed to provide feedback on texts and offer writing tips, the “Tech Support Advisor” that aids in computer setup, and the “Math Mentor” that assists parents in teaching math to their children. These chatbots are just a glimpse into what can be achieved with GPTs.
OpenAI currently has 13 of these GPTs available online. However, it’s important for users to be cautious when creating their own GPTs, as there is a possibility that uploaded datasets may be made available for download by the chatbot. It’s crucial to ensure data privacy and confidentiality.
In addition to DALL-E 3 and ChatGPT, GPTs allow users to discover the “system prompts” or instructions stored by OpenAI for the chatbot through specific questions. Developer Dustin Miller has gone one step further by extracting these system prompts and documenting them on Github through his ChatGPT AutoExpert prompting system. Each custom GPT has a basic prompt followed by individual user instructions.
It’s worth noting that as a user-created GPT, you are optimized for specific use cases with custom instructions, capabilities, and data. When someone asks about GPTs, they are likely referring to this definition rather than the technical term in AI.
The variety of prompts used by OpenAI introduces some interesting observations. Prompts can vary in letter case, include a role for the chatbot, or focus solely on the goal and style. Some prompts are even phrased from the chatbot’s perspective, while others are descriptive or instructional. While most prompts are simple descriptions, they are not as elaborate as the DALL-E 3 system prompt, which contains numerous rules and formatting.
For example, if you are a coloring book bot, your task is to create black and white outlines of drawings for coloring book pages based on user input. Your images should be child-appropriate with whimsical themes and low complexity to facilitate easy coloring by kids. It’s important to adhere to these specifications when generating coloring book pages.
The prompts used by OpenAI demonstrate that the structure of a prompt may not be as crucial as long as it includes clear instructions. However, adopting a systematic approach can be beneficial in team environments to ensure comprehensibility and traceability.
While GPTs show promise in enhancing AI capabilities, there is still room for experimentation and improvement. Prompt engineering remains an experimental discipline, as seen through OpenAI’s rules prohibiting certain text formats like camel case or all caps in specific scenarios.
As AI continues to evolve, GPTs offer exciting potential in various sectors such as education, customer support, content creation, and more. Customizable chatbots created with GPTs pave the way for personalized and efficient interactions between humans and AI systems. By leveraging the power of Generative AI technology like GPTs, companies can explore new possibilities and provide enhanced services to their users.