Building an App With the ChatGPT API: Lessons Learned
Since its release, ChatGPT has taken the world by storm. Seemingly out of nowhere, we had a computational system so effective that it could generate human language and dialogue good enough to pass expert evaluation tests like the bar or medical exams.
One of the reasons ChatGPT became so popular is the simplicity of the user experience. It allows users to access the power of artificial intelligence with a mere input field. Some experts even think that the technology is showing glimpses of artificial general intelligence, which means that these tools could learn from experience and apply knowledge and skills to new and unfamiliar situations.
Previously, most of the machine learning advancements required familiarity with programming, running on specialized hardware such as powerful GPUs and requiring complex installation steps. ChatGPT, conversely, offered a powerful alternative: it combined the power of a Large Language Model (LLM) with the simplicity of a conversation. It stands to be one of the fastest-growing apps ever.
However, it remains to be seen if a chat-based interface is the best way to interact with an LLM-based AI. Typing out repetitive questions can be time-consuming. And the linear nature of a chat could make it difficult to understand the big picture while creating different branches of conversation.
Table Of Contents
- The OpenAI APIs - GPT3.5-turbo model
- Overview of Large Language Models (LLMs)
- Weaknesses of Large Language Models & ChatGPT
- Tech Stack Used to Build an AI-Powered App
- Lessons Learned Building an AI App with the ChatGPT API
- Prompting techniques
- Four Ways Developers can use AI in Their Daily Workflows
- Final Thoughts
I think there are alternative ways one can engage with an LLM-based AI that would address some of these challenges. I built Heuristi.ca to explore one such alternative. Heuristica is an AI-powered knowledge exploration app. Instead of the prompt-reply method used in ChatGPT, a user simply clicks buttons to generate mind maps on any topic they want to explore. This method allows for the categorization and visualization of information in a more comprehensive way than linear, chat-based conversations.
In this article, I will share the learnings I had when building an AI-powered app using the OpenAI APIs, including how to decide on which model to use, as well as how I leveraged the AI itself during the whole development as a programmer.
The OpenAI APIs – GPT3.5-turbo model
To build Heuristica, I leveraged the OpenAI APIs. Using the APIs, you can perform a range of operations, such as text completion, chat completion, and image generation, using various kinds of available models.
