With over 5 years of experience in chatbot development, I’ve been building on top of GPT models since the early days of GPT2, when it was once considered too dangerous.
The introduction of ChatGPT piqued my interest, and I was curious to learn how people perceive this technology and how they plan to use it.
In order to gain this knowledge, I offered free development assistance to members of the OpenAI community in exchange for their project insights.
In this article, I’ll share what I’ve learned and the tools I’ve created to help others achieve their goals.
You can find the OpenAI thread here: https://community.openai.com/t/let-me-help-you-build-your-project-no-fees/27085.
Although ChatGPT is an amazing tool, most users expressed dissatisfaction with two key aspects. Firstly, ChatGPT only includes training data up to 2021, which implies that the model’s responses may be outdated.
However, since Microsoft is integrating GPT into Bing, this concern may not be significant for too long. Secondly, users face the challenge of being unable to use ChatGPT with their own data source.
Although ChatGPT has been trained on a large amount of data and has a general understanding of our world, it is difficult to train ChatGPT to prioritize its responses to custom data. What is the significance of this?
Assume you are a marketer who wants ChatGPT to write blog posts promoting your brand’s products and services.
Teaching ChatGPT unique aspects of your brand and including product details that may not be in their training data is not currently possible using OpenAI’s standard API.
Similarly, for chatbots to assist with answering customer questions, they must be knowledgeable about the company they represent, or else they will not be effective.
After speaking with various users across different industries and use cases, it became clear to me that all their issues could be resolved if they had their own version of ChatGPT.
To assist these users, I divided the challenge into two parts. Firstly, generating content on a large scale from their own knowledge base.
Secondly, answering questions for themselves or the people they serve using their own knowledge base.
Content Generation from Custom Data Source
To simplify content generation at scale, I created a Google spreadsheet where users can input their custom data, and the spreadsheet will automatically call OpenAI’s API to generate content in multiple rows.
You can download this sheet at https://community.openai.com/t/google-spreadsheet-gpt3/25460.
I chose Google spreadsheets because they’re user-friendly, and you can easily create workflows by chaining together multiple cells.
Here’s an example of how you can generate a lot of content using a GPT custom function in Google spreadsheets.
In one column, you can input the context, and in the next column, you can ask GPT a question based on that information.
This allows you to upload multiple rows of data in one column and receive responses for multiple rows at scale.
Additionally, you can chain multiple cells together and create a workflow of GPT calls from your spreadsheet. This is useful when you want to ask GPT a series of questions based on previous answers.
ChatGPT Answering on Custom Data Source
Many users have expressed interest in using ChatGPT as an extension of their own knowledge base by teaching it information that is currently unavailable to ChatGPT.
To help these users achieve this goal, I created a web app that allows users to upload their documents or attach web pages to their library and teach ChatGPT to answer questions using these references.
For example, you can attach an article about the LA Lakers trade news and use ChatGPT to reference it and get your questions answered. You can also view the references to see where ChatGPT sourced its answers from.
Furthermore, you can export your custom-built ChatGPT as a plugin and embed it onto your website so that others can use it as well. Think of it as a combination of Google Drive and OpenAI ChatGPT.
This web app is currently free to use, and you can register at https://superinsight.ai.
Utilizing Generative AI to understand data has become increasingly essential for individuals and organizations alike.
With the massive amounts of data being generated every day, traditional data analysis techniques are no longer sufficient.
Generative AI offers a powerful solution that can analyze and interpret data at scale, providing insights that would be impossible for humans to uncover manually.
By leveraging Generative AI, businesses can gain a competitive edge by making data-driven decisions faster and more accurately.
Here are two links to the tools discussed in this article