ChatGPT is probably one of the most innovative AI-driven chat and question-answering platforms developed by OpenAI. Released on November 30, 2022, ChatGTP has taken the world by storm. This is primarily because it is easy to use and is quickly becoming a tool to answer all sorts of questions and increase productivity.
ChatGPT uses a state-of-the-art, open-source dialogue model based on OpenAI’s GPT 3.5 AI engine. The model is trained in a fully unsupervised manner using human-human dialogue data from various sources. In this post, you will learn what ChatGPT is, who is behind it, what it does, and who can use ChatGPT.
Who Developed ChatGPT?
ChatGPT is developed by OpenAI, a non-profit artificial intelligence research company that was founded in 2015 by Elon Musk, Sam Altman, and Greg Brockman. It aims to develop artificial general intelligence in a way that benefits humanity, rather than benefiting one particular company or organization. Key products of OpenAI include DALL-E, GPT-3, GPT-2, OpenAI Five, and the ChatGPT user interface.
GPT-3 is an AI that performs a variety of natural language tasks. There are also a pair of Codex that translates natural language to code or explain code (think comments). DALL·E 2 is a new AI system that can create realistic images and art from a description in natural language.
Microsoft Investment in OpenAI
In 2019, Microsoft invested $1 billion in OpenAI. More recently, Microsoft has stated they plan to invest even more in OpenAI and the AI space in general.
In their press release, Microsoft said: “Microsoft is making this investment based on our conviction that AI will become increasingly significant over time. We also believe this is the right thing to do at scale as we bring together technologies across Office 365, Windows 10 and Azure.". As a matter of fact, Azure AI is developed in collaboration with OpenAI to bring AI to Microsoft’s Azure cloud.
In January 2023 Microsoft announced plans to integrate OpenAI’s engine with the Bing search engine experience. This includes image-generation OpenAI’s DALL-E 2 image service.
Microsoft already offers instances of GPT-3 and DALL-E 2 virtual servers on Azure. This means businesses and individuals can stand up private instances of these powerful AI tools.
Microsoft has also added OpenAI’s coding models as the GitHub CoPilot service. The service costs $100/year or $10/month and can integrate directly with several popular coding environments.
Microsoft is the first big brand to apply the AI technology to solving business problems. The ChatGPT tool has created buzz and now everyone is trying to decide how to leverage the tool. First, there are several questions we need to answer before determining practical applications.
Who can use ChatGPT?
ChatGPT is a free open-source framework for building conversational chatbots. It's available for anyone who wants to build a chatbot, improve their existing chatbot or create their own custom tool or solution.
When you visit ChatGPT or OpenAI.org you should create a new account. While the tool is open source there is a fee to use the OpenAI instance. You pay for tokens and those tokens are consumed as the AI model produces answers.
When you sign up your account starts with $18 (USD) worth of tokens. I have been using the tool for almost a month now and have just consumed $2 worth of tokens. So tokens can go a long way.
As I mentioned earlier Microsoft has pre-configured virtual machines available in Azure. This means you can create your own private instance of the engine. You would only need to pay for your compute time in Azure.
The practical applications are endless, and since anyone is free to use the tool it means everyone can benefit.
How does ChatGPT Work?
ChatGPT is a simple user interface where a registered user can open a chat thread and ask questions.
This is only a front-end to the AI models. OpenAI does have a public API, but the average person would not know how to use it. ChatGPT has given us a simple user interface that just about anyone can use.
You can see an example of how to use ChatGPT text completion API in C# ChatGPT Completions With C#. There are many other SDKs that abstract the API in other languages, like JavaScript (nodejs).
The basic workflow of ChatGPT involves providing the interface with a prompt or context. After you submit your query let the AI engine to generate text based on that input. For example, you might provide ChatGPT with a prompt such as "I'm thinking of starting a business selling handmade soaps. What should I consider before getting started?" ChatGPT would then generate text in response to this prompt, providing suggestions or advice about what to consider when starting a business selling handmade soaps.
I find the answers it provides can start me down several paths for more and more details. By the time I am done, I have a very complete answer and can start to execute.
The architecture of ChatGPT is based on the GPT transformer model, which consists of a series of layers that process the input data and generate output text. The model is trained on a large dataset of human-generated text, which allows it to learn the patterns and structure of natural language and generate text that is similar in style and content to human-generated text.
There are multiple AI models that drive answers. ChatGPT has been designed to triage inputs to determine which model will provide the best response. For example, if you ask it to write code it will use either the Davinci Code or the Cushman codex. For a text model, there are multiple choices. The quality of the answer and of course the cost associated with each model varies.
When you use the API you must tell the engine which model to use. This makes creating task-specific tools easier.
Working With ChatGPT
After trying a few questions, I must admit, ChatGPT is going to challenge Google search and other question-and-answer platforms such as Reddit, StackOverflow, and others.
The answers are often very detailed. They can also express some emotion, such as humor as well.
A word of caution about the answers. They may not be correct. The models are only trained on a dataset up to 2021. So current events are out.
One of my tests was to ask ChatGPT to write an article about an upcoming NFL game. It wrote the article based on the team’s records and stats from 4 or 5 years ago. Other tests have shown the engine can easily get confused about details.
So do not trust any answer as 100% true. Always fact-check the answer. This includes code. I find the generated code to be a combination of very simple and often riddled with bugs or obsolete APIs or coding techniques.
If the answer is something you plan on publishing, you should also edit the content. Savvy teachers and professors already have plagiarism detection tools. Many of these tools also include AI detection services.
I find using a grammar tool and some slight rewritings provide enough sanitation to make the content my own.
I currently use it as a tool to perform initial research for me. This can save me hours and hours of tedious tasks. Instead of wasting those hours doing the groundwork to create an article, write some demo code or just get an understanding of a new topic, OpenAI can perform those tasks in a few minutes. From there it is up to me to make it professional.
Conclusion
The ChatGPT model is open-source and available on GitHub. If you're interested in learning more about the project, check out our website where you can find more information and get involved!
Right now, the possibilities are wide open. Everyone is racing to find practical applications for this technology. This is where software developers can really excel. New tools, product extensions, and more that integrate OpenAI behind the scenes are where the tools are going to shine.
Join the conversation! Your thoughts help the community grow.