ChatGPT, like its “sibling” InstructGPT, is an artificial intelligence (AI) model. As such, ChatGPT is a program or an algorithm that uses a dataset to recognize specific patterns. So, when given enough information, often a massive amount of data, ChatGPT can reach a conclusion or make a prediction.

What Is ChatGPT?

ChatGPT, as mentioned above, is an AI model released on 30 November 2022. Specifically, it allows chatbots to interact with users conversationally. Since it uses the dialog format, ChatGPT can answer follow-up questions, admit mistakes, challenge incorrect premises, and reject inappropriate requests. In a sense, it’s a more intelligent chatbot in that talking to it is just like talking to an average person that instantly reacts to what users say.

Who Created ChatGPT?

OpenAI is ChatGPT’s developer. OpenAI, established on 11 December 2015, is an AI research and deployment company. According to its website, it “aims to ensure that artificial general intelligence benefits all of humanity.”

Since its founding, it has launched AI software and toolkits, including OpenAI Gym Beta and Universe. It’s backed by several of the world’s biggest tech companies, such as Microsoft.

Why Is ChatGPT More Advanced Than InstructGPT?

InstructGPT is an AI model that follows instructions instantly to provide detailed responses. It enables predictive text, such as when you type a search term into your web browser. In most cases, you don’t have to finish typing your entire question. Suggestions that complete your queries appear as drop-down choices. To get results, all you need is to select the text that corresponds to what’s on your mind.

ChatGPT goes beyond in that it provides suggestions to complete statements and seemingly lets computers talk to you.

How Does ChatGPT Work?

Like other chatbot models, interacting with ChatGPT begins by asking it a question. Unlike most models, though, ChatGPT can solve complex problems like composing a song, writing about a given topic, fixing your computer code, and so much more.

Anyone can try using ChatGPT from its website. On it, you can also see how ChatGPT can help a developer fix errors in his code.

ChatGPT example

Note that ChatGPT’s response is just like how a more senior developer might answer his mentee. It asks for more context to give the user a more accurate answer.

What Are the Uses of ChatGPT?

Apart from chatbots, companies can potentially use ChatGPT to interview potential employees. Take the AI chatbot Olivia developed by Paradox as an example. It can effectively screen, interview, and hire a McDonald’s employee.

Other real-world use cases that have been demonstrated so far include the following:

  • A bank compliance chatbot that operates as a chief compliance officer
  • An AI system for human resource (HR) research and professional development
  • An AI system that can be made to take existing courseware and automatically create quizzes
  • A virtual teaching assistant that can create course outlines and learning objectives
  • An AI system for recruiting, onboarding, sales training, manufacturing training, compliance training, leadership development, and personal and professional coaching

Is ChatGPT Foolproof?

Although the world seems to be raving about ChatGPT, like any other technology, it has several limitations. Even its creator OpenAI acknowledges that ChatGPT “sometimes writes plausible-sounding but incorrect or nonsensical answers.”

In the world of AI, this behavior is not uncommon and is called “hallucination.” Why does it happen? First, you have to know that any AI model’s reward model can be overoptimized, thus hindering performance. A reward model refers to how emotions and other behaviors are grouped. Overoptimization, in this case, can occur when you include “happy” in, say, the positive group and “sad” in the negative group. If the model is programmed to react only positively or negatively, then it can’t be both simultaneously. And we know that’s not really the case for humans. Despite being sad overall, for instance, likely due to a death in the family, you can still feel happy that lots of people remembered your recently departed loved one.

Also, ChatGPT has limited knowledge of events that occurred after 2021 due to its dataset. Other issues include:

  • ChatGPT exhibits a pro-environmental, left-libertarian orientation when prompted to take a stance on political statements.
  • ChatGPT’s human reviewers preferred longer answers even if it’s not as factual.
  • ChatGPT’s training data may also suffer from algorithmic bias that may affect how it describes people.

In the Future

While OpenAI may have pioneered this new technology, many will swoop in to innovate on their idea. ChatGPT will surely be advanced, extended, and refined, and quite soon.