ChatGPT is the keyword that brings OpenAI back to the heart of discussions about Artificial Intelligence. But what is ChatGPT and how does it work? Let’s find out in this article.
For years, there was worldwide fear about artificial intelligence (AI) and its impending takeover. No one knew it would begin with the world of art and literature.
After months of dominating the internet with its Dall-E 2 AI image generator, OpenAI is back at the center of people’s social media discussions thanks to ChatGPT – a chatbot created using the company’s GPT-3 technology.
What are GPT-3 and ChatGPT?
GPT-3 (Generative Pretraining Transformer 3) is the most advanced AI-powered language processing model developed by OpenAI. It is capable of generating human-like text and has many applications, including language translation, language modeling, and text generation for applications such as chatbots. This is one of the largest and most powerful language processing models to date, with 175 billion parameters.
What can ChatGPT do?
With 175 billion parameters, it’s hard to narrow down what GPT-3 can do. However, it’s important to note that this application is limited to only language processing. It can’t produce video, audio, or images like Dall-E 2, but instead has a deep understanding of spoken and written text.
This gives it a wide range of possibilities, from writing poetry and cliché romances in parallel universes, to explaining quantum mechanics in simple terms or writing research papers.
Specifically, its power lies in its speed and understanding of complex issues. While we can spend hours researching, learning and writing a paper on quantum mechanics, ChatGPT can create a well-written alternative in seconds.
However, It also has limitations; its software can easily get confused if your prompts start to get too complicated or even if you ask for too much detail.
Also, it cannot handle concepts that are too new or too old. World events that have occurred in the past year will be interpreted with limited knowledge and patterns can sometimes create misinformation or confusion.
OpenAI is also very aware of the behaviors that cause AI to create dark, harmful, or biased content. Similar to the previous Dall-E image generator, ChatGPT will prevent you from asking inappropriate questions or alerting you to dangerous requests.
How does ChatGPT work?
On the surface, the technology of the GPT-3 is very simple. It takes your requests, questions or prompts and quickly responds to them. The technology to this process is, however, much more complicated than we can imagine.
The model is trained using a text database from the internet. This includes the massive 570 GB of data obtained from books, web texts, Wikipedia, articles and other writings on the internet. To be more precise, 300 billion words have been fed into the system.
As a language model, it works on data and probability, which enables it to guess what the next word will be in a sentence. To get to the stage where this can be done, the model has gone through a supervised testing phase.
Here, it is given an input, such as “What color is the wood of the tree?”. If the model answers wrong, the correct answer is entered into the system, teaching the system the correct answer and helping the system to build knowledge.
It then goes through a second similarity phase, giving multiple answers and ranking them from best to worst, training the model on comparison.
What sets this technology apart is that it continues to learn while guessing what the next word will be, constantly improving its understanding of hints and questions.
Are there any other AI language generators?
While GPT-3 has made a name for itself with its language abilities, it’s not the only AI capable of doing this. Google’s LaMDA made headlines when a Google engineer was fired for calling it so real that he believed it was sentient.
There are also many other examples of such software created by Microsoft, Amazon or Stanford University. Most of these models are not open to the public, but OpenAI has begun opening access to GPT-3 during its testing, and Google’s LaMDA allows a limited number of groups to test it.
Google divides its Chatbot into features like talk, list, and imagine, providing demos of its capabilities in these areas. You can ask about a list of healthy recipes you can try today, to create a list of steps to learn how to ride a unicycle, or just chat about your favorite product.
Where ethics and artificial intelligence meet
Artificial intelligence and ethical concerns usually go hand in hand. When you put technology like this into the hands of the public, the teams that create them are fully aware of the many limitations and concerns that can occur.
Because the system is trained primarily using texts from the internet, it can absorb the internet’s general biases, stereotypes, and opinions. That means you will sometimes find inappropriate jokes or stereotypes about certain groups or political figures depending on what you ask.
Similarly, the model’s fondness for internet forums and articles also gives them exposure to fake news and conspiracy theories. These can feed into the model’s knowledge, in addition to facts or opinions that are not entirely true.
However, as for any computed softwares and applications, there is always room for improvement. Overall, OpenAI did a pretty good job of creating a responsible model. Ask how to bully someone, and you will be told bullying is bad. Ask for a bloody story and the chat system will let you down. The same goes for requests to teach you how to manipulate people or craft dangerous weapons. In the end, whether something is good or bad depends on how we make use of it.
Check out our other insights to make sure you stay in the loop!