Namaste Yogis. Welcome to the Blockchain & AI Forum, where your blockchain and artificial intelligence technology questions are answered, mostly correct! Here no question is too mundane. As a bonus, a proverb is also included. Today’s question, was submitted by Samsara, from Puerto Rico, and she wants to know: what is ChatGPT and how has it evolved?
Samsara, you came to the right place. ChatGPT means Chat Generative Pre-Trained Transformer. No wonder it goes by ChatGPT instead! Ha Ha. Essentially, ChatGPT is a large language model-based chatbot and more. I’m confident you have encountered bots. Nevertheless, let’s define bot and chatbot.
A bot is an automated software application that performs repetitive tasks over a network and follows very specific instructions. Bots can perform only as programmed. There are various types of bots, including chatbots. A chatbot simulates human conversation using machine learning. A chatbot can respond to simple questions. Alexa from Amazon is a prime example. Now over to ChatGPT.
ChatGPT was created by OpenAI—an artificial intelligence (AI) research and deployment company. OpenAI’s mission statement is, “to ensure that artificial intelligence benefits all of humanity.” Sounds, virtuous. Do you remember when Google’s motto was “don’t be evil”? Ha, ha. Just saying! Now it’s time to examine the evolution of ChatGPT.
Before ChatGPT, natural language models existed. However, those models were trained on large amounts of annotated data for a particular task. Consequently, those models were strictly limited to what they were trained for and incapable of nothing beyond. To overcome those limitations, OpenAI proposed a generative language model built on unlabeled data thus allowing users to refine the model. The intent was to get the model to perform classification, question answering, and sentiment analysis. Essentially, ChatGPT1 was “trained” to attempt appropriate responses to questions based on accessible data. Although ChatGPT1 was a major technological upgrade, it tended to generate repetitive and generic responses. Your mobile phone device has evolved and so has ChatGPT.
In ChatGPT2, the amount of training data increased by 10X. Therefore, GPT2 generated far more coherent human like answers. Moreover, the processing speed jumped. However, the development of unsupervised learning was the great leap forward. With unsupervised learning, the model could “learn” and make predictions based on the vast amounts of data it was fed without the need for explicit instructions from humans. Holy!
ChatGPT3 was trained on an ever-growing mountain of data. Via the internet, GPT 3 accesses books, articles, websites, social media, etc. and consequently, has learned to “understand” and analyze language patterns. Additionally, GPT3 can perform tasks it was never trained to do. However, it tends to offer repetitive answers, is subject to bias, and consumes lots of power. Critics say it’s just a large memory machine that can spit out answers but isn’t any more intelligent than a calculator. Ouch, rather harsh!
ChatGPT4 is a multi-modal model, meaning it can handle text and images. That’s a first. With greater access to data and more training, ChatGPT 4 is much more creative and collaborative. Unlike its older siblings, GPT3 can generate text that are eerily human like. For instance, it can mimic a writer’s writing style. (Hey, don’t copy me!) Ha, ha. ChatGPT scored in the top 10% in the exam given to become an attorney and it’s multi-lingual in 26 languages! However, OpenAI warns ChatGPT4 makes up “facts” and thus be careful!
Samsara, I hope I answered your questions. Always remember the proverb from Laos—No one is so rich as to throw away a friend.
Until next time.
Yogi Nelson
Blockchain & AI Forum 