Written by Jason Hollander:
Introduction
On November 30, 2022, OpenAI, an AI research and development company, released ChatGPT to the public. Ever since, this new technology has taken the tech world by storm, and many have been speculating about how advanced AI has become and what is next for our computer-operated world.
What is ChatGPT?
In simple terms, ChatGPT is an extremely advanced chatbot. It uses a sophisticated AI language processing system to understand the context of messages, generate human-like responses, translate texts, write code, and more.
GPT stands for Generative Pre-training Transformer. Generative refers to how the model can generate text. Pre-training relates to how the model has been trained on massive amounts of data. Using Natural Language Processing, ChatGPT pulls data from sources such as online databases, textbooks, websites, and various articles, which is used to generate a response to the prompt.
How does ChatGPT work?
ChatGPT uses AI deep learning techniques to pull data from the web to generate answers. It specifically uses a variant of a “transformer architecture,” which is capable of reading several terabytes of data containing billions of words. It then processes the input text and predicts the most likely next word in the sequence, based on its immense catalog of data.
The training model for ChatGPT is quite complex. ChatGPT mainly uses Supervised Learning and Reinforcement Learning techniques to train the model. There are three main steps in its training. First, it collects data and implements Supervised Fine Tuning (SFT). A prompt is presented and a coder assists in the response along with the response of ChatGPT. Next comes the reward system. The purpose of this is to score the response of ChatGPT based on the quality of the answer. This will allow the AI to learn what responses are best and teach itself for future responses. The goal is for the AI to choose responses that humans would like best. Third, the program fine-tunes the SFT model via Proximal Policy Optimization (PPO). PPO is a specific reward learning reinforcement system that allows ChatGPT to update its policy based on the score of the response.
In short, ChatGPT uses the combination of machine learning and human trainers to provide a reward system based on the quality of answers and updates the program based on the positive feedback it recieves.
What can ChatGPT do for me?
ChatGPT is capable of answering questions, writing code, writing stories/poems, writing emails, solving math problems, and more. It can be a great resource as a virtual assistant as a researcher, coder, client services role, emailer, and more.
ChapGPT has accomplished some incredible feats in the academia world. Recently, ChatGPT passed law exams in four courses at the University of Minnesota and another exam at University of Pennsylvania’s Wharton School of Business, though it did not score a 100%.
The faculty in QMSS warn that we should do our own homework and write our own papers during our college career! This is how we learn and avoid plagiarism charges! (This point added by one of the QMSS faculty).
What are the limitations?
While ChatGPT is capable of digesting trillions of data points and generating very accurate answers, it does not have the ability to understand like a human can. The model is very limited in comprehension, reliability, and range. ChatGPT can also generate false information, toxic or biased outputs, and offensive responses due to the amount of such content on the internet. It remains extremely important that ChatCPT and similar AI models require human supervision.
Will it take my job?
No, not yet… but maybe someday. ChatGPT’s ability to generate stories that are human-like has definitely startled me and could threaten my role as a student writer.
ChatGPT will likely disrupt jobs that are repetitive or consist of routine tasks that can be automated. These jobs include data entry, customer service, content creation, etc. While ChatGPT has the potential to eliminate a lot of jobs, it also has the potential to create a lot of jobs. AI technology like ChatGPT will not be the replacement of human jobs but rather a complement to the work we do. According to The World Economic Forum’s Future Jobs Report from 2020, 85 million jobs will be lost due to AI by 2025, but another 97 million jobs will emerge.
What’s next?
It is important to remember that only ChatGPT 3.5 has been released. It can do all of this after only three major updates. This is only the beginning. As the 3rd model of ChatGPT is now open to the public, the chatbot will only get more advanced. The explosion of usage will allow the developers to fine-tune even more.
OpenAI is not alone. Competitors such as Amazon, Google, Facebook have already created their similar chat bots, but they remain closed to the public, unlike ChatGPT. With this spark of AI and a vision of how much impact this technology can have, investors and businesses are racing to become the leader in artificial intelligence.