Explain Yourself, ChatGPT
If you aren't already afraid of AI taking over the world, then you probably shouldn't read this.
ChatGPT (Chat-Generative Pretrained Transformer) is a nightmare turned reality. Last November, OpenAI, a San Francisco-based artificial intelligence company, released a groundbreaking AI-system that is about just as scary as its name suggests. The tool is considered to be a language model—it uses context and probability to generate the most likely next set of words—but it operates like a search engine. Because ChatGPT is able to pull information from a massive dataset, it can answer virtually any question you ask it12.
Here is one of my first interactions with the system:
Me: Is it safe to take tylenol every day?
ChatGPT: It is generally safe to take Tylenol (acetaminophen) as directed on the label or as advised by your healthcare provider. However, taking Tylenol for extended periods of time or at high doses can increase the risk of certain side effects, such as liver damage. If you are considering taking Tylenol on a daily basis, it is important to discuss this with your healthcare provider first to determine if it is appropriate for you and to determine the appropriate dosage and duration of use.
In just four sentences, ChatGPT provides a much clearer and more helpful answer than the list of search results generated by Google.
ENDLESS APPLICATIONS
ChatGPT is different from virtual assistants like “Siri” and “Alexa” because it also has the ability to accomplish real-world tasks. Since being released to the public for free, millions of users have signed up and begun using the system’s software to do things like write them stories and research papers, develop personalized meal-plans or safety protocol instructions, or even just provide them with some company for awhile.
Unsurprisingly, the tool has created chaos within the classroom. Just last week, the largest public school system in the country—NYC public schools—blocked internet access to ChatGPT citing “concerns about cheating and learning development.”3
But why take my word for it when ChatGPT can explain itself?
Me: Can you explain what you are?
ChatGPT: I am a language model trained by OpenAI. I am a computer program designed to process and generate human-like text. My primary function is to assist users by providing information and answering questions to the best of my ability based on the dataset and knowledge that I have been trained on. I do not have personal experiences or feelings, and I am not a real person. I exist solely to provide assistance and information to users through the use of natural language processing and machine learning techniques.
Me: How do you generate text?
ChatGPT: As a language model, I generate words by using the statistical relationships learned from a large dataset to predict the most likely next word or sequence of words given a prompt or context. This process is known as “language generation.”
To generate words, I use a combination of machine learning techniques and natural language processing algorithms to analyze patterns and relationships in the text data that I have been trained on. Based on this analysis I am able to generate text that is similar in style and content to the text that I was trained on, while also attempting to produce output that is coherent, grammatically correct, and flows naturally.
Okay, so there is totally something unsettling about a self-described “non-sentient,” non-human machine referring to itself with the word “I." The system also seems super keen on emphasizing the importance of itself not being a living entity…totally not creepy at all.
THREE-STEP TRAINING PROCESS
You may be surprised to learn this, but computers do not understand English. In fact, they do not speak Spanish, French, or German either. Computers only follow commands that are written in 1s and 0s and in sequences of 1s and 0s. ChatGPT and other similar language models, chatbots, and virtual assistants are highly dependent on natural language processing (NLP) technology—which is the mechanism that allows computers to translate words and letters into comprehendible 1s and 0s and facilitate communication with their "masters."
This is how I envision the updated version of ChatGPT to look in 10 years.
The team at OpenAI took ChatGPT’s NLP technology to the next level. To train the model, “human AI trainers” were used to model representative behavior.
First, the trainers play both sides of the dialog—the system and the user. They begin by writing responses to example prompts and demonstrating the system’s desired behavior. Next, the computer logs their responses. It studies the content and analyzes the word-patterns within it before attempting to replicate an adequate response in a number of its own outputs.
From there, the AI trainers rank the quality of each ChatGPT response/output. They prioritize sentence structure and the use of proper grammar and rank the ones that most closely mimic human conversation higher4. Finally, the system uses the ranked responses to elicit more desirable behavior in the future.
CONSEQUENCES AND PREDICTIONS
Ultimately, ChatGPT represents a massive step forward in efficient NLP processing and our ability to communicate with NLP-dependent systems. The model is surely one of the most intensely-trained AI-systems that exists today, but it is almost certain that the time spent on future artificial intelligence systems will dwarf the time spent on this model.
It is not difficult to imagine how the use of ChatGPT’s technology within something as realistic and intimidating as a deepfake could wind up being extremely consequential. ChatGPT is one of the first systems with the ability to do things like use context, recognize previous conversations, and clarify its own responses. But if that technology is implemented into something as lifelike as the AI-impersonated video of Barack Obama directly below this paragraph (and it will be), then the opportunities for deceit and malice will become virtually endless.
A deepfake is a realistic piece of synthetic media.
On a lighter note, I believe that ChatGPT could actually be the first step towards having superhero-like automated assistants become actively used in everyday life. A simple voice-to-text translation would allow ChatGPT’s technology to be put into an AI-system like Tony Stark’s “JARVIS” or the AI-assistant named “Thalia” that aides and abets navigation of Logic’s spaceship on his album “The Incredible True Story.”
If you aren’t a Logic fan, you are wondering why I am talking about him—but if you are, then you will know that one of his most famous tracks, “Fade Away,” simulates a very cool potential application of ChatGPT’s state-of-the-art NLP capabilities—should they be improved upon and transferred into a physical assistant or robot.
Quentin (Q) and Kai (K) (the ship’s commanders) are discussing Thalia’s (T) (the AI-system) new interface.
Q : I told you man, the upgrade is crazy. Thalia—revert last command and run simulation.
T (in feminine-robot voice) : Running simulation.
K : What the hell is that?
Q : You can pick anyone in history and, based on a collection of data, it allows you to personally speak with them.”
K : Anybody?
Q : Anybody.
K : Thalia, do Big Sean (another rapper featured on the album)
Q : Alright, let me get this straight. You could speak with anyone in recorded history, like Jesus, Steve Jobs, Einstein, and you pick Big Sean?
K : Man, f*** yeah! That’s my dawg!
Q : Okay. Thalia, simulate Big Sean.
T (in Big Sean’s voice) : Whadup Kai! You lil b****
K : Yeah. Imma have to upgrade my s***.5
Soon, our robots will be assisting us with nearly all of our day-to-day activities. Think of how helpful features like real-time language translation or infrared and ultraviolet image processing would be. After meeting someone new or traveling to somewhere you have never been before, you will no longer need to recall people’s faces, their names, or the landmark you pass by because your assistant will undoubtedly be equipped with facial and geographical recognition and memory. All you will have to do is ask your AI-assistant, who will surely be given a witty name, to “tell my car that I want to go to the supermarket,” or “tell me who I was having that conversation about Mark Wahlberg with a few days ago.” In time, your robot may become your best friend.
One of the most impressive features of ChatGPT is its ability to script computer code. The system is now capable of providing the instructions needed to build other web tools, such as private web browsers, customized business websites, and even fully-functioning games, apps, and armies of angry Twitter bots.
In the snippet above, ChatGPT (allegedly) directs an ironic and sinister film about a post-apocalyptic world controlled by AI-systems.
In this tweet, ChatGPT instantaneously produces a short story about MLB-superstar Shohei Ohtani
In this series of images, ChatGPT has written computer code to create an app that allows its user to create different shapes on their smartdevice’s camera.
A YouTube channel with the name “ChatGPT Tutorials” has gained 30k+ subs in less than 3 weeks and features videos with names like “Build Notepad App using ChatCPT in less than 1 minute” and “Build a BrickBreaker Game using ChatGPT.”
ChatGPT’s dataset has a cut-off date and can’t tell you about anything after September 2021—Sudhakar Ramasamy published on dev.to https://dev.to/sudhakar3697/fyi-chatgpts-knowledge-cutoff-date-is-september-2021-2f5f
(ChatGPT) “is a variant of the popular GPT-3 (Generative Pertained Transformer 3) model, which has been trained on a massive amount of text data to generate human-like responses to a given input.”—Eeman Majumder published in “Geek Culture” on medium. https://medium.com/geekculture/chatgpt-what-is-it-and-how-does-it-work-exactly-62e7010524d3
“ChatGPT Banned in New York City Public Schools Over Concerns About Cheating, Learning Development”—Alyssa Lukpat published in WSJ https://www.wsj.com/articles/chatgpt-banned-in-new-york-city-public-schools-over-concerns-about-cheating-learning-development-11673024059
Or so I assume—there is very little info about training methods and what kinds of specific behaviors were chosen as positive or negative.
“Fade Away” queued up to referenced scene: