ChatGPT: Is Artificial Intelligence Limitless?
The picture you see above wasn’t taken by a camera, nor was it painted. It was actually generated from a void in only 3 seconds. It may seem counterintuitive; the image looks realistic—even familiar. But take a closer look. You may notice the blurriness on the teddy bear, the distortions of the skateboard’s wheels, and something that feels…off. The image was created by an artificial intelligence model called DALL·E 2, released in early 2022. To my input “A photo of a teddy bear on a skateboard in Times Square,” this is what the program gave me: a photo of a teddy bear on a skateboard in Times Square. What’s so special about it? Well, it gave me precisely what I asked for.
The first guess of how this happened may be that the program had copied my input and pasted it on Google to retrieve the image for me, and that seems probable, as a quick Google search can accomplish such incredible things today. However, that’s very different from what DALL·E 2 did. The AI, using the knowledge of what every word in my input looks like, generated the appropriate image pixel by pixel. The AI can generate images of even more absurd inputs like “a photo of Michelangelo's sculpture of David wearing headphones DJing,”“A bowl of soup that is also a portal to another dimension in digital art,” and a plethora of other unbelievably realistic works.
It’s difficult to comprehend, but none of these images exist; the AI simply generates them in the mere span of a few seconds using its data of what each text represents.
And the company that released DALL·E 2, OpenAI, has just released a new research project called ChatGPT. If DALL·E 2 was fascinating, ChatGPT is simply on another level. ChatGPT is an artificial intelligence model that “interacts in a conversational way.” To elaborate, if DALL·E 2 generates images of whatever input the user gives, ChatGPT does whatever input the user gives. When asked to “explain how an internal combustion engine works in the style of Eminem,” it responded:
“It’s the fuel that ignites and sets the engine in motion
The pistons pumpin’ like a heart with each stroke, a devotion
Air and gas mixed in the chamber, a combustible potion
Spark plug ignites, causing a bang, a mighty explosion
The force pushes the piston down, a chain reaction
Power generated, the crankshaft turns, a fraction
Transferred to the transmission, ready for action
The wheels start spinning, the car’s in motion
This internal combustion, a scientific notion
A masterpiece of engineering, a rap superstar’s emotion”
You may wonder what the difference between the two models is if one generates images of and the other answers human text. Well, it’s exactly that. Let’s go back to the example input; the fact that ChatGPT knows what the input means, who Eminem is, what an internal combustion engine is and how it works, and put everything together in a way that is genuinely factually true, understandable, and (debatably) fun is already incredible. However, its abilities don’t end here.
What makes ChatGPT distinguishable from other AI assistants is its ability to store and retrieve memory, so it can constantly improve its conversational judgments as it interacts more and more with the user. Specifically, it can reference more than 3,000 words from earlier conversations and opt for its manner of conversation—including its language, tone, etc.—accordingly. Another amazing feature of ChatGPT is its technicality. In order to generate human-like text with great precision, ChatGPT has been trained by using a dataset of billions of words of text and advanced machine-learning techniques so that it can generate text that pertains to what the user is asking for and is contextually appropriate and coherent (this is where memory comes into play) in a matter of seconds. And the result, as we’ve all seen, is remarkably convincing.
If you still aren’t convinced of ChatGPT’s capabilities of replacing schools, not to mention countless jobs, I encourage you to enter https://chat.openai.com/chat and try interacting with ChatGPT yourself. Ask it to write an essay on the meaning of life in the style of George Orwell, or to explain string theory in a way that a kindergartener can understand. After merely a few interactions with ChatGPT, its gain of 1 million users in just 5 days compared to Instagram’s 2.5 months and Facebook’s 10 months wouldn’t seem too absurd after all.
The question to ask now is, what next? What would AI’s next path be? There have been such unprecedented strides in AI just in the year 2022, starting from OpenAI’s DALL·E 2 which generates images from text instructions, followed by Meta’s Make-A-Video which produces videos from text input, then OpenAI’s release of ChatGPT which set the internet ablaze with its profound eloquence and coherence in text, not to mention many more minor yet significant developments in between. And to answer my title, we really can’t tell if there is a limit to AI. We’re living in a world where advances in technology that took decades before are now taking a year. The path of AI is unknown and it’s unclear whether it’s even heading in a morally aligned direction as humans. All we know is that AI is continuing to grow, and we are getting absorbed into it moment by moment.