A Stormy Night in Paris

It was a stormy night in Paris. The rain pounded against the old cobbled streets, creating a cacophony of noise that echoed through the city. The sky was dark, with only a few stars twinkling in the night sky, and the wind was howling. Danielle stood outside an apartment building, her long, dark hair blowing in the wind. She was shivering against the cold, wearing a thin coat that did little to protect her from the bitter chill. She had been out late at a friend’s house, and now she was faced with the task of walking home in the pouring rain. She cursed herself for not taking a taxi, but she had been too eager to get back home.

 Danielle pulled her coat tighter around her and started walking. The cobbled streets were slippery, and she had to be careful not to slip and fall. As she walked, she thought about her life. She was twenty-five years old and had just moved to Paris a few months ago. She was desperate to make a new start, to find a job and make some friends. It had been hard so far, but she was determined to make it work. Suddenly, Danielle heard a voice coming from down the street. She stopped and looked around, but there was no one in sight. She was about to keep walking when she heard the voice again. It was a deep, masculine voice, and it sounded like it was coming from a nearby alleyway. Danielle hesitated. She wasn’t sure if it was safe to investigate, but she was also curious. She slowly walked towards the alleyway, her heart racing as she did.

 When she reached the entrance, she peered inside and saw a man crouched in the shadows. He was wearing a long black coat and a fedora, and he was holding a violin. The man saw Danielle and smiled. “Hello there,” he said. “It’s a bit late for a young lady to be out on a night like this, isn’t it?” Danielle was taken aback. She hadn’t expected to encounter anyone in the alleyway, least of all a stranger. She was about to turn and leave when the man spoke again. “I’m sorry for intruding,” he said. “I was just playing my violin. Would you like to join me?” Danielle hesitated. She wasn’t sure if she should trust the stranger, but there was something about him that made her feel at ease. She nodded her head and stepped into the alleyway.

The man introduced himself as Pierre. He told Danielle he was a musician and had been playing his violin in the alleyway for years. He said the sound of the strings helped to soothe his soul on nights like these. Danielle listened, captivated by his story. She asked Pierre to play her a song, and he obliged. His music filled the alleyway, and Danielle felt as if she was in a different world. She closed her eyes and let the music wash over her, feeling her worries and fears drift away. When the song ended, Danielle thanked Pierre for the music. He smiled and told her to come back anytime, and she promised she would. Danielle walked away, a smile on her face. She felt lighter now, and her worries seemed to have vanished in the music. As she walked back to her apartment, she promised herself she would come back and visit Pierre soon, and she fell asleep with a smile on her face.

 ********

The story above is decent writing although not totally realistic but probably would get a passing grade in a high school English class. What makes it impressive is it was written in total by an artificial intelligence program called GPT3. My only contribution was the first sentence. This example is just a very small sample of what it can do.

GPT3 is a large language model. What is that? I will let GPT3 tell you:

A large language model is a type of artificial intelligence (AI) that is trained to generate human-like text. This is typically done by training the model on a massive dataset of text, such as books, articles, and other written works. The goal of a large language model is to be able to generate text that is similar to human writing in terms of style, tone, and content. This can be used for a variety of purposes, such as generating news articles, writing creative fiction, or even creating chatbots that can hold natural conversations with people.

In addition to the above, GPT3 can generate computer code and analyze existing code for errors.  It can write poetry in the style of a famous poet or compose music. It is nothing short of amazing! In December of 2022, the company responsible for GPT3, OpenAI, released an enhancement called ChatGPT which improved the conversational ability of the model and raised public awareness of what it could do.

By the way, the picture at the beginning of this article which looks like a photograph of a woman walking in the Paris rain, is not a real photo. It was generated by an artificial intelligence program called Stable Diffusion. I gave it a short description of what I wanted, and it generated the image. Stable Diffusion is a text-to-image model that was released in 2022. There are several other competing models like it. Below are two other images I generated using Stable Diffusion:

Very soon it will be possible to generate videos with this technology and that will likely have an impact on the entertainment industry. It’s already generating controversy in the art world. The picture below took first place in the Colorado State Fair for digital art in September 2022. It was created by a program similar to Stable Diffusion called Midjourney. Real artists were not happy. Is this art even though it wasn’t created by a human being?

Théâtre D’opéra Spatial

There are other issues. When you can generate almost any image with a few keystrokes, you can create all sorts of undesirable content although the programs themselves try to limit what you can do. Stable Diffusion is open source, so it’s technically possible to bypass any limitations that are imposed by the official version.

Large language models have some downsides as well. It’s theoretically possible to generate all sorts of unsafe content (e.g., How do I build a Molotov cocktail?) and think of the high school or college student who generates a term paper using a large language model and passes it off as their own.

Training a large language model takes a huge amount of computer resources, so it is only done periodically. The current version of GPT3 at this writing, called GPT3.5, was trained at the beginning of 2022, so it doesn’t have any knowledge of events since then.

Finally, large language models have the habit of occasionally generating text that is totally untrue and presenting it as fact. Meta (formerly known as Facebook) was forced to remove its large language model called Galactica from the internet after only three days. It was supposed to generate scientific text, but it sometimes generated fake research and wiki articles that sounded right but were baseless. It even attributed some of these articles to real scientists.

The companies that have created these technologies are doing everything possible to mitigate these issues, but it’s not clear that they can be 100% effective in making these tools completely safe and accurate. There is a great deal of potential but currently there are still some flaws that limit their utility in some environments (e.g., I wouldn’t trust GPT3 to give me medical advice.). Hopefully, many of these issues can be solved over time.

The rapid pace of artificial intelligence advancement has led to speculation on when artificial intelligence will surpass human intelligence. This is usually called the “technological singularity”. I asked GPT3 when this will happen:

The technological singularity is a hypothetical future event that is predicted to occur when technological advances, particularly in the fields of artificial intelligence and machine learning, become so advanced that they fundamentally alter human civilization. The exact timing of this event is difficult to predict, and there is much debate among experts about when or if it will ever occur. Some believe that it could happen within the next few decades, while others believe it may not happen for hundreds or even thousands of years. Ultimately, the timing of the singularity will depend on many factors, including the rate of technological progress, the development of new technologies, and the ability of humanity to adapt to these changes.

This is a long way of saying that any timeframe for this breakthrough is pure speculation. Hang on though because the ride will surely be fun!

Bruce Gobioff is a graduate of MIT and spent 35 years in the computer industry with IBM as a software developer and manager. He’s been an OLLI-USF member for seven years and is currently an OLLI-USF instructor teaching classes in both liberal arts and technology subjects including artificial intelligence. His next class on AI is in February 2023. He wants you to know that this is not a computer-generated image. It’s the face he’s been stuck with since birth.