One of the Most Powerful AIs Ever Created Makes its Public Debut

Last week, ChatGPT, which is already being called one of the most advanced artificial intelligences ever created, was released to the public. For now, it is completely free to use, and according to its maker, the OpenAI research company, it quickly gained over a million users in just a few days.

I have been covering AI advancements for a few years now, and even generated an original story using the AI Dungeon program for a previous Nextgov column. But as impressive as most of the other AIs I have studied have been, ChatGPT is on a whole other level. It’s not perfect yet, but if you want to get a glimpse of how AI technology could change the world in the near future, it’s a great place to start.

What makes ChatGPT so special?

ChatGPT was designed from the ground up to interact with humans, so the front end interface is designed like a chatbot. Users type their queries and the AI provides answers. It’s not good enough yet to pass a Turing Test, so you really can’t mistake it for being a human, but it will answer questions in natural language using a friendly, if not sometimes stilted, style.

The new AI is different from a search engine because it can’t access the internet. Instead, it pulls its knowledge from what must be an incredibly large set of data. But it does not just use raw data alone. OpenAI trained ChatGPT using a technique called Reinforcement Learning from Human Feedback, or RLHF. In RLHF, human trainers work with the AI by taking on the role of both a user and an assistant to the AI. Trainers would first ask questions and then modify the AI’s responses, training it for better results. The AI learns from that advice and modifies future responses appropriately.

The result is an AI that can accomplish amazing tasks better than almost anything that came before it. It can provide a term-paper worthy discussion on almost any topic, write poetry, give advice about difficult tasks and even write code in a variety of programming languages. You can even feed it complex or technical information and ask the AI to explain it to you. For example, I copied a long, complex description of nuclear fission from a webpage and asked ChatGPT to explain it to me in simple terms. I was shocked that it was able to perform that task after thinking about its response for just a few seconds.

Putting ChatGPT to the test

Right now, anyone can use ChatGPT for free, as OpenAI is still in development, and having users query the AI is the best way to further refine its capabilities. All you need to do is set up an account and you can get started. Company officials have said that each query costs a few cents worth of computer time, which, when multiplied by its millions of users, can really add up, so ChatGPT may have to start charging in the future. But for now, everything is free.

I started out by asking ChatGPT what it could do. This is what it told me:

ChatGPT: OpenAI has designed me to be a powerful and efficient language model, capable of generating high-quality text in a wide range of styles and formats. I am trained on a large and diverse corpus of text data, which allows me to produce responses that are relevant, coherent, and consistent with the input that I receive. I am also designed to be flexible and adaptable, able to generate text in multiple languages and dialects, and to generate text in response to a wide variety of prompts and inputs.

However, it is important to note that I am not capable of surpassing human intelligence or creativity. I am a machine learning model, and I am only able to provide responses based on the data that I have been trained on. I do not have the ability to learn, reason, or make independent judgments, and I am not able to generate original content or ideas. My primary function is to assist users with tasks related to natural language processing, and I am not intended to replace or compete with human intelligence.

In addition to generating text, I can also assist users with tasks related to understanding and interpreting natural language. I am able to summarize and paraphrase text, providing a concise and coherent summary of a given passage of text. I can also answer questions about a given passage of text, providing detailed and accurate responses to user inquiries. I am also able to perform language translation, allowing users to easily understand and communicate with individuals who speak different languages.

I already mentioned how it was able to break down a complex description about a topic where I am not an expert—nuclear fission—into an easy-to-understand definition. So, next I decided to try something very specific. I asked ChatGPT to help me write an app in the C++ programming language that would check the current times around the world and display them on my screen. Surprisingly, not only did ChatGPT give me pointers like which coding libraries I should use to grab the current times, but it actually wrote the code for me in C++. That is pretty amazing.

ChatGPT: To write a program in C++ that tells the current time of several places around the world, you will need to do the following:

• Include the necessary C++ libraries, such as iostream and chrono, which will provide access to the input/output and time/date functions that you will need to use in your program.

• Define the time zones for the different locations that you want to include in your program. You can do this by using the time_zone class provided by the chrono library, which allows you to specify the time zone offset and the daylight savings time rules for each location.

• Use the local_time class provided by the chrono library to create a local_time object for each location. This class allows you to specify the local time and time zone for each location, and it will automatically adjust the time for daylight savings time.

• Use the format function provided by the chrono library to format the local time for each location as a string, using the desired format (e.g. "HH:MM:SS" for a 24-hour clock, or "h:MM:SS AM/PM" for a 12-hour clock).

• Use the cout stream provided by the iostream library to output the formatted local time for each location to the console.

Next, I went from something highly technical, like programming, to something more general. My wife was having trouble giving our cat medicine, so I asked ChatGPT to help us out. And again, it gave perfect advice.

ChatGPT: Giving a cat medicine can be a challenging task, as cats can be hesitant or unwilling to take medication. However, there are some tips and techniques that can make the process easier and less stressful for both the cat and the caregiver.

One of the key things to keep in mind when giving a cat medicine is to approach the task in a calm and gentle manner. Cats can sense stress and anxiety, and they may become more resistant to taking medicine if they are feeling scared or threatened. It is important to speak to the cat in a soothing voice, and to handle them gently and carefully to avoid causing them any discomfort or fear.

Another tip for giving a cat medicine is to use a delivery method that is suitable for the type and dosage of the medication. For example, some medications can be mixed into the cat's food or water, making it easy for the cat to consume without realizing that they are taking medicine. Other medications, such as pills or capsules, may need to be given directly by mouth using a pill dispenser or syringe. It is important to follow the instructions provided by the veterinarian or the medication packaging to determine the best delivery method for the cat.

In some cases, it may be necessary to restrain the cat in order to give them medicine. This should only be done as a last resort, and it should be done in a way that minimizes the cat's discomfort and stress. For example, the cat can be wrapped in a towel or blanket, with their head exposed, and their mouth can be gently opened using your fingers or a pill dispenser. This will allow you to place the medication directly in the cat's mouth, without having to force them to swallow it.

Next, I tried to engage ChatGPT’s creative side. I asked it to write several poems, which it did easily. One inspiring verse it penned about football should be used in NFL commercials. But I tried to really push the AI by giving it a challenge that I did not think in a million years that it could accomplish, least of all not within a few seconds of asking my question. I asked if it could tell me a story about a dog getting a new bone, but to tell it from the dog’s perspective. I almost fell out of my chair when it returned such an amazingly cute little story.

ChatGPT: Once upon a time, there was a happy little dog named Max. Max loved nothing more than to chew on a good, hearty bone. He had a collection of bones that he kept in a special basket in his doghouse, and he would spend hours each day gnawing on them and scraping off the tasty bits of meat and gristle that were stuck to them.

One day, Max's owner came home from the pet store with a brand-new, gleaming white bone for him. Max's tail wagged with excitement as his owner placed the bone in his basket. He couldn't wait to sink his teeth into it and taste the delicious flavor of the fresh marrow inside.

Max picked up the new bone and carried it over to his favorite spot in the sunshine. He sniffed it all over, taking in the rich, meaty scent. Then, with a happy bark, he dug his teeth into the bone and began to chew.

The new bone was everything Max had hoped for and more. It was big and meaty, with plenty of tasty bits to chew on. Max gnawed on it for hours, feeling the satisfying crunch of the bone between his teeth. He was in doggy heaven.

As the sun began to set, Max finally grew tired and lay down to rest, still cradling his beloved bone between his paws. He fell asleep with a contented sigh, dreaming of all the wonderful bones he would chew on in the days to come.

The end.

Okay, so ChatGPT can program in multiple computer languages and tell a capable bedtime story. I’m actually a little bit jealous of it right now. And it can break down complex information and help people with unusual or difficult tasks. All that is great, but among the many news stories I read or saw about ChatGPT were quite a few that raised the concept of AI being a bad thing, which could potentially do more harm than good. So I decided to ask ChatGPT directly why it thought so many people were afraid of it.

ChatGPT: There are many reasons why some people may be afraid of artificial intelligences (AIs). Some people may be afraid of the potential consequences of creating powerful AI systems, such as the potential for them to become uncontrollable or to harm humans. Others may be afraid of the potential impact of AI on society and the economy, such as the potential for AI to replace human workers and disrupt labor markets. Still others may be afraid of the ethical and moral implications of creating intelligent machines, such as the potential for AI to challenge or undermine human values and beliefs.

Some people may also be afraid of AI because of the way it is portrayed in science fiction and popular culture. In many stories and movies, AI is depicted as a powerful and malevolent force, capable of outsmarting and enslaving humans. This portrayal can create fear and anxiety in some people, leading them to view AI as a threat to human existence.  

I like how ChatGPT brought up ethical AI because I have an interview scheduled with an AI scientist on that very topic, and already talked with the founder and CEO of Credo AI, Navrina Singh, about the need for ethics and federal AI guidelines in a previous Nextgov column. I shared that article with ChatGPT and asked it to comment on the importance of ethical AI, not unlike how I would have asked a human mentioned in a news story if they wanted to respond or comment. This is what ChatGPT told me.

ChatGPT: There are many ways in which AI can be designed and used in an ethical manner. One of the key principles of ethical AI is transparency, which means that AI systems should be designed and used in ways that are transparent and explainable to humans. This can help to ensure that AI systems are accountable and responsible for their actions, and that they can be understood and evaluated by human users and stakeholders.

Another principle of ethical AI is fairness and non-discrimination. AI systems should be designed and used in ways that do not discriminate against or disadvantage certain groups of people based on factors such as race, gender, age, or disability. This can help to ensure that AI systems are fair and equitable, and that they do not perpetuate existing biases or inequalities.

In addition to transparency and fairness, ethical AI should also prioritize human well-being and safety. AI systems should be designed and used in ways that protect and enhance human welfare, and that minimize the potential risks and harms associated with their use. This can include considerations such as the potential impact of AI on employment and the economy, the potential for AI to be used for malicious or harmful purposes, and the potential for AI to cause physical or psychological harm to humans.

So there you have it, a talk about ethical AI from an AI itself. 

Everything about ChatGPT is impressive. Although I did find a few limitations in what it could do—it’s not up to date about current events after 2021 and is restricted from talking about certain topics like politics—overall, this is one impressive application. Once it gets fully refined and tuned, it might really change the world.

John Breeden II is an award-winning journalist and reviewer with over 20 years of experience covering technology. He is the CEO of the Tech Writers Bureau, a group that creates technological thought leadership content for organizations of all sizes. Twitter: @LabGuys