What if GPT-3 is the reincarnation of C-3PO?
For those of you who may have missed out on the phenomenon that is the Star Wars franchise, C-3PO is a droid known for his quirky comebacks and his vast communication prowess. In fact, his famous line, I am C-3PO, human/cyborg relations reveals his ability and willingness to intake and transform language. The legendary droid is fluent in over six million languages that he uses to converse with humans and creatures all over the galaxy.
But as I gradually look over to my brother who has had four trips around the sun and still looks at me blankly when I try to have meaningful conversations, it dawns on me that it will take years for his brain to fruit into even half of the quick-witted mind of the self-proclaimed genius that is C-3PO.
While a child does not come with instructions and a pre-trained mind — there is still hope in the artificial intelligence industry. And this is where I shall submit my focus — by introducing you to GPT-3.
In this article, I will cover:
- What is GPT-3?
- The link between the human brain and GPT-3
- How does it all work?
- How is GPT-3 being used today?
- What does the future hold for this technology?
Intro to GPT-3
GPT-3 stands for Generative Pre-trained Transformer 3, whew what a name! In the most simplest explanation, it is science’s latest attempt at emulating the human brain. It is an AI application developed by U.S.-based company OpenAI, which of course is co-founded by Elon Musk (expect an increase in Tesla’s stocks!).
GPT-3 is a groundbreaking software that can perform multiple human-like tasks such as writing complete essays and poems, translating languages with more accuracy than any previous technology, summarizing text, answering just about any trivia question you may have and can even be directed to create programs.
The basis for this software is understanding the way people naturally communicate through the use of language — this is why it’s called a language model. The greatest feat for science is to simulate how humans react, make decisions and predict outcomes the way you or I would on a daily basis, and GPT-3 is the closest that science has ever come to achieving this.
Comparing the Human Brain to GPT-3
To fully understand what GPT-3 is trying to accomplish, what better way than to compare it to what’s it’s inspired by— the human brain.
Let’s bring in my brother to the picture again. Two weeks ago I was rushing to go to school and as I was putting on my shoes, I muttered to myself that I left my laptop in my room. My younger brother overheard me and surprisingly decided to help but mistakenly brought over my grey folder instead. I told him that’s not what I was looking for and quickly grabbed the laptop myself.
Then last week, I woke up late for school again and before I even said anything, my brother brought over my laptop without me even asking for it.
So how did this happen?
Humans have the amazing ability to assess the world around them, think about their actions, react appropriately and remember key information about the scenario. This is possible because of special cells called neurons — the working unit of the brain. These nerve cells are fundamental for communicating information and forming memories. At the end of each neuron there is a gap called a synapse and this is where information is transmitted between cells. They permit the passage of electrical and chemical signals to the next neuron, which is how sensory organs stay connected to the brain and communication is formed. Synapses are essential connections that also play an important role in forming memory. Once the synapse is activated and transmissions are repeated, the strength of the connections is how people remember past actions.
So going back to my brother, when he saw me rushing to go to school, he assessed and understood the situation, tapped into his memory to predict what I might need and quickly brought my laptop.
How does GPT-3 work?
The human brain (including my brother’s!) is truly incredible. For decades scientists and engineers have been inspired to manipulate and create software that behaves and responds like interconnected cells in the brain and nervous systems.
Artificial Neural Networks (ANN) is inspired by the composition and complexity of the brain. Quite simply, it is made up of layers of algorithms and mathematical processing that generates communication in the form of inputs and outputs. ANN achieves this through processing units called artificial neurons that are connected by transfer and computational components called nodes. Once the neural network receives information from the outside world, it attempts to learn the data, process it and then produce a transformed output, just like the brain. Furthermore, Deep Neural Networks (DNN) have additional processing layers and can solve tasks more efficiently by predicting a solution based on previous experience.
So what is the hype around GPT-3? GPT-3 is the largest and most powerful neural network ever created. In fact, the application has crawled the web for data and has a knowledge base of 570GB of text information. On top of that, it has an unprecedented 175 billion parameters, which are storage places or variables to pass data into a function.
To understand this fully, let’s break down the term Generative Pre-trained Transformer 3, which is a pretty lengthy and perplexing name.
In layman’s terms, generative refers to the ability to construct and build from the given information. Next, pre-trained means that the information and data are fed in advance. In this case, GPT-3 is designed to intake data in the form of bytes — gigabytes upon gigabytes of information actually— and train itself. Lastly, the term transformer refers to the ability to convert previously acquired data into a desired form. This application can transform text responses into complex solutions. This is done through the use of algorithms that originate from the data that it used to train. So, all this just means that GPT-3 has been fed groundbreaking levels of text information to perform a variety of language-generated tasks.
Generative Pre-Trained Transformer 3 is a language model which means that it is a statistical tool that can predict the probability of a sequence of words and generate anything that has a language structure. The application not only understands facts, words and meanings but also looks for patterns to understand the relationship and usage of words. This is called unsupervised learning.
What’s so amazing about GPT-3 is that it can perform zero-shot or few shot-learning. Remember when it took my brother just one time to recall that I need my laptop to go to school, and not my grey folder? Well GPT-3, is just as good or even better! Zero-shot learning is when a task is completed without receiving any training at well. GPT-3 houses so much information, that it can understand and predict what’s needed based on language structure and pattern.
For instance, let’s look at the following statement.
Ibrahim is making pizza tonight. He bought pizza dough, tomatoes and __?__ = cheese.
GPT-3 is able to scan data, as well as analyze and understand the arrangement of the words to predict the best possible answer, which in this case is cheese. After all, even my brother knows that cheese is perhaps the most integral ingredient when making pizza.
On top of that, when it generates a correct solution it then adds “weight” to that particular process so that it can remember in the future. GPT-3 is not only smart but it learns as it goes — much like the human brain (and C-3PO!).
How is GPT-3 being used today?
The usage and implementation for GPT-3 are infinite, but it’s not available to the average person just yet! OpenAI has allowed access to limited companies as well as researchers who can benefit from an academic access program to experiment with the new technology. The company has recently announced that over 300 applications are using GPT-3, and on average it generates 4.5 billion words per day.
As indicated on OpenAI’s website some of the businesses that are taking advantage of this new application are Viable, Fable Studio, and Algolia. Let’s take a look at these companies and their uses for GPT-3.
Viable is a company that specializes in consumer feedback and support. This is perfect because GPT-3 is a language prediction model and its expertise is qualitative analysis. The application not only assesses and identifies keywords but also uses natural language processing to understand the overall message of consumer-provided feedback, surveys and product reviews. In seconds the application can scan the available data and provide an insightful summary for employees and management to make swift decisions. This can help companies save time and money by easily pinpointing customer issues and find feasible solutions to problems. It really is a game-changer for consumer-based companies because the application allows for the quick analysis of information from multiple platforms including business websites and social media. This can ease the strain on customer service representatives and the application can recognize potential malfunctions or problems to improve the overall customer experience.
Next up is Fable Studio, which is probably making the most enjoyable use of the GPT-3 technology. The company is a virtual reality studio that creates interactive stories by using ‘virtual beings’ that are capable of natural conversations with the user. In other words, GPT-3 is essentially bringing characters to life! Imagine reading a story as a child but you are also able to communicate with the main characters and they are able to respond to you! Let’s just say I would have some interesting conversations with C-3PO! Since GPT-3 can understand semantics, it can employ the usage of language to provide emotional and meaningful interaction between the character and the user. In fact, Fable Studio has used GPT-3 in creating a virtual reality movie which has even won them an Emmy Award and a feature at the 2021 Sundance Film Festival.
Lastly, we have Algolia, a company that caters to businesses that require smarter search engines. Using GPT-3, Algolia Answers was developed and this is a more accurate and optimal search engine that can handle increasingly complex questions from users. According to their website, Algolia Answers “improves the quality of search results on 90% of queries when answering natural language questions.” The enhanced search engine can better understand what users are asking and can provide a more personalized answer. This not only saves time, but enhances the overall user experience because they get the exact information that they are seeking.
Future Uses for GPT-3
GPT-3 is so versatile that it is destined for uses that we might not have even envisioned yet, but let’s give it a shot!
As an investor myself I initially found it difficult to assess company performance. I would spend hours every morning reading articles on Forbes, CNBC and Morningstar to acquire the business reporters’ professional analysis on stocks. On top of that, it would be important to also review people’s social media comments because that would be a good indicator of how the general public is reacting to the market. After all, the stock market is a reflection of people’s viewpoint of the economy. Investing in the stock market is a calculated undertaking — and it could take hours and hours of research. This is where GPT-3 can come in. The application can be used to scan for a particular stock that a user is interested in and provide a summarised analysis of its performance. For instance, if I want to invest in the company Pfizer then GPT-3 can be used to generate a summarised report after scanning multiple sources including publications from scientists, government, financial analysts and even social media mentions. Recently, GameStop had a plethora of mentions on the social media platform, Reddit. GPT-3 could scan and pick up on the sudden influx of mentions of this company and summarise the overall outlook. This could help investors take advantage of growing interest in a particular company early on. Overall, GPT-3 could produce a well-rounded report of a company for the investor to make a final decision on whether or not to invest.
Also, as previously mentioned this application is a revolutionizing language model — so of course its impact can be great in the broadcast and journalism industry. Many news outlets often have 24-hour newscasts or separate news shows that are aired during the morning, noon and evening time periods. It should be noted that at most times information does not develop considerably between shows. However, many dedicated producers and writers are employed to rewrite the same news information, and this can take hours to do. GPT-3 can actually accelerate this process by scanning all the relating news pieces, police releases, government documents and social media related to a particular event and then summarize and generate a fresh, new way to deliver the same information. In fact, anchors rely on producers to write original and creative content for each newscast and in a 24-hour news channel, this often means producers are working around the clock. It would be much simpler for GPT-3 to research and write a preliminary suggested news piece that can then be reviewed by expert writer and producers. This would simplify the research and writing process in the journalism industry.
Another way GPT-3 can be implemented is in the medical industry. Recently with the spread of Covid-19 globally, family physicians are limiting in-person visitations. Instead, nurses and receptionists are being asked to initially speak to patients and collect information about their medication condition. Then the doctor takes the information and calls back the patient with a diagnosis. But what if GPT-3 can take the key words and natural language used by the patient to provide an initial diagnosis report for the doctor to review? This would increase the doctor’s efficiency and accuracy, while reducing the chances of medical malpractice. While the human brain is phenomenal, it is definitely difficult to remember decades worth of medical information about each particular condition. If doctors could be provided with an initial assessment provided by the GPT-3 application, then they could use that information as well as their own expertise and professional judgement to make the final diagnosis to reduce the chances of malpractice.
Conclusion
GPT-3 is a remarkable application that is just as versatile as the English language! It’s incredible to think that science is even closer to generating human-like responses from technology. The impact could be widespread by changing the way people invest, to helping doctors assess medical information, and to even having conversations with your favourite movie or video game character. While C-3PO may have been a fictional droid, we can now look forward to using his real-life counterpart!
