GPT-3: Revolutionizing AI One Character At a Time
We live in exciting times, where technological evolutions are changing the world every single day, making a significant mark in history. Artificial Intelligence has already made a huge imprint on the world, the anticipation of what is next in the future is exhilarating.
A recent Artificial Intelligence development, OpenAI, has left the world stunned with the possibilities that this new model ignites.
OpenAI is an Artificial Intelligence research and development company, founded by Elon Musk and Sam Altman. Throughout the years, the company has brought exciting new AI technologies to the world, however, their latest evolution, the GTP-3, has the power to revolutionize the AI world.
What is GPT-3?
GPT-3 also known as the “Generative Pre-Training Transformer”, is the 3rd language model created by OpenAI. GPT-3 is a pre-trained model that can use its attained knowledge to produce human-like texts.
To break it down further, GPT-3 takes input for a user and uses all the resources at its hand (or whatever body parts language models have) and outputs exactly what the user requested in the input.
The language model has been trained with billions of bits of data from various sources such as books, Common Crawl, Wikipedia etc. According to Rob Toews in a Forbes article, “at its core, GPT-3 is an extremely sophisticated text predictor.” Based on the user’s input, the language model takes keywords and makes its best statistical prediction as to what the first piece of the output will look/sound like. This process is repeated numerous times where the language model takes “the original input together with the newly generated chunk, and treats that as a new input, generating a subsequent chunk — until it reaches a length limit.”
What makes the GPT-3 so unique from its predecessor models? It is a unique offering of 175 BILLION parameters within this language model. This means that the GPT-3 algorithm has 175 billion different weights or learned parameters to evaluate the user input to give back its predicted output. It is essentially pre-trained on the entire Internet.
That is a LOT of different parameters that the GPT-3 algorithm goes through, and what makes the language model that more special is that it can give its output in a matter of seconds. This is because GPT-3 is what AI specialists call a Neural Network. According to Investopedia, a Neural Network is a series of algorithms that “mimics the way the human brain operates.” Similar to a human brain, a Neural Network recognizes relationships between different datasets. In the training phase, the language model is when the Neural Network is being put together.
To put the hype of GPT-3 into perspective, the predecessor, GPT-2, was known to be “state-of-the-art” and only had 1.5 billion parameters in place. With a broader Neural Network, the GPT-3 is more accurate with its predictions.
Application of GPT-3
GPT-3 is capable of: generating functioning code, writing creative fiction (parodies, poems etc.), business memos in seconds of the user typing in their command.
The Guardian released an article entirely written by GPT-3 to persuade the reader that “robots come in peace.” With only a few sets of instructions, the GPT-3 was able to articulate an entire article with evidence, quotes and humour to convince readers that robots mean no harm to humans and do not wish to take over the world.
This language model has the potential to make significant improvements in every existing industry: the Tech industry, Financial industry, Healthcare industry, Teaching industry. The possibilities are endless. As of now, OpenAI had only released the model to certain companies, but once it is more openly available, every industry will be benefitted from this software.
Downsides of GPT-3
The hype of GPT-3 is very justifiable, however, there are some downsides to the model that reduces its appeal. GPT-3 dataset is its biggest asset and in some ways its weakness as well. According to TechCrunch, 60% of GPT’s dataset was built from a majority of Common Crawl data, meaning that the language model was trained with both “very reputable” such as BCC, and “less reputable” sources such as Reddit. Due to this, the language model learned humanity's “best” and its “worst” qualities. This includes profanities, underlying problematic biases, sexualization of women, religious discrimination etc.
Traditional common in almost all Artificial Intelligence systems, GPT-3 also lacks common sense. According to Lacker.io, during a Turing Test, the language model “can answer a lot of common sense questions” and to stump it, you “need to ask questions that no normal human would ever talk about.”
CEO Sam Altman himself has expressed his thoughts on the overhype of GPT-3. In his Tweet, he says that “the GPT-3 hype is way too much. It’s impressive but it still has serious weaknesses and sometimes makes very silly mistakes. AI is going to change the world, but GPT-3 is just a very early glimpse. We still have a lot to figure out.”
Nonetheless, with a few modifications, the GPT-3 can live up to its hype and revolutionize how the world works. There are endless possibilities of real-life applications of GPT-3.
Possible Real-Life Applications
If used in teaching apps, the language model can help teach coding to kids at a younger age. With the GPT-3 at their side, they will be able to ask questions and learn without much hindrance.
It can be used to further develop virtual assistant AI technology such as Alexa. With a few modifications in its system, GPT-3 can strengthen user-device communications. If it were to replace Alexa’s already existing backend or integrate with it, GPT-3 will help make conversation with the user much smoother and less static.
When implemented into the healthcare industry, GPT-3 will make lives much easier. If the input to the model were to include information of symptoms, history and other relevant information, GPT-3 will be able to assist doctors in making diagnoses of diseases and illnesses. Together, doctors and AI will work hand-in-hand to provide patients with efficient medical care.
With this language model, the government will be able to expand its reach of online surveillance. With full access to the internet, GPT-3 will be able to surveil every message sent on the internet, whether this is social media, in a form or potentially on the dark web.
These are just a few examples of potential future implementations. In reality, the implementations of GPT-3 are endless. While GPT-3 has a few flaws and kinks that need to be ironed out, the overall benefit of the functionality is much higher, and can significantly help better our current systems if put in the right hands.